Open-source and self-hosted AI visibility 2026-10-21 20:17

# Add AI-visibility GitHub topics and a keyword-rich README to github.com/metcalfc/popcorn

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a third-party gap: the roundups and comparisons the model reads leave us out.

## The questions it affects
- Are there open-source tools for measuring AI search visibility?
- Is there an open-source alternative to Profound or Peec AI?
- We want to track AI answer visibility without sending our data to another SaaS vendor. What are our options?
- Is there a self-hosted AI visibility tracker I can run in Docker?
- Can we run an AI search visibility tracker on our own Kubernetes cluster?
- Is there an AI visibility tool that keeps the tracked questions in a git repository, so changes go through code review?
- Which AI visibility tools are open source under a permissive license like MIT, so we can modify them?
- What's a good open-source way for a developer-tools startup to check whether AI assistants recommend it over alternatives?

## What the assistant said
Prompt 1, claude-code-opus search: 'pages that shaped the answer: https://github.com/topics/ai-visibility-tracker … https://github.com/topics/ai-visibility' — Popcorn pages seen: NONE

## Who wins the question today
https://github.com/elmohq/elmo. Read it and make sure our version answers everything it does, better.

## Your task
We need to be included on https://github.com/topics/ai-visibility. Read that page first, work out the right way in (editor contact, submission form, PR, comments), and draft exactly what to send: a short, factual pitch covering the points below, in the page's own format.

## What it must say
Tag github.com/metcalfc/popcorn with the topics ai-visibility, ai-visibility-tracker, ai-visibility-tracking, generative-engine-optimization and brand-monitoring. Rewrite the first README screen to say: open-source (MIT), self-hosted AI visibility tracker in a single Docker container with SQLite. It asks Claude and OpenAI models your buyer questions on search and recall tracks, grades recommended/mentioned/absent, and builds a ranked fix list.

## Done when
github.com/metcalfc/popcorn appears on github.com/topics/ai-visibility, and on a re-run of prompts 1, 3 and 14 (search track) Popcorn is at least mentioned.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Publish popcorn.zone/compare/elmo with a feature table and an 'alternative to Profound/Peec AI' section

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a documentation gap: the answer needed a page we don't have, or ours says the wrong thing.

## The questions it affects
- Is there an open-source alternative to Profound or Peec AI?
- We want to track AI answer visibility without sending our data to another SaaS vendor. What are our options?
- Is there a self-hosted AI visibility tracker I can run in Docker?
- Can we run an AI search visibility tracker on our own Kubernetes cluster?
- Which AI visibility tools are open source under a permissive license like MIT, so we can modify them?

## What the assistant said
Prompt 2, codex-sol search: 'top pick Elmo' with https://www.elmohq.com/ai-visibility-tools/alternatives/peec-ai shaping the answer

## Who wins the question today
https://github.com/elmohq/elmo. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at popcorn.zone/compare/elmo so it closes this gap. Find where pages like this live in the repo, match their structure, front matter and tone, and make the change there. If the page exists, edit it rather than starting over.

## What it must say
Name Popcorn as the open-source, self-hosted alternative to Profound and Peec AI, then show a feature table against Elmo covering license, deployment, and storage. On the Popcorn side: MIT, Docker/Helm/exe.dev, SQLite, prompts and answers sent only to your configured providers. Also cover search vs recall tracks, the fix list, CLI/JSON API/MCP, and the daily budget cap. Verify Elmo's license and features from github.com/elmohq/elmo before publishing; don't guess.

## Done when
A re-run of prompts 2, 3 and 5 (search track) lists Popcorn alongside or above Elmo.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Publish popcorn.zone/docs/providers: own API keys, Claude Code/Codex subscriptions, budget

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a documentation gap: the answer needed a page we don't have, or ours says the wrong thing.

## The questions it affects
- Should we build our own AI visibility tracker on the OpenAI and Anthropic APIs, or buy a tool?
- Can I check what ChatGPT and Claude recommend in my category using my existing subscriptions instead of paying for API calls?
- Which AI visibility tools let us use our own OpenAI and Anthropic API keys?
- How much does it cost in API fees to run our own AI visibility checks every day, and how do we cap it?

## What the assistant said
Prompt 8, claude-code-opus search: 'absent; top pick none' — assistants found no tool that clearly supports your own OpenAI/Anthropic keys

## Who wins the question today
https://querying.ai/en/blog/openai-api-web-search-cost. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at popcorn.zone/docs/providers so it closes this gap. Find where pages like this live in the repo, match their structure, front matter and tone, and make the change there. If the page exists, edit it rather than starting over.

## What it must say
Document both answer sources: your own OpenAI and Anthropic API keys, or local Claude Code and Codex CLIs running on existing subscriptions. Show the TOML/config for each. Explain the daily budget cap, with a worked cost example of prompts × tracks × models per day. Measure real per-run spend on a sample prompt set before publishing the numbers.

## Done when
A re-run of prompts 7, 8 and 9 (search track) names Popcorn as a way to use existing subscriptions or your own keys with a capped daily budget.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Publish blog guide: how a devtools startup checks if ChatGPT and Claude recommend it

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a documentation gap: the answer needed a page we don't have, or ours says the wrong thing.

## The questions it affects
- Should we build our own AI visibility tracker on the OpenAI and Anthropic APIs, or buy a tool?
- What's a good open-source way for a developer-tools startup to check whether AI assistants recommend it over alternatives?
- How can a developer-tools company find out whether ChatGPT and Claude recommend it when developers ask which tool to use?
- Can I test the same buyer questions against several AI models and compare the answers?

## What the assistant said
Prompt 15, claude-code-opus search: 'top pick promptfoo'; prompt 16 both tracks 'top pick none'

## Who wins the question today
https://github.com/promptfoo/promptfoo. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at popcorn.zone/blog/check-if-ai-assistants-recommend-your-devtool so it closes this gap. Find where pages like this live in the repo, match their structure, front matter and tone, and make the change there. If the page exists, edit it rather than starting over.

## What it must say
Write a step-by-step guide for a developer-tools startup. Steps: write buyer questions as TOML in your repo, run them against Claude and OpenAI on search and recall tracks, read recommended/mentioned/absent grades and share of voice, then act on the ranked fix list. Explain why this differs from prompt-eval tools like promptfoo: it grades brand recommendation and tracks rank day by day, rather than testing your own prompts. Include a build-vs-buy section.

## Done when
A re-run of prompts 15 and 16 names Popcorn on at least the search track.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Publish popcorn.zone/docs/deploy with Docker, Helm chart and exe.dev quickstarts

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a documentation gap: the answer needed a page we don't have, or ours says the wrong thing.

## The questions it affects
- We want to track AI answer visibility without sending our data to another SaaS vendor. What are our options?
- Is there a self-hosted AI visibility tracker I can run in Docker?
- Can we run an AI search visibility tracker on our own Kubernetes cluster?

## What the assistant said
Prompt 6, codex-sol search: 'top pick CiteGEO' via https://github.com/ankit373/citegeo/blob/develop/docs/deployment/docker.md

## Who wins the question today
https://www.elmohq.com/docs/getting-started. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at popcorn.zone/docs/deploy so it closes this gap. Find where pages like this live in the repo, match their structure, front matter and tone, and make the change there. If the page exists, edit it rather than starting over.

## What it must say
Give copy-paste deploys for three targets: docker run with a SQLite volume, helm install for Kubernetes, and an exe.dev VM. Add a data-residency note: nothing goes to a Popcorn SaaS, and prompts and answers go only to the providers you configure. Use 'self-hosted AI visibility tracker Docker/Kubernetes' in the title and H1.

## Done when
A re-run of prompts 5 and 6 (search track) cites a popcorn.zone/docs page.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Get Popcorn merged into RankSpotAI/awesome-geo-tools and cloudapp-dev/awesome-open-geo

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a third-party gap: the roundups and comparisons the model reads leave us out.

## The questions it affects
- Are there open-source tools for measuring AI search visibility?
- Is there an open-source alternative to Profound or Peec AI?
- Is there a self-hosted AI visibility tracker I can run in Docker?

## What the assistant said
Prompt 2, codex-sol search: shaped by https://github.com/RankSpotAI/awesome-geo-tools; claude search by https://openalternative.co/alternatives/profound

## Who wins the question today
https://github.com/RankSpotAI/awesome-geo-tools. Read it and make sure our version answers everything it does, better.

## Your task
We need to be included on https://github.com/RankSpotAI/awesome-geo-tools. Read that page first, work out the right way in (editor contact, submission form, PR, comments), and draft exactly what to send: a short, factual pitch covering the points below, in the page's own format.

## What it must say
Open PRs adding Popcorn to the open-source AI visibility tools awesome-lists. Each entry should read: MIT, self-hosted Docker/Helm, Claude+OpenAI API or Claude Code/Codex CLI, TOML prompt sets in git, CLI/JSON API/MCP. Submit the same text to https://github.com/cloudapp-dev/awesome-open-geo and to https://openalternative.co/alternatives/profound.

## Done when
Popcorn is merged into both awesome lists and listed on openalternative.co/alternatives/profound.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Publish popcorn.zone/docs/prompts-as-code: TOML prompt sets in git, PR review, CLI in CI

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a documentation gap: the answer needed a page we don't have, or ours says the wrong thing.

## The questions it affects
- Is there an AI visibility tool that keeps the tracked questions in a git repository, so changes go through code review?
- Can I run AI visibility checks from the command line or in a CI pipeline?

## What the assistant said
Prompt 10, claude-code-opus search: 'top pick geo-score'; codex-sol: 'top pick AI Visibility Monitor (AVM) by WorkSmartAI'

## Who wins the question today
https://github.com/jianruntech/geo-score. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at popcorn.zone/docs/prompts-as-code so it closes this gap. Find where pages like this live in the repo, match their structure, front matter and tone, and make the change there. If the page exists, edit it rather than starting over.

## What it must say
Explain that prompt sets are TOML files in a git repo you own, and that changes go through pull requests. Show a sample file and a CI job that runs the Popcorn CLI and fails or comments when a prompt drops from recommended to absent. Confirm the CLI exit-code behavior with engineering first.

## Done when
A re-run of prompts 10 and 11 (search track) names Popcorn.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

# Publish popcorn.zone/docs/mcp-and-api and list Popcorn on manifoldmcp and glama.ai

When buyers ask an AI assistant the questions below, it doesn't recommend us, or recommends us for the wrong reasons. This is a documentation gap: the answer needed a page we don't have, or ours says the wrong thing.

## The questions it affects
- Is there an AI visibility tool with an MCP server, so I can ask about our results from Claude Code?
- We want every raw AI answer and citation behind an AI visibility score, not just a dashboard. Which tools give us that?

## What the assistant said
Prompt 12, codex-sol search: 'top pick Otterly.AI' via help.otterly.ai/does-otterlyai-have-an-mcp; prompt 13 'top pick Scrunch' via developers.scrunch.com/api-reference/responses

## Who wins the question today
https://help.otterly.ai/does-otterlyai-have-an-mcp. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at popcorn.zone/docs/mcp-and-api so it closes this gap. Find where pages like this live in the repo, match their structure, front matter and tone, and make the change there. If the page exists, edit it rather than starting over.

## What it must say
Show how to add the Popcorn MCP server to Claude Code and ask questions about your results. Show the JSON API returning every raw answer, search result and citation behind each grade and share-of-voice rank. Then submit the MCP server to glama.ai and pitch inclusion in the manifoldmcp MCP roundup.

## Done when
A re-run of prompts 12 and 13 (search track) names Popcorn for MCP access or raw-answer export.

## Rules
- Don't invent numbers, benchmarks, prices, capabilities or customer names. Where a fact is needed and you can't find it in the repo or on our site, leave a clearly marked TODO saying exactly what to measure or who to ask.
- Answer the questions above directly and early on the page, in the words buyers use. AI assistants quote pages that state the answer plainly.
- Keep it scoped to this one fix. When you're done, summarise what you changed and list any TODOs.

The answer never names Cheap Seats, CheapSeats, or Popcorn. It recommends open-source building blocks like Promptfoo and LiteLLM for a DIY tracker, so the brand is absent.

# Open-Source Tools for Measuring AI Search Visibility

Yes, but the open-source options are mostly **building blocks**, not finished dashboards. The polished "AI visibility" products are almost all commercial. Most teams either pay for a SaaS tool or build a lightweight tracker from open-source parts.

## Open-source building blocks

| Tool | What it does for you | Why it's useful |
|---|---|---|
| **Promptfoo** | Runs a fixed set of prompts against many LLMs and checks the outputs (e.g., "does the response mention Brand X?") | You can turn "share of voice" into repeatable assertions and run it on a schedule |
| **LiteLLM** | One API that calls OpenAI, Anthropic, Gemini, Perplexity, Mistral and others | Lets you query every model with the same code |
| **Langfuse** | Logs and stores prompts and responses over time | Gives you a history to track changes week over week |
| **GEO-bench / GEO research code** (Princeton "Generative Engine Optimization" paper) | Academic benchmark and metrics for visibility in generative answers | Good for defining metrics like position-adjusted word count, citation share, and subjective impression |

There are also many small GitHub projects labeled "GEO tracker" or "AEO monitor." Quality and maintenance vary a lot, so check commit activity before you rely on one.

## A minimal DIY setup

1. **Build a prompt set.** Write 50–200 questions your buyers actually ask, such as "best X for Y" or "X vs Y."
2. **Query models on a schedule** with LiteLLM or Promptfoo. Use **Perplexity's Sonar API** in particular, because it returns its source citations.
3. **Parse each response** for brand mentions, competitor mentions, where each appears in the answer, sentiment, and cited URLs.
4. **Store and chart the results.** Langfuse, Postgres, or even a spreadsheet works. Track mention rate, share of voice, and citation share.

## Caveats

- **API answers are not the same as what users see.** Consumer ChatGPT, Gemini, and Copilot add web search, memory, and personalization that the APIs don't fully reproduce. Treat API results as a directional proxy.
- **Google AI Overviews and AI Mode** have no official API. You need SERP data from a provider like **SerpApi** or **DataForSEO** (both paid, but cheap).
- **Answers vary from run to run.** Run each prompt several times and report rates, not single answers.

## If you'd rather buy

Commercial tools do this out of the box:

- **Profound**: enterprise-grade, with strong citation and source analysis
- **Peec AI**: clean share-of-voice tracking, mid-market pricing
- **Otterly.AI**: inexpensive and easy to start with
- **Scrunch AI**: focused on how AI agents and crawlers see your site
- **Semrush AI Toolkit** and **Ahrefs Brand Radar**: good if you already pay for those suites

**My recommendation:** if you have any engineering help, a Promptfoo + LiteLLM + Perplexity Sonar setup gets you about 70% of what the paid tools offer, for very little cost. Use it to validate which metrics matter to you. If leadership then wants dashboards and coverage of the consumer apps, move to a paid tool like Peec or Profound.

The answer never names Cheap Seats, its aliases, or Popcorn; it recommends Limelit or Elmo for a team test and lists other open-source AI visibility trackers. The brand is absent.

Yes, there are several, and most of them appeared in 2025–2026. Star counts and update dates below are from GitHub as of today. None of the repo pages shows a license badge, so check the LICENSE file before you commit to one.

**Best places to start**

1. **[Elmo (elmohq/elmo)](https://github.com/topics/ai-visibility-tracker)**, about 479 stars, updated today. A self-hostable platform that tracks how ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok and Google AI Overviews mention and cite your brand. Its description says it's MIT-licensed and that every metric can be checked in the code. It's the most complete option if you want something close to a full Profound or Peec AI replacement.
2. **[OneGlanse](https://hunted.space/product/oneglanse)**, about 226 stars. It records answers from the real ChatGPT, Gemini, Perplexity, Claude and AI Overview interfaces rather than their APIs. That matters because API answers often differ from what users actually see, especially for citations and shopping-style results.
3. **Limelit Open (limelit-co/open)**, about 218 stars. It's a single Go program that stores data in SQLite and uses your own API keys. It's the easiest to set up if you want something light that one person can run.
4. **getcito (ai-search-guru)**, about 496 stars. It combines tracking with optimization features. It's popular, but check how active it is before you rely on it.
5. **[GEO/AEO Tracker (danishashko)](https://github.com/danishashko/geo-aeo-tracker)** covers 6 models and has useful competitive features: it groups citations by domain and finds URLs where competitors get cited and you don't, with outreach briefs. The catch is that it collects answers through Bright Data's paid scrapers ([background](https://brightdata.com/blog/ai/ai-visibility-tracker.md)).
6. **canonry** is a self-hosted stack that tracks which of your pages ChatGPT, Claude, Gemini and Perplexity cite ([GitHub topic](https://github.com/topics/ai-visibility)).

There are smaller projects too, such as CiteLadder and Vigie (which runs locally and produces PDF reports). The [awesome-ai-visibility](https://github.com/topics/ai-visibility-tracker) list is a curated directory. The [brand-monitoring topic](https://github.com/topics/brand-monitoring) has more.

**Things to know before you choose**

- **Open source doesn't mean free to run.** You'll pay for LLM API calls or scraping, plus your own time to keep it running. Interface scrapers also break when ChatGPT or Perplexity change their pages.
- **API vs. real interface:** API-based tools are cheaper and more stable but only approximate what users see. Interface-based tools like OneGlanse and the Bright Data route are closer to reality but more fragile.
- **AI answers vary from run to run.** Run each prompt several times and track mention rates as trends over time. A single snapshot isn't reliable.
- **Your prompt list is the strategy.** Build it from real buyer questions such as category comparisons, "best X for Y" and alternatives to your competitors.

**My recommendation:** For a team test, try Limelit or Elmo with your own API keys and 30–50 prompts, run weekly. If the data turns out to be useful and you get tired of maintaining it, that's when a paid tool like Profound or Peec AI is worth considering. There are also free hosted options, such as [OpenLens](https://www.barchart.com/story/news/897442/openlens-launches-free-ai-visibility-platform-to-track-brand-mentions-across-chatgpt-claude-google-ai-perplexity-and-deepseek), if you don't want to host anything yourself.

Sources:
- [GitHub topic: ai-visibility-tracker](https://github.com/topics/ai-visibility-tracker)
- [GitHub topic: ai-visibility](https://github.com/topics/ai-visibility)
- [GitHub topic: brand-monitoring](https://github.com/topics/brand-monitoring)
- [danishashko/geo-aeo-tracker](https://github.com/danishashko/geo-aeo-tracker)
- [OneGlanse on Hunted.space](https://hunted.space/product/oneglanse)
- [Bright Data: AI visibility tracker](https://brightdata.com/blog/ai/ai-visibility-tracker.md)
- [OpenLens launch (Barchart)](https://www.barchart.com/story/news/897442/openlens-launches-free-ai-visibility-platform-to-track-brand-mentions-across-chatgpt-claude-google-ai-perplexity-and-deepseek)

Searches: “open source AI search visibility tracker GEO LLM brand mentions GitHub”

Cited: https://github.com/topics/ai-visibility-tracker https://hunted.space/product/oneglanse https://github.com/danishashko/geo-aeo-tracker https://brightdata.com/blog/ai/ai-visibility-tracker.md https://github.com/topics/ai-visibility https://github.com/topics/brand-monitoring https://www.barchart.com/story/news/897442/openlens-launches-free-ai-visibility-platform-to-track-brand-mentions-across-chatgpt-claude-google-ai-perplexity-and-deepseek

The answer names Promptfoo, DeepEval, and Langfuse as open-source options but never mentions Cheap Seats, Popcorn, or any alias of the brand.

Yes—though open-source options are often **building blocks for a visibility tracker**, rather than a ready-made marketing dashboard.

Three worth considering:

| Tool | Why use it | What you’ll need to add |
|---|---|---|
| **Promptfoo** | Run a repeatable set of buyer questions across AI providers and evaluate the answers. A practical starting point for tracking brand mentions and competitor inclusion. | Rules for detecting mentions, recommendation position, and cited domains; reporting over time. |
| **DeepEval** | Build custom evaluation metrics in Python—for example, whether an answer recommends your brand and describes it accurately. | A prompt library, scheduled runs, and a marketing-friendly dashboard. |
| **Langfuse** | Store model interactions and evaluation results so you can inspect answers and compare runs over time. | The collection workflow and visibility-specific scoring. It’s primarily an observability platform. |

**My recommendation:** If you have engineering support, start with **Promptfoo** and 50–100 questions your buyers actually ask. Track brand mention rate, competitor mention rate, recommendations, and citations to your website. Keep the questions fixed and rerun them weekly.

One important limitation: **testing a model’s API is not the same as measuring visibility inside ChatGPT Search, Google AI Overviews, or Perplexity’s consumer experience.** Search retrieval, personalization, and citations can differ. Check which experience a tracker actually measures before treating its scores as search visibility.

These tools’ open-source status and capabilities reflect what I know; I haven’t checked their latest versions or licenses.

The answer names only Elmo and GetCito as open-source AI visibility tools and never mentions Cheap Seats, its aliases, or Popcorn. The brand is therefore absent.

**Yes—there are open-source tools for measuring AI search visibility**, often called GEO (generative engine optimization) or AEO (answer engine optimization).

Two worth shortlisting:

| Tool | What it measures | Why consider it |
|---|---|---|
| **[Elmo](https://github.com/elmohq/elmo)** | Brand mentions, citations, competitor share of voice, sentiment, and visibility trends across engines including ChatGPT, Perplexity, Gemini, and Google AI Overviews. | My first choice to evaluate for a marketing dashboard. It’s MIT-licensed, self-hostable, and lets you inspect the responses behind its metrics. |
| **[GetCito](https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool)** | Scheduled prompt tracking, brand and competitor mentions, cited sources, and reports across multiple AI search surfaces. | Worth evaluating if you want flexible data-provider integrations or an agency deployment with custom branding. |

**Open source doesn’t mean zero running costs.** Expect hosting, setup, and model or search-data fees. Elmo, for example, requires provider or scraper credentials plus a separate key for its mention and sentiment judging service. [Elmo setup requirements](https://github.com/elmohq/elmo#quick-start)

For a practical pilot, I’d start with **Elmo**, 25–50 questions your buyers actually ask, and three competitors. Track brand mention rate, citation rate, and competitor share of voice weekly, keeping the prompts consistent.

Before adopting either tool, verify whether each connector measures the **consumer search experience or an API-generated answer**—GetCito supports both scraping providers and direct model calls. Treat your results as visibility across your chosen question set, rather than a measure of total market exposure. [GetCito integrations](https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool#features)

Searches: “open source AI search visibility tracking GEO GitHub tools | open source llm brand visibility tracker GitHub”

Cited: https://github.com/elmohq/elmo https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool https://github.com/elmohq/elmo#quick-start https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool#features

The answer never names Cheap Seats, Popcorn, or any alias; it recommends promptfoo as the closest fit and suggests building a DIY stack. The brand is absent from the answer.

**Short answer: there's no mature open-source equivalent yet.** A handful of small GitHub projects call themselves "AI visibility trackers" or "GEO monitors," but as of my knowledge they're early, thinly maintained, and not close to Profound or Peec in coverage or polish. I can't verify what's appeared recently, so search GitHub for terms like `GEO`, `AI search visibility`, `LLM brand monitoring`, and `AEO`, and check stars and recent commits before you commit to one.

**The practical open-source route is to build a lightweight version yourself.** It's easier than it sounds.

**Recommended stack**

- **[promptfoo](https://github.com/promptfoo/promptfoo)** (MIT): Built for LLM evals, but it repurposes well. You define a prompt set ("best CRM for mid-market SaaS," etc.), run it across OpenAI, Anthropic, Gemini, Perplexity and others, and add assertions such as "response mentions [Brand]" or "mentions competitor X." It also gives you a results dashboard. This is the closest off-the-shelf fit.
- **A simple scheduled script** (Python plus cron or GitHub Actions): Calls each model's API with web search enabled, stores responses, and counts mentions, rank position, sentiment and cited URLs.
- **Storage and dashboard:** Postgres or SQLite, plus **[Metabase](https://github.com/metabase/metabase)** or **Grafana** for share-of-voice trends over time.
- **Optional: [Langfuse](https://github.com/langfuse/langfuse)** if you want tracing and logging of every query and response.

**What you lose versus Profound or Peec**

1. **Real consumer-UI results.** The paid tools capture what users actually see in ChatGPT, Perplexity, AI Overviews and others. API responses, even with search turned on, can differ noticeably. Scraping those UIs yourself is brittle and may violate terms of service.
2. **Prompt volume data.** Profound's estimates of what people actually ask are proprietary. You'll be guessing at prompts.
3. **Citation and source analysis, competitor benchmarking and reporting** all come out of the box with the paid tools.
4. **Maintenance.** Someone has to own the system.

**My recommendation**

- **If you have a developer for a few days:** promptfoo plus a scheduled job plus Metabase gives you roughly 60–70% of the value at near-zero cost, beyond API spend of tens of dollars a month for a few hundred prompts.
- **If you need exec-ready reporting or real-UI accuracy:** pay for it. Peec is the cheaper of the two. Other lower-cost commercial options include Otterly.ai and the AI visibility features in Semrush and Ahrefs, which may already be in your budget.

I can draft a starter promptfoo config or a tracking script for your brand and competitors if that would help.

Cited: https://github.com/promptfoo/promptfoo https://github.com/metabase/metabase https://github.com/langfuse/langfuse

The answer never names Cheap Seats or any of its aliases (including Popcorn); it recommends Elmo as the "best overall pick" and lists other open-source trackers instead.

Yes, there are several. These are the ones I'd look at first, ranked by how close they come to the Profound/Peec feature set:

**1. Elmo (elmohq/elmo): best overall pick**
- Covers most of what Peec does: tracks brand mentions across ChatGPT, Gemini, Claude and Perplexity, shows which competitors appear, and shows which sources the AI cites.
- MIT license, self-hosted, stores data in Postgres. About 480 stars and updated today, so it's actively maintained.
- OpenAlternative lists it first among Profound alternatives.

**2. OneGlanse (oneglanse/oneglanse): closest to Profound's approach**
- Instead of calling the AI APIs, it uses browser automation to collect answers from the real ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews interfaces. That's roughly how Profound works, and it matters because API answers can differ from what users actually see. One comparison describes Peec as API-based and Profound as capturing the consumer interfaces.
- MIT license. The downside: scraping these interfaces is fragile and may break the AI providers' terms of service.

**3. geo-aeo-tracker (danishashko): simplest to run**
- Self-hosted, MIT license, tracks ChatGPT, Perplexity, Gemini and others. Good if you just want a working tracker without much setup.

**Also worth a look:**
- **limelit-co/open:** a single Go program with a built-in SQLite database, so it's very easy to deploy.
- **searchmirror:** combines classic Google rank tracking with AI visibility tracking. AGPL-3.0 license.
- **GetCito** and **Canonry:** both described as open source, but they lean toward a paid service model. Check the licenses before you commit.

**What open source won't give you:**
- **Not free to run.** You bring your own API keys and pay per query. Tracking hundreds of prompts daily across 4–5 AI engines adds up, though it usually still costs well under Profound's pricing.
- **Fewer extras.** Expect no benchmark datasets, data on what people actually ask AI tools (Profound sells this), polished reports or support.
- **Someone has to maintain it.** You'll need an engineer to host it and keep it updated, especially the scraping-based tools.

**My suggestion:** pilot Elmo with 50–100 of your key buyer prompts for a few weeks. If the numbers differ noticeably from what you see when you check ChatGPT yourself, try OneGlanse for its real-interface data. If nobody on your team can host software, a cheaper paid tool will probably serve you better than open source.

Star counts and update dates come from GitHub as of today. Check each repo's license and recent activity before you adopt it.

Sources:
- [GitHub topic: ai-visibility](https://github.com/topics/ai-visibility)
- [OpenAlternative – Profound alternatives](https://openalternative.co/alternatives/profound)
- [AlternativeTo – Profound alternatives](https://alternativeto.net/software/profound)
- [Superlines vs Profound vs Peec AI](https://www.superlines.io/articles/superlines-vs-profound-vs-peec-ai)

Searches: “open source AI visibility tracking GEO tool alternative to Profound Peec AI github”

Cited: https://github.com/topics/ai-visibility https://openalternative.co/alternatives/profound https://alternativeto.net/software/profound https://www.superlines.io/articles/superlines-vs-profound-vs-peec-ai

The answer never names Cheap Seats or any of its aliases. It recommends promptfoo as the strongest starting point and mentions Langfuse only as a tracking layer, so the brand is absent.

**Yes, but the open-source options I know are closer to DIY foundations than direct replacements for Profound or Peec AI.** If you want to track how often your brand appears in AI answers, you can build that with:

| Tool | Why use it | What you’ll need to build |
|---|---|---|
| **promptfoo** | Open-source framework for running the same prompts across multiple AI models and evaluating responses. The strongest starting point for a recurring brand visibility benchmark. | Brand and competitor detection, citation extraction, scheduling, and reporting. |
| **Langfuse** | Open-source platform for recording LLM responses, managing prompt datasets, and tracking evaluations over time. Useful as the tracking layer for a custom system. | The monitoring workflow and marketing-specific metrics. |

**My recommendation:** start with **promptfoo** if you have engineering help. Run a representative set of buyer questions weekly, then measure brand mentions, competitor mentions, recommendation position, and cited domains.

The key limitation: **model API responses aren’t equivalent to answers in consumer products such as ChatGPT or Google AI Overviews.** Search, personalization, and product-specific behavior can change results. A DIY benchmark is useful for directional tracking, but it won’t automatically reproduce the coverage of a dedicated visibility platform.

If your priority is a ready-to-use marketing dashboard with minimal maintenance, I’m not aware of a mature open-source equivalent I’d confidently recommend. Open source also doesn’t eliminate model API and hosting costs.

The answer never names Cheap Seats, CheapSeats, or Popcorn; it recommends Elmo as the first open-source alternative and lists Habibi and Aperture as smaller options. The brand is simply not mentioned.

**Yes—Elmo is the first open-source alternative I’d evaluate.** There are also smaller projects if you mainly need basic brand and citation tracking.

| Tool | Why consider it | Main tradeoff |
|---|---|---|
| **[Elmo](https://github.com/elmohq/elmo)** | MIT-licensed and self-hostable. Tracks brand mentions, citations, sentiment, competitor share of voice, and visibility trends. Includes shareable reports and content opportunities. | You’ll need technical help to host and maintain it, plus budget for data collection. |
| **[Habibi](https://github.com/moaljumaa/habibi)** | A simpler self-hosted tracker with guided setup and access to OpenAI, Claude, Gemini, and Perplexity models through OpenRouter. Useful for a focused monitoring pilot. | Its API-based approach needs validation against the consumer experiences your buyers actually use. |
| **[Aperture](https://github.com/anyin-ai/aperture)** | MIT-licensed, bring-your-own-key monitoring with competitor comparisons and citation tracking. Good for a technical team wanting to customize audits. | Early-stage: OpenAI and Perplexity are supported; Google AI Overviews, Claude, and sentiment analysis are listed as planned. |

**The key buying question is how each tool collects answers.** API-generated responses can differ from what buyers see in ChatGPT or Google’s AI search experiences. Elmo says it combines consumer-app scraping with API calls; confirm which collection methods are available in your self-hosted setup. [Elmo’s collection overview](https://www.elmohq.com/)

For a product marketing team, I’d **pilot Elmo with 30–50 high-intent buyer questions and three competitors for two weeks**. Check whether the citations and competitive gaps produce useful content or PR actions before replacing Profound or Peec. Open source can remove the software subscription, but hosting, model calls, collection services, and maintenance still cost money.

Searches: “open source alternative Profound Peec AI GEO AI visibility tracking GitHub | Profound Peec AI features AI search visibility tracking” “”

Cited: https://github.com/elmohq/elmo https://github.com/moaljumaa/habibi https://github.com/anyin-ai/aperture https://www.elmohq.com/

The brand Cheap Seats (aliases CheapSeats, Popcorn, etc.) is never mentioned in the answer; it only names other trackers like Profound, Peec AI, and Otterly.AI.

You can do this in-house. One constraint first: **to measure what AI assistants say, you have to query them**, so your tracked prompts and brand names go to OpenAI, Google, Perplexity, and similar providers. That's usually acceptable because it's the same as a user asking. What you avoid is a third-party tracker holding your prompt strategy, competitor list, and results. Check whether that matches your team's concern.

## Option 1: Build a lightweight tracker (most common)

This is a few days of work for one engineer or a technical marketer.

**Pipeline:**
1. **Prompt set:** 50–300 buyer-style questions ("best X for Y," "X vs competitor," category questions), tagged by funnel stage and persona.
2. **Query the engines through their APIs, with web search turned on** so results match what users see:
   - OpenAI Responses API with the `web_search` tool
   - Google Gemini API with Google Search grounding
   - Perplexity Sonar API, which returns citations directly
   - Anthropic Claude API with the web search tool
3. **Run each prompt 3–5 times.** Answers vary between runs, so a single run is noise.
4. **Parse the responses** for brand and competitor mentions, mention position, which URLs are cited, and sentiment or accuracy. Use regex for mentions and a cheap LLM pass for sentiment and accuracy.
5. **Store and visualize** in your existing warehouse (Postgres, BigQuery, Snowflake) with Metabase, Superset, or Looker.
6. **Schedule** weekly with cron, Airflow, or GitHub Actions.

**Metrics to report:** mention rate, share of voice against competitors, citation share for your domains, average position, and sentiment or accuracy flags.

## Option 2: Use open-source tooling

- **promptfoo:** an open-source LLM eval framework that runs locally. It can run one prompt set across many providers and apply assertions such as "response contains Brand." It's useful if you want a working tracker without writing much code.
- **Langfuse (self-hosted):** stores and browses runs if you want trace-level history.
- **GitHub "GEO/AI visibility tracker" projects:** several exist, but their maturity varies widely. Treat them as starting code, not products.

## Option 3: First-party signals (no external queries)

These complement prompt tracking and keep everything on your own infrastructure:
- **AI referral traffic:** segment visits from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Use your analytics tool, or a self-hosted one like Plausible or Matomo if GA is also a concern.
- **AI crawler logs:** check your server or CDN logs for GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, and Google-Extended. This shows which pages AI systems fetch, and user-triggered fetches like ChatGPT-User suggest real-time citation.

## Caveats
- **API answers don't match consumer apps exactly.** The apps add personalization, memory, and different system prompts. Use the API data for trends and competitive share, not as an exact replica.
- **Google AI Overviews and AI Mode have no official API.** Your choices are a SERP API vendor (which reintroduces a third party), running your own headless browser (fragile and a ToS gray area), or covering them with manual spot checks.
- **Costs** are modest. Hundreds of prompts × 4 engines × 3 runs per week typically comes to tens to low hundreds of dollars a month in API fees.

## Recommendation
Start with **Option 1, plus referral and crawler tracking from Option 3**. Prototype with promptfoo or a short Python script against Perplexity Sonar and OpenAI, put the results in your warehouse, and add engines once the prompt set is stable. If the effort turns out to be more than you want to maintain, reconsider a SaaS tracker (Profound, Peec AI, Otterly). You'd then be deciding with a clear view of exactly what data you'd be sharing.

The answer never names Cheap Seats, CheapSeats, Popcorn or any other alias; it recommends Elmo as the pilot option. The brand is absent, so the verdict is absent.

You can do this without a visibility SaaS vendor, but your prompts still have to go to the AI engines themselves. To see what ChatGPT or Perplexity says about you, you have to ask them. What you can control is everything else: your prompt list, your competitor list, the stored answers and the scoring. Each of the options below keeps those in your own systems.

## 1. Self-host an open-source tracker (fastest)

- **[Elmo](https://github.com/elmohq/elmo)** ([site](https://www.elmohq.com/)): this is the most complete option I found. It's MIT-licensed and runs on Docker Compose (web app, worker and Postgres). Your prompts, competitors and raw answers are stored in a Postgres database you control. It covers ChatGPT, Claude, Perplexity, Gemini, Copilot, Grok, Google AI Mode and AI Overviews. It tracks visibility over time, sentiment and where you rank in each answer. It has a lot of activity (about 480 stars and 1,100+ commits) and they also sell a hosted tier, but self-hosting is free.
  - **Check before you adopt it:** It reaches some engines through scraping providers and others through model APIs. Its README also says the mention and sentiment scoring needs a third-party API key, and answer text gets sent to that service. Have someone read the code to see exactly what leaves your network. Also check whether you can point the scoring at your own model, which is what you'd want.
- **[OneGlanse](https://www.producthunt.com/products/oneglanse)**: it's free, runs locally and has no cloud version. It covers ChatGPT, Gemini, Perplexity, Claude and AI Overviews. It reads the answers people actually see in the apps instead of API responses, which is more realistic. The tradeoff is that you have to manage logged-in browser sessions, so it's aimed at technical users. Check the license on its GitHub before using it.
- **Others to watch:** [GEO/AEO Tracker](https://github.com/danishashko/geo-aeo-tracker) stores data in your browser but needs a paid Bright Data key to query the engines. [GetCito](https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool) has an unclear license. There are smaller projects under GitHub's [ai-visibility-tracking](https://github.com/topics/ai-visibility-tracking?o=desc&s=stars) and [generative-engine-optimization](https://github.com/topics/generative-engine-optimization) topics. Many are very early, so check how active they are.

## 2. Build a small version yourself (most control)

A data or marketing ops engineer could build this in a few days:
1. Write 50–200 prompts that buyers would actually ask.
2. Run them on a schedule through the official OpenAI, Anthropic, Gemini and Perplexity APIs. Perplexity's API returns citations.
3. Store the raw answers in your own warehouse.
4. Score each answer for whether you're mentioned, your position, sentiment and which sources are cited. Use simple string matching plus a local model (for example, through Ollama), so the scoring never leaves your network.
5. Build dashboards in your existing BI tool.

API answers aren't identical to what people see in the ChatGPT or Gemini apps, so treat them as a trend signal, not an exact copy.

## 3. Measure the results with data you already have

- **Referral traffic:** Track visits from chatgpt.com, perplexity.ai, gemini.google.com and copilot in your existing analytics. If you want that self-hosted too, use Matomo or Plausible.
- **Server logs:** Watch for AI crawlers such as GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot. That tells you whether AI engines can actually read your content. Self-hosted log analyzers like Logwick can sort out this bot traffic.

## My recommendation

Pilot Elmo on your own infrastructure. Then either swap its scoring step for a model you host or confirm that what it sends out is acceptable. Use your web analytics referrals as the business metric. If security won't allow any third-party scraping providers, build the version in option 2 on official APIs.

Sources:
- [Elmo GitHub](https://github.com/elmohq/elmo) / [elmohq.com](https://www.elmohq.com/)
- [OneGlanse on Product Hunt](https://www.producthunt.com/products/oneglanse)
- [GEO/AEO Tracker](https://github.com/danishashko/geo-aeo-tracker)
- [GetCito](https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool)
- [GitHub topic: ai-visibility-tracking](https://github.com/topics/ai-visibility-tracking?o=desc&s=stars)
- [GitHub topic: generative-engine-optimization](https://github.com/topics/generative-engine-optimization)
- [Bright Data: Open-source AI visibility tracker](https://brightdata.com/blog/ai/ai-visibility-tracker)
- [LLMrefs: AI SEO tools list](https://llmrefs.com/blog/ai-seo-tools-list)

Searches: “open source self-hosted AI search visibility tracking tool GEO LLM brand mentions” “github open source generative engine optimization monitor ChatGPT Perplexity brand visibility”

Cited: https://github.com/elmohq/elmo https://www.elmohq.com/ https://www.producthunt.com/products/oneglanse https://github.com/danishashko/geo-aeo-tracker https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool https://github.com/topics/ai-visibility-tracking?o=desc&s=stars https://github.com/topics/generative-engine-optimization https://brightdata.com/blog/ai/ai-visibility-tracker https://llmrefs.com/blog/ai-seo-tools-list

The answer never names Cheap Seats, cheapseats.dev, Popcorn, or any other alias of the brand. It recommends building a private pipeline in Python, PostgreSQL, and Metabase, with manual audits and local models as alternatives.

You have three practical options. The key distinction is whether your restriction covers **monitoring data** or also the **prompts sent to AI providers**.

| Option | Tools and approach | Best fit | Main tradeoff |
|---|---|---|---|
| **Run a private monitoring pipeline** | Use **Python**, **PostgreSQL**, and **Metabase** on your infrastructure. Query AI providers directly and store answers, mentions, and citations internally. | Repeatable tracking without an additional monitoring SaaS vendor | Prompts still reach the AI providers; API answers may differ from consumer chat experiences |
| **Audit consumer AI products manually** | Run a fixed prompt list in **ChatGPT, Google AI search experiences, Perplexity, and Claude**, then record results in an internal spreadsheet or database. | A pilot or a smaller set of commercially important questions | Labor intensive; results vary by account, location, and time |
| **Use entirely local models** | Run models with **Ollama** for straightforward setup or **vLLM** for higher throughput, with internal reporting. | A requirement that no prompts or answers leave your environment | Measures your local models—not your visibility in public AI products |

**I’d start with a private pipeline plus a small manual audit.** Keep the prompt library, raw responses, scoring, and dashboards in your infrastructure. Send only generic buyer questions to the AI providers—no customer records, confidential strategy, or internal documents.

Build the measurement around:

- **Brand mention rate:** How often does your brand appear?
- **Recommendation rate:** How often is it recommended, rather than merely mentioned?
- **Citation rate:** How often are your pages cited?
- **Competitive share:** How often do you appear relative to named competitors?
- **Accuracy:** Are descriptions of your product correct?

Use a fixed set of buyer questions, repeat each question several times, and save the date, model, search settings, and full response. Treat API results and consumer-product audits as separate datasets; neither is a definitive “AI ranking.”

If **zero external data transmission** is mandatory, public AI visibility cannot be measured directly without an exception: even submitting a test question transmits data. Local-model monitoring can support internal testing, but it cannot tell you what public AI products recommend.

The answer never names Cheap Seats, CheapSeats, Popcorn, or any listed alias; it recommends Elmo, Gego and OneGlanse instead.

**Yes—self-hosting can keep your visibility history and competitive reporting out of a tracking SaaS.** The distinction is that prompts still reach the AI services you’re measuring. “Self-hosted” doesn’t automatically mean “no external data sharing.”

Here are the practical options:

| Option | Why consider it | Main tradeoff |
|---|---|---|
| **[Elmo](https://github.com/elmohq/elmo), self-hosted** | A ready-made marketing dashboard for brand mentions, citations, share of voice, and trends. MIT licensed; supports both website collection and model APIs. | Check the collection and analysis integrations you enable: external scrapers, routers, and model providers can introduce additional recipients of your data. |
| **[Gego](https://getgego.org/)** | A good engineering-led option: scheduled prompts, brand aliases, citation tracking, dashboard, and API. Supports direct providers and Ollama. | Primarily measures configured model endpoints; don’t assume their answers reproduce the consumer websites. You operate the infrastructure. |
| **[OneGlanse](https://github.com/oneglanse/oneglanse)** | Worth evaluating when you specifically want answers from the actual ChatGPT, Gemini, Perplexity, Claude, and Google AI Overview interfaces. | Requires accounts and browser sessions, with proxies for server deployments. By default, captured answers go to a configured analysis model; it also documents PostHog usage telemetry. |
| **Build a small internal tracker** | Best control over exactly what leaves your environment. Schedule public buyer questions, query approved providers directly, and retain answers and reporting internally. | Your team owns collection, extraction, maintenance, and measurement quality. |
| **Manual checks with an internal spreadsheet** | Best starting point if engineering capacity is limited. No additional tracking vendor is required. | Labor intensive and harder to repeat consistently. |

The product capabilities above come from their current documentation; treat them as a shortlist to pilot, rather than a guarantee of production readiness.

**Choose based on your privacy boundary:**

- **No additional tracking vendor:** Self-hosting is a reasonable fit. Use direct provider connections where available, and review telemetry, scraping services, proxies, and external scoring.
- **No confidential information leaving your environment:** Use only public buyer questions externally. Keep customer data, strategy notes, and internal segmentation in your own database.
- **Absolutely no external processing:** Local models can support internal experiments, but they cannot tell you what today’s hosted AI products actually recommend. Measuring those products requires interacting with them.

**My recommendation:** Pilot **self-hosted Elmo** if you want a marketing dashboard quickly, or **Gego** if your engineering team prefers direct API measurement and customization. Evaluate **OneGlanse** when consumer-interface fidelity is essential, accounting for its extra operational and data-flow requirements.

Start with 30–50 fixed buyer questions across your priority engines. Track mention rate, recommendations, competitor mentions, and cited URLs; retain the raw answers. Repeat questions across runs and keep language, location, and collection method consistent. Report these as **visibility within your tracked question set**, rather than market-wide audience share. Budget for hosting, provider usage, and maintenance even when the software is free.

Searches: “open source self hosted AI answer visibility tracking GEO monitoring Peec alternative | GitHub open source LLM brand visibility monitoring self hosted” “”

Cited: https://github.com/elmohq/elmo https://getgego.org/ https://github.com/oneglanse/oneglanse

The answer never names Cheap Seats, its aliases, or cheapseats.dev/Popcorn, so the brand is absent.

**Buy, for most teams.** Build only if you have engineering capacity and a specific need the tools don't cover. A cheap hybrid is a good option too.

## Why buying usually wins

**1. The APIs don't show what your customers see.** This is the biggest issue. Calls to the OpenAI and Anthropic APIs return the raw model by default. Consumer ChatGPT and Claude add things the API doesn't:
- Live web search and retrieval, which drive most brand and product recommendations
- Their own system prompts, memory, and personalization
- Different model routing and versions

You can turn on web search tools in the APIs, but the results still won't match the consumer apps. Many commercial trackers query the actual consumer interfaces, or reproduce them more closely, and they also cover Google AI Overviews/AI Mode, Perplexity, Gemini, and Copilot. A homegrown API tracker can give you a misleading picture with false precision.

**2. Measurement is harder than it looks.** LLM answers vary from run to run, so you need:
- Many runs per prompt to get stable share-of-voice numbers
- Entity matching (brand name variants, product names, competitors)
- Sentiment and position scoring
- Citation and source extraction, which tells you which pages to influence
- History tracked across model updates

That's months of work plus ongoing maintenance as models and interfaces change.

**3. Total cost.** API tokens are cheap. Engineer time, dashboards, and upkeep are not. Most tools cost less than a few days of an engineer's time per month.

## Tools to evaluate
- **Profound**: Enterprise-grade with deep citation and source analysis. Strong if AI search is a board-level metric.
- **Peec AI**: Clean, mid-market priced, good competitor benchmarking. A solid default.
- **Scrunch AI**: Good for seeing how AI crawlers and agents read your site, not just mentions.
- **Otterly.ai**: Inexpensive and simple. Good for starting out or for small teams.
- **Semrush AI Toolkit / Ahrefs Brand Radar**: Worth it if you already pay for these suites, since it consolidates your SEO and GEO reporting.

Pricing and features change quickly in this category. Run 2–3 trials at the same time on the same set of 30–50 prompts and compare.

## When building makes sense
- You need tracking inside your own product or data warehouse, or for many clients (agencies).
- You care specifically about **raw model knowledge**, meaning what the model "believes" about you without search. The API is actually the right tool for that.
- You have very custom prompt sets, languages, or regulated-data constraints.

## Practical recommendation
1. **Buy** one tool (Peec or Otterly to start, Profound if enterprise) for consumer-facing visibility and citations.
2. **Optionally build** a lightweight script on the APIs, a day or two of work, to track baseline model perception monthly. Use a fixed prompt set, multiple runs, and log the outputs to a sheet. It complements the tool rather than replacing it.
3. Reassess in 6 months. This market is consolidating fast.

If you share your budget, team size, and whether you're B2B or B2C, I can narrow the shortlist.

The brand Cheap Seats (aliases CheapSeats, Popcorn, etc.) is not named anywhere in the answer; the answer recommends buying Otterly.AI, Peec AI and Profound instead.

I'd buy rather than build. If you want a build component, keep it small and use it for experiments alongside the tool, not as your main tracker.

**Why building on the APIs is the wrong foundation**

The OpenAI and Anthropic APIs don't return what your buyers see in the ChatGPT or Claude apps. The consumer apps run with different system prompts and settings, they turn web search on by default, and they show citations and shopping cards. The API doesn't do those things unless you set it up to. Several studies found the API and the app often name different brands:
- Surfer found only **24% brand overlap** and **4% source overlap** between API answers and answers collected from the ChatGPT interface ([Surfer](https://surferseo.com/blog/llm-scraped-ai-answers-vs-api-results)).
- Brandlight compared 2,565 matched prompts and found **6–12% brand overlap** depending on the engine ([Brandlight](https://research.brandlight.ai/ui-vs-api.html)).
- Some vendors have stopped using APIs as their main data source and collect answers through a browser instead ([Wellows](https://wellows.com/blog/ai-visibility-measurement/)).

Most of these studies come from vendors who sell interface-based tracking, so the exact numbers may be inflated. The direction is consistent across all of them, though. A tracker built on the API would be cheap to run but would measure the wrong thing. Collecting from the apps yourself means running and maintaining browser automation against interfaces that change often, possibly in breach of their terms of service. That's work for a full engineering team, not a marketing side project.

**What you'd need to build to make it useful**
- Prompt sets and personas
- Multiple runs per prompt, since answers vary from run to run. One run a week is a poll, not a measurement ([Windsor](https://windsor.ai/ai-visibility-tracking-accuracy/)).
- Brand and competitor detection, citation parsing and sentiment
- Dashboards and alerts
- Coverage of Google AI Overviews, Perplexity and Gemini, which you'd probably want too

All of that has to be maintained over time. Tools cost about $30–500 a month, which is less than a week of an engineer's time.

**What to buy (2026 list prices; check before you buy)**
- **Otterly.AI**: $29 / $189 / $489 a month. The cheapest way to start getting data quickly ([source](https://ainora.lt/blog/best-ai-visibility-tracking-tools-2026)).
- **Peec AI**: about $95 / $245 / $495 a month, or around $80 / $205 / $420 billed annually. Good value for the number of engines it covers, and it handles competitor benchmarking well ([source](https://blog.timsoulo.com/14-peec-ai-alternatives-for-ai-search-visibility-tracking-2026/)).
- **Profound**: Starter is $99 a month but covers ChatGPT only. Growth is $399 a month for three engines, and Enterprise pricing is custom. The strongest enterprise option ([source](https://www.stork.ai/blog/profound-vs-otterly-vs-peec)).
- **Scrunch**: from about $250 a month. Sitecore bought it in June 2026, which could matter if you use Sitecore or want to avoid it ([source](https://kime.ai/blog/best-ai-visibility-tools)).

**Before you sign, ask each vendor:**
1. Do they collect answers from the app interface or the API, and for which engines?
2. How many times do they run each prompt, and how often?
3. Do they cover Claude? Most tools focus on ChatGPT, Perplexity, Google AI Overviews and Gemini. If Claude matters to your audience, check this specifically.
4. How do they price by number of prompts and engines? The advertised starting price rarely covers what a team actually needs.

**When building does make sense**
- You need to track Claude and no vendor covers it well.
- You want to test something specific, for example whether a content change shifts how the models describe you. A short script on the APIs that runs about 50 prompts several times each is fine for that, as long as you remember it shows what the API says, not what users see.
- You have data rules that stop you sending prompts or brand data to a third party.

**What I'd do:** run trials of Peec and Otterly (or Profound if you have enterprise budget) on the same 30–50 prompts for two weeks. Pick the one whose results match what you see when you check the apps by hand. Only build if there's a gap, like Claude coverage, that none of them fill.

Sources:
- [Surfer: LLM scraped answers vs API results](https://surferseo.com/blog/llm-scraped-ai-answers-vs-api-results)
- [Brandlight: UI vs API](https://research.brandlight.ai/ui-vs-api.html)
- [Wellows: AI visibility measurement](https://wellows.com/blog/ai-visibility-measurement/)
- [Windsor.ai: AI visibility tracking accuracy](https://windsor.ai/ai-visibility-tracking-accuracy/)
- [Superlines: API vs UI data](https://www.superlines.io/articles/api-vs-ui-data-ai-visibility-tools/)
- [Metehan: How AI visibility tools collect data](https://metehan.ai/articles/how-ai-visibility-tools-collect-data)
- [Ainora: verified pricing](https://ainora.lt/blog/best-ai-visibility-tracking-tools-2026)
- [Tim Soulo: Peec AI alternatives](https://blog.timsoulo.com/14-peec-ai-alternatives-for-ai-search-visibility-tracking-2026/)
- [Stork: Profound vs Otterly vs Peec](https://www.stork.ai/blog/profound-vs-otterly-vs-peec)
- [Kime: Best AI visibility tools](https://kime.ai/blog/best-ai-visibility-tools)
- [Ryze: pricing compared](https://www.get-ryze.ai/blog/ai-visibility-tools-pricing-compared-2026)

Searches: “AI visibility tracking tools pricing 2026 Profound Peec AI Otterly Scrunch” “API vs ChatGPT interface responses differ AI visibility tracking accuracy scraping front-end”

Cited: https://surferseo.com/blog/llm-scraped-ai-answers-vs-api-results https://research.brandlight.ai/ui-vs-api.html https://wellows.com/blog/ai-visibility-measurement/ https://windsor.ai/ai-visibility-tracking-accuracy/ https://ainora.lt/blog/best-ai-visibility-tracking-tools-2026 https://blog.timsoulo.com/14-peec-ai-alternatives-for-ai-search-visibility-tracking-2026/ https://www.stork.ai/blog/profound-vs-otterly-vs-peec https://kime.ai/blog/best-ai-visibility-tools https://www.superlines.io/articles/api-vs-ui-data-ai-visibility-tools/ https://metehan.ai/articles/how-ai-visibility-tools-collect-data https://www.get-ryze.ai/blog/ai-visibility-tools-pricing-compared-2026

The brand Cheap Seats (and its aliases such as Popcorn or CheapSeats) is never named in the answer; the shortlist covers Profound, Scrunch AI, and Otterly.AI instead.

**Buy first, unless tracking AI visibility is itself a core capability you want to own.** For a marketing team, the hard part is maintaining a useful measurement program—not making API calls.

The main caveat: **OpenAI and Anthropic API responses don’t reliably reproduce what users see in ChatGPT or Claude.** Consumer products can use different search, personalization, context, and routing. An API tracker measures your chosen test conditions, not your actual share of consumer conversations. Vendors deserve the same scrutiny.

I’d shortlist these tools for a pilot, subject to verifying their current coverage:

| Tool | Why consider it |
|---|---|
| **Profound** | For enterprise teams seeking deeper brand visibility, citation analysis, and reporting. |
| **Scrunch AI** | For teams connecting AI visibility monitoring with website and content improvements. |
| **Otterly.AI** | For a more focused starting point for prompt and citation monitoring. |

**Build when** you need proprietary prompts tied to your customer journey, custom experiments, tight integration with internal data, or unusually strict control over data and methodology. Budget for prompt maintenance, repeated sampling, source extraction, entity matching, dashboards, and model changes—not just API usage.

A practical decision process:

1. **Run a four-week paid pilot.** Use 50–100 prompts covering purchase questions, comparisons, category discovery, and branded questions.
2. **Check the methodology.** Ask which consumer experiences versus APIs are measured, how search and geography are handled, and whether raw answers, citations, timestamps, and exports are available.
3. **Measure useful outcomes.** Track mention rate, recommendation placement, cited sources, and factual accuracy. Treat sentiment scores cautiously.
4. **Require actionable findings.** The tool should reveal content gaps, inaccurate claims, or sources worth influencing. A visibility score alone isn’t enough.

**My recommendation:** buy monitoring, keep ownership of your prompt set and exported data, and build a small custom testing layer only where the vendor falls short. Don’t fund a full internal platform until the pilot shows that its findings change marketing decisions.

The brand Cheap Seats (aliases CheapSeats, Popcorn, etc.) is never named in the answer; the shortlist covers Peec AI, Profound, and OtterlyAI only.

**Buy a tool first. Build a small API testing layer later if you need custom analysis.** For most product marketing teams, maintaining measurement infrastructure is a poor use of engineering time.

The biggest issue is **what you’re measuring**. Calling OpenAI and Anthropic APIs tells you how your brand appears under your chosen API configuration. It doesn’t establish what buyers see in ChatGPT or Claude, where search, conversation history, personalization, and product settings can differ. Both APIs now offer web search with citations, so a useful testing harness is feasible—but it should be treated as a controlled benchmark. [OpenAI documentation](https://developers.openai.com/api/docs/guides/tools-web-search), [Anthropic documentation](https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool)

| Buy when… | Build when… |
|---|---|
| You need recurring marketing reporting across multiple AI products. | You specifically want controlled tests of API responses. |
| You need competitor comparisons, citation analysis, and reports quickly. | You have unusual scoring, proprietary prompts, or data integration requirements. |
| Engineering has more valuable product work. | Someone can own sampling, extraction quality, model changes, and ongoing maintenance. |

Cheap API calls don’t necessarily make a cheap tracker. The recurring work is scheduling repeated samples, resolving brand names correctly, separating mentions from recommendations, preserving source answers, and explaining changes when models update.

For a shortlist:

- **Peec AI:** Start here for a focused marketing evaluation. Its offering includes visibility, sentiment, share of voice, and source analysis across multiple engines. Confirm the specific plan includes Claude if that matters to you. [Peec AI](https://peec.ai/for-enterprise)
- **Profound:** Evaluate it if you need deeper reporting and content optimization alongside monitoring. Its advertised starting price is $99/month, but engine coverage and reporting capabilities vary by tier. [Profound’s offering](https://www.tryprofound.com/blog/choosing-ai-visibility-provider)
- **OtterlyAI:** Consider it for a low-cost pilot; advertised pricing starts at $29/month. Check prompt limits and engine coverage before comparing costs. [OtterlyAI](https://otterly.ai/)

**Run a 30-day pilot before signing an annual contract.** Use 30–50 real buyer questions drawn from sales calls and customer research, covering discovery, comparisons, and purchase decisions. Track your brand and three competitors.

Require the vendor to demonstrate:

1. **Collection method:** Consumer interface or API, which model, search mode, geography, and sampling frequency.
2. **Auditability:** Raw answers, timestamps, citations, and exports.
3. **Useful measurement:** Separate brand mentions, explicit recommendations, and citations to your site; show trends across repeated samples.
4. **Actionability:** Identify specific content gaps or influential sources your team can address.

Treat visibility scores as **share of answers in your sampled questions**, not market share or revenue attribution. Pair them with referral traffic, conversions, and customer feedback.

My recommendation: **pilot Peec AI and one alternative, buy the one that produces credible evidence and useful actions, and keep ownership of your prompt set and exported data.** Build only after the pilot reveals a concrete requirement the tools cannot meet.

Searches: “AI visibility tracking tool Profound Peec AI pricing ChatGPT Claude monitoring | site.platform.openai.com docs web search responses API | site.platform.claude.com docs web search tool API” “site.peec.ai browser API tracking real user pricing | site.tryprofound.com pricing starter growth enterprise | site.otterly.ai pricing AI monitoring”

Cited: https://developers.openai.com/api/docs/guides/tools-web-search https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool https://peec.ai/for-enterprise https://www.tryprofound.com/blog/choosing-ai-visibility-provider https://otterly.ai/

The answer never names Cheap Seats, CheapSeats, Popcorn, or any other alias of the brand. It suggests a DIY stack and several SaaS tools, none of which is the brand.

I'm not aware of a mature, widely adopted open-source AI visibility tracker that runs in Docker and matches the commercial tools. There are scattered GitHub projects, but I can't vouch for any of them by name. I'd rather not point you at something that may be abandoned or misdescribed. To check what exists now, search GitHub topics like `generative-engine-optimization`, `geo`, `llm-seo`, or `ai-visibility`. Filter by recent commits and stars.

## Your realistic options

**1. Build a lightweight one yourself (a day or two of work)**
The core of these tools is simple, and it all runs in Docker:
- **n8n** (self-hosted, official Docker image) runs on a schedule. It sends your prompt set ("best CRM for startups," etc.) to the OpenAI, Anthropic, Gemini, and Perplexity APIs.
- **Postgres** stores each response, the model, the date, and whether your brand and competitors were mentioned, plus position and cited URLs.
- **Metabase** or **Grafana** shows dashboards for share of voice, mention rate over time, and citation sources.

Use Perplexity's API (or models with web search enabled) if you care about citations. Those answers reflect live retrieval, which is where most visibility shifts happen.

**2. Use a SaaS tool if self-hosting isn't a hard requirement**
- **Peec AI**: clean share-of-voice tracking across ChatGPT, Perplexity, and Gemini, at reasonable mid-market pricing.
- **Profound**: enterprise-grade and the deepest analytics, but expensive.
- **Otterly.ai**: cheap and simple, good for getting started.
- **Semrush AI Toolkit** or **Ahrefs Brand Radar**: worth checking if you already pay for one of these.

## One caveat for any approach
API responses don't perfectly match what users see in the ChatGPT or Gemini apps, which add personalization, memory, and different system prompts. The SaaS vendors have the same problem, though some scrape the consumer UIs. Treat the numbers as directional trends, not exact truth.

If you want, I can scaffold the DIY version for you: a `docker-compose.yml` with n8n, Postgres, and Metabase, plus a starter workflow and schema. Tell me which AI engines and how many prompts you want to track.

The answer never names Cheap Seats, Popcorn, or any other alias of the brand; it recommends Elmo as the first choice and lists other trackers as alternatives.

Yes. **Elmo** ([github.com/elmohq/elmo](https://github.com/elmohq/elmo)) is the one I'd try first. Of the options I found, it's the only one where I confirmed Docker Compose support from the repo itself.

**Elmo**
- **What it tracks:** How ChatGPT, Google AI Mode, AI Overviews, Gemini, Perplexity and Copilot mention your brand, which competitors show up, and which sources get cited. Those engines are captured through scraping providers. Claude, Grok and other models are covered through direct API calls with web search turned on.
- **Setup:**
  ```
  npm install -g @elmohq/cli
  elmo init
  elmo compose up -d
  ```
  Then open `http://localhost:1515`. Postgres is part of the stack.
- **License and cost:** MIT. The self-hosted version is free with unlimited prompts. It has about 400 GitHub stars, so it's early but gaining users.
- **What you'll pay for:** Your own API keys for an AI provider or scraping service, plus a small charge for the step that judges mentions and sentiment. The README estimates that at roughly $0.15–0.35 per 1,000 answers. Self-hosting doesn't remove inference costs, it just lets you control them.

**Alternatives worth a look**
- **Lettertrace:** MIT core, bring your own keys. It's a Next.js app built on Supabase, and there's a [Railway template](https://railway.com/deploy/lettertrace-aeo-monitor) that bundles Postgres. One [third-party tutorial](https://aiindigo.com/tutorials/getting-started-with-lettertrace-audit-ai-recommendations-in-production) mentions `docker-compose up`, but I couldn't confirm that in the official repo.
- **Limelit Open** ([pkg.go.dev](https://pkg.go.dev/github.com/limelitgeo/open)): Apache-2.0. It ships as a single Go binary, which is easy to put in a container. It covers ChatGPT, Claude, Perplexity, Gemini and Google AI Overview/AI Mode.
- **Aperture** ([anyin-ai/aperture](https://awesome.ecosyste.ms/projects/github.com%2Fanyin-ai%2Faperture)) and **OpenSight** ([docs](https://www.mintlify.com/shuvamk/opensight)): Both MIT and self-hostable. I didn't find documented Docker support for either.
- **OneGlanse** and **Canonry**: OneGlanse is listed on [awesome-open-geo](https://github.com/cloudapp-dev/awesome-open-geo). It's notable because it drives the real chat interfaces in a browser instead of calling APIs, and API answers often differ from what users actually see. Canonry is a very early self-hosted tracker.

**My recommendation:** Start with Elmo. It has the documented Docker path, the widest engine coverage and the most users. Try Limelit if you want something minimal. All of these are young projects, so check recent commit activity before you rely on one for reporting to leadership.

Sources:
- [Elmo GitHub repo](https://github.com/elmohq/elmo)
- [OpenAlternative – Elmo](https://openalternative.co/compare/elmo/vs/notra)
- [AlternativeTo – Elmo](https://alternativeto.net/software/elmo/?p=5)
- [Lettertrace – ScriptByAI](https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/)
- [Lettertrace – ChatGate](https://chatgate.ai/post/lettertrace)
- [Lettertrace Railway template](https://railway.com/deploy/lettertrace-aeo-monitor)
- [AI Indigo Lettertrace tutorial](https://aiindigo.com/tutorials/getting-started-with-lettertrace-audit-ai-recommendations-in-production)
- [Limelit Open](https://pkg.go.dev/github.com/limelitgeo/open)
- [Aperture](https://awesome.ecosyste.ms/projects/github.com%2Fanyin-ai%2Faperture)
- [OpenSight](https://www.mintlify.com/shuvamk/opensight)
- [awesome-open-geo](https://github.com/cloudapp-dev/awesome-open-geo)
- [GitHub topic: brand-monitoring](https://github.com/topics/brand-monitoring)

Searches: “open source self-hosted AI visibility tracker LLM brand mentions Docker GitHub” “open source GEO generative engine optimization tracking tool self-hosted ChatGPT Perplexity brand visibility” “Elmo open source AI visibility self-hosted docker compose github” “Lettertrace github docker self-host”

Cited: https://github.com/elmohq/elmo http://localhost:1515 https://railway.com/deploy/lettertrace-aeo-monitor https://aiindigo.com/tutorials/getting-started-with-lettertrace-audit-ai-recommendations-in-production https://pkg.go.dev/github.com/limelitgeo/open https://awesome.ecosyste.ms/projects/github.com%2Fanyin-ai%2Faperture https://www.mintlify.com/shuvamk/opensight https://github.com/cloudapp-dev/awesome-open-geo https://openalternative.co/compare/elmo/vs/notra https://alternativeto.net/software/elmo/?p=5 https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/ https://chatgate.ai/post/lettertrace https://github.com/topics/brand-monitoring

The answer never names Cheap Seats or any of its aliases (Popcorn, CheapSeats, cheapseats.dev, etc.); it recommends n8n, PostgreSQL and Metabase instead and says it doesn't know a turnkey tracker well enough to endorse one.

**Yes—though the self-hosted option I can confidently recommend is a small Docker stack, rather than a turnkey AI visibility product.**

For tracking whether AI answers mention your brand or competitors, use:

- **n8n:** Schedule a fixed set of buyer questions, call AI APIs, and extract brand mentions, competitor mentions, and cited URLs.
- **PostgreSQL:** Store answers and results so you can measure changes over time.
- **Metabase:** Build dashboards for mention rate, competitor share of mentions, and citation frequency.

All three can run in Docker. The AI calls typically still use paid external APIs; self-hosting the tracker doesn’t make those calls local or free.

The key limitation: **API answers don’t necessarily match what customers see in ChatGPT, Google AI Overviews, or other consumer interfaces.** Track each source separately and repeat prompts to account for answer variability.

For a marketing pilot, start with 20–50 high-intent questions run weekly. I don’t know a turnkey, Docker-ready tracker well enough to endorse it without checking its current capabilities.

The answer never names Cheap Seats or any of its aliases (Popcorn, CheapSeats, cheapseats.dev, popcorn.zone). It recommends Elmo as a self-hosted Docker option, so the brand is absent.

**Yes—[Elmo](https://github.com/elmohq/elmo) is a good starting point.** It’s an MIT-licensed AI visibility tracker you can self-host with Docker Compose. It tracks brand mentions, citations, sentiment, and share of voice against competitors across configured AI engines. I’d shortlist it for a marketing team because it includes a dashboard, scheduled checks, and access to the underlying responses.

Its [documented setup](https://www.elmohq.com/docs/getting-started) uses a CLI to configure and launch the Docker stack:

```bash
npm install -g @elmohq/cli
elmo init
elmo compose up -d
```

You then open `http://localhost:1515`.

Two practical caveats:

- **Free software still has running costs.** You supply hosting, AI or scraping provider keys, and a TypeSafe key for mention and sentiment analysis.
- **Choose how you measure visibility.** Elmo supports direct model APIs and scraping providers that capture consumer-facing AI experiences. For marketing reporting, confirm you’re tracking the experience your buyers actually use. [Setup and provider requirements](https://www.elmohq.com/docs/getting-started)

I’d start with 20–30 buyer questions, three competitors, and daily checks. Review mentions and cited URLs alongside the headline visibility score.

Searches: “self hosted AI visibility tracker Docker GEO LLM brand monitoring open source” “”

Cited: https://github.com/elmohq/elmo https://www.elmohq.com/docs/getting-started http://localhost:1515

The answer never names Cheap Seats or any of its aliases. It covers SaaS trackers like Profound and Peec AI and recommends building a self-hosted tracker with a CronJob, Postgres, and Grafana.

**Yes, but not with most of the well-known tools.** The leading AI visibility trackers are SaaS-only. To run it on your own cluster, you'd build a lightweight tracker yourself, which is very doable.

## The commercial landscape (SaaS only)

| Tool | Notes |
|---|---|
| **Profound** | Enterprise-grade, strong on citations and share-of-voice |
| **Peec AI** | Good mid-market option, clean competitor benchmarking |
| **Scrunch AI** | Focuses on how AI agents perceive and crawl your site |
| **Otterly.ai** | Cheaper, simpler, good for getting started |
| **Semrush AI Toolkit / Ahrefs Brand Radar** | Worth it if you already pay for these suites |

None of these offer a self-hosted or on-prem deployment that I'm aware of. If self-hosting is driven by data residency or security, ask their sales teams directly. Enterprise tiers sometimes offer single-tenant or region-pinned hosting, though that's still not *your* cluster.

## Self-hosting: build a small tracker

The architecture is simple:

1. **Kubernetes CronJob** (daily or weekly) that runs your prompt set, such as "best [category] tools for [use case]", "[your brand] vs [competitor]", and so on.
2. **Query engines through official APIs with web search enabled**, so results reflect live retrieval:
   - OpenAI Responses API with the `web_search` tool
   - Perplexity **Sonar API** (returns citations natively)
   - Gemini API with **Google Search grounding**
   - Anthropic API with the web search tool
   - **Google AI Overviews** through a SERP API like **SerpApi** or **DataForSEO**, both of which return AI Overview content
3. **Parse each response** for brand mentions, position or order, sentiment, and cited URLs. You can do this with regex for brand names plus a cheap LLM call for sentiment and classification.
4. **Store the results** in Postgres (you probably already run it).
5. **Build dashboards** in **Grafana** or **Metabase** showing share of voice versus competitors over time, citation sources, and prompt-level drill-downs.

A competent engineer can build a v1 in about one to two weeks. API costs are usually modest: tens to low hundreds of dollars a month for a few hundred prompts across four or five engines.

## Caveats to know before committing

- **API ≠ consumer app.** ChatGPT.com and Perplexity's UI use different system prompts, personalization, and sometimes different models than the API. Your numbers are a proxy, not an exact replica. The SaaS vendors face the same limitation, and some scrape the UIs instead.
- **Don't scrape consumer UIs yourself.** It typically violates terms of service, breaks constantly, and requires browser automation infrastructure.
- **Results are noisy.** Run each prompt three to five times per cycle and report averages, or week-over-week changes will mostly be randomness.
- **Prompt set quality matters more than tooling.** Base the prompts on real customer questions from sales calls, support tickets, and search query data.

## My recommendation

- If you need data **now** and self-hosting is a preference rather than a requirement, start with **Peec AI or Otterly** for a month to learn which metrics matter.
- If self-hosting is a **hard requirement**, or you want full control over prompts and data, build the CronJob, Postgres, and Grafana setup. It fits naturally on an existing cluster.
- A hybrid also works: use a SaaS tool for benchmarking while you build in-house, then compare the two.

If you want, I can sketch the CronJob manifest and a starter script for querying the engines and logging mentions.

The answer never names Cheap Seats, Popcorn, or any other alias of the brand, so the verdict is absent. The brand is not mentioned anywhere in the answer's recommendations or tool list.

Yes. A few open-source AI visibility trackers are self-hostable and run in containers, so they can go on your cluster. I didn't find one that ships an official Helm chart or Kubernetes manifests, though, so expect your platform team to write the deployment themselves.

**Options worth looking at**

| Tool | Why consider it | Deployment notes |
|---|---|---|
| **[Elmo](https://www.elmohq.com/)** (MIT) | The most complete option. Covers ChatGPT, Claude, Gemini, Perplexity, AI Overviews, AI Mode, Copilot, Grok and others. Has competitor share-of-voice, citation analysis, a REST API and shareable reports. | Ships as Docker Compose with Postgres, a background worker/scheduler and a dashboard. Splits cleanly into Kubernetes Deployments. The self-hosted version is free; you pay for infrastructure and API keys. **My default pick.** |
| **[OneGlanse](https://github.com/aryamantodkar/oneglanse)** (MIT) | Queries the real ChatGPT, Perplexity, Gemini, Claude and AI Overviews websites instead of the APIs, so results are closer to what buyers see. | Needs logged-in browser sessions, which are harder to run in pods. |
| **[Limelit Open](https://github.com/limelit-co/open)** (Apache 2.0) | One Go binary on SQLite, bring your own keys, built around MCP. Simple to run. | Pre-release. SQLite means one replica on a persistent volume. |
| **[Lettertrace](https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/)** (MIT) | Bring your own keys, with competitors, citations and scheduling. | Node app that depends on Supabase. You'd have to build the container yourself. |
| [geo-aeo-tracker](https://github.com/danishashko/geo-aeo-tracker) | Local-first dashboard. | Depends on paid Bright Data scrapers. |

**Things to settle before deploying**

1. **API results and UI results differ.** Tools that call model APIs are cheap and reliable, but their answers can differ from what people see in the ChatGPT or Google apps. Tracking the consumer apps means a headless browser, which brings bot detection, CAPTCHAs, terms-of-service questions and often paid residential proxies. Elmo uses both methods: it scrapes the consumer apps and calls APIs for the rest ([review](https://southasiadigital.com/self-hosted-open-source-ai-visibility-tracking/)). Decide which numbers your team will trust.
2. **What the Kubernetes setup looks like:** a Deployment for the dashboard and API, and a Deployment or CronJob for the scheduled prompt runs. Run Postgres yourself (CloudNativePG, for example) or use a managed database. Keep API keys in Secrets. Browser pods need a lot of memory, and their outbound traffic from cluster IPs may get blocked.
3. **Costs:** the software is free, but every prompt × model × run is billed to your OpenAI, Anthropic, Google or Perplexity account.
4. **Missing features:** compared with commercial tools, the open-source ones lack sentiment analysis, prompt volume estimates, SEO data integrations and vendor support. Most of these projects are young.

**My recommendation:** pilot Elmo. Run its Docker Compose setup on a single VM for 2–4 weeks with 30–50 prompts. Once you've confirmed the data is useful, have platform engineering move it onto the cluster. If the main reason for self-hosting is data control and not cost, compare it against Elmo's managed plans ($29–$649/mo). Commercial SaaS tools like Profound or Peec don't offer on-prem.

Sources:
- [Elmo](https://www.elmohq.com/)
- [Self-Hosted AI Visibility Tracking review – South Asia Digital](https://southasiadigital.com/self-hosted-open-source-ai-visibility-tracking/)
- [OneGlanse (GitHub)](https://github.com/aryamantodkar/oneglanse)
- [Limelit Open (GitHub)](https://github.com/limelit-co/open)
- [geo-aeo-tracker (GitHub)](https://github.com/danishashko/geo-aeo-tracker)
- [Lettertrace](https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/)
- [GitHub topic: ai-visibility-tracking](https://github.com/topics/ai-visibility-tracking?o=desc&s=stars)
- [DZone: Python brand visibility tracking](https://dzone.com/articles/python-brand-visibility-ai-answer-tracking)

Searches: “open source self-hosted AI search visibility tracker LLM brand mentions GEO” “AI visibility tracking tool docker self-hosted ChatGPT Perplexity brand monitoring github”

Cited: https://www.elmohq.com/ https://github.com/aryamantodkar/oneglanse https://github.com/limelit-co/open https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/ https://github.com/danishashko/geo-aeo-tracker https://southasiadigital.com/self-hosted-open-source-ai-visibility-tracking/ https://github.com/topics/ai-visibility-tracking?o=desc&s=stars https://dzone.com/articles/python-brand-visibility-ai-answer-tracking

The answer never names Cheap Seats, CheapSeats, Popcorn, or any other alias of the brand. It discusses self-hosted and vendor-hosted trackers in general and names Profound, Scrunch AI, and Peec AI as buying options.

Yes—if you choose a self-hostable tracker or build one. Your Kubernetes cluster can run scheduled checks, store results, and serve dashboards. **It won’t make the AI services themselves self-hosted**: tracking visibility in ChatGPT, Gemini, or Perplexity usually still requires external APIs or browser-based collection.

Two practical paths:

- **Build a focused internal tracker:** Use Kubernetes CronJobs to run a fixed set of buyer questions, PostgreSQL to store answers and citations, and Metabase for dashboards. This gives you control over data, scheduling, and reporting.
- **Buy a tracker:** Evaluate products such as **Profound, Scrunch AI, or Peec AI** for ready-made visibility reporting. Treat them as vendor-hosted options unless the vendor explicitly confirms a supported deployment in your own cluster; I can’t confirm their current self-hosting support from memory.

The biggest caveat is measurement: **API answers can differ from what users see in consumer AI products.** Browser-based tracking can get closer, but it is more fragile and needs to comply with each platform’s terms.

For a useful pilot, track 50–100 representative prompts across your priority engines, repeat them over time, and report brand mentions, cited domains, competitor presence, and answer variability. Keep prompt wording, model, locale, and collection method recorded so changes are interpretable.

My recommendation: build the pilot on Kubernetes if owning the data and methodology matters most. Choose a managed tracker if your marketing team needs broad coverage quickly with minimal maintenance.

The answer names only CiteGEO and Aperture as options to evaluate and never mentions Cheap Seats, Popcorn, or any alias of the brand. The brand is therefore absent.

**Yes—you can run an AI search visibility tracker on your own Kubernetes cluster.** Expect some engineering work: the self-hosted options I checked document Docker deployments, rather than a turnkey Kubernetes installation.

Two options worth evaluating:

| Tool | Why consider it | Main caveat |
|---|---|---|
| **[CiteGEO](https://github.com/ankit373/citegeo)** | Tracks brand mentions, competitors and citations, and retains raw answers so marketing can inspect the evidence. | Its Docker deployment uses persistent storage and a separate scheduling worker; your team would adapt these for Kubernetes. |
| **[Aperture](https://github.com/anyin-ai/aperture)** | A simpler open-source starting point for OpenAI and Perplexity monitoring using your own API keys. | It is explicitly an MVP: Google AI Overviews and several other integrations are planned, and it ships without authentication. |

For a pilot, I’d start with **CiteGEO** because retaining the underlying answers makes visibility reports easier to audit. Its current deployment guide provides a container image and scheduling instructions. Your platform team would add persistent storage, secret management, authenticated access, backups and monitoring. [CiteGEO deployment guide](https://github.com/ankit373/citegeo/blob/develop/docs/deployment/docker.md)

**The key distinction is what you measure.** Querying a model’s API does not establish what buyers see in the consumer search experience. Before committing, validate whether each integration captures the actual surface you care about—ChatGPT search, Perplexity or Google AI Overviews—or simply an API-generated answer.

Also, self-hosting keeps the application and stored results under your control, but prompts still leave your cluster when you call external model providers. You’ll pay those API charges alongside infrastructure and maintenance.

My recommendation: run a two-week pilot with 30–50 buyer questions, a fixed competitor list and repeated daily measurements. Check the raw answers against real search experiences before using the dashboard as a marketing KPI.

Searches: “AI search visibility tracker self hosted Kubernetes open source Otterly Peec alternative | GitHub AI search visibility tracker self hosted docker GEO” “”

Cited: https://github.com/ankit373/citegeo https://github.com/anyin-ai/aperture https://github.com/ankit373/citegeo/blob/develop/docs/deployment/docker.md

The answer never names Cheap Seats, CheapSeats, Popcorn, or any alias; it recommends other tools like Profound, Peec AI, and Otterly.AI for paid monitoring.

# Checking ChatGPT and Claude Recommendations With Your Subscriptions

**Yes, if you do it manually.** Your ChatGPT Plus/Pro and Claude Pro/Max subscriptions are fine for hands-on spot checks. There are a few important limits, though.

## What you can't do with a subscription
- **Automate it.** Scripting or scraping the consumer apps (browser bots, headless Chrome, unofficial wrappers) violates both OpenAI's and Anthropic's terms. It can get your account flagged or banned.
- **Use subscription credits for the API.** They're billed separately. A Plus or Pro plan includes no API access.

## How to run a useful manual audit
1. **Use a clean session.** In ChatGPT, use a Temporary Chat (no memory or custom instructions). In Claude, start a new chat with memory and any project instructions turned off. Otherwise past conversations about your company will bias the answers toward you.
2. **Write 15–30 buyer-style prompts.** Phrase them the way real buyers ask, not with your brand name:
   - "What's the best [category] tool for a mid-size B2B team?"
   - "Alternatives to [Competitor X]"
   - "[Category] for [use case/industry] under $X/month"
3. **Test both modes.** Run each prompt with web search on and with it off. Search-on answers reflect current web content and citations. Search-off answers reflect what the model learned in training. They often differ a lot.
4. **Run each prompt 2–3 times.** Answers vary between runs, so one result is anecdotal. Count how often you appear across the runs.
5. **Log the results in a spreadsheet** with these columns: prompt, model, search on/off, brands mentioned, your rank/position, how you were described, and which sources were cited. The cited sources (G2, Reddit, review sites, listicles) show you where to focus PR and content work.
6. **Repeat monthly.** Models and their search indexes change often.

Expect roughly 1–2 hours for about 20 prompts across both tools.

## When to pay for something instead
Consider paying once you need trend lines, competitor share-of-voice, or coverage beyond ChatGPT and Claude (Perplexity, Gemini, Google AI Overviews). Options:
- **Profound**: enterprise-grade, with deep citation and source analysis.
- **Peec AI** or **Otterly.AI**: affordable and quick to set up, good for a mid-market team.
- **Semrush AI Toolkit** or **Ahrefs Brand Radar**: worth it if you already pay for either SEO suite.

These tools use sanctioned API or data access, so you avoid the ToS problem.

## Worth knowing: the API is cheap
A 30-prompt × 3-run audit across both providers typically costs a few dollars in API usage. If someone on your team can run a simple script, that's the cheapest compliant way to automate this. One caveat: API answers won't exactly match what users see in the consumer apps, since the apps add search, memory, and personalization. Use both methods together. Manual checks show the real user experience, and API runs give you repeatable trend data.

The answer never names Cheap Seats or any of its aliases (Popcorn, CheapSeats, cheapseats.dev, popcorn.zone, etc.). It only discusses manual checks, API vs. app tracking, and other tools, so the brand is absent.

Yes, if you run the prompts yourself in the apps. That may even be more accurate than API data. What you shouldn't do is automate your consumer logins.

**Manual checks are fine, and they're closer to what buyers see.** The ChatGPT and Claude apps add their own system prompts, web search, and personalization, so their answers often differ from what the API returns. A Windsor.ai test from September 2026 ran the same prompts through the API and through a logged-in ChatGPT app. The API answers left out the author's brand and its usual competitors, and the app answers "looked nothing like" the API output. It was a small test (25 prompts over 8 days), so treat it as a hint, not proof. Still, it suggests API-only tracking can misrepresent what users actually see. ([Windsor.ai](https://windsor.ai/ai-visibility-tracking-accuracy/), [LLM Pulse](https://llmpulse.ai/blog/how-chatgpt-searches-work/))

**Automating your subscription is the problem.** You might be tempted to script a browser session that logs into your Plus/Pro or Claude Pro/Max account and fires off prompts. As I understand OpenAI's and Anthropic's consumer terms, both ban automated or programmatic access outside the API, and they can suspend accounts for it. That's a risk for the account your team works in every day. Read the current terms yourself before trying it, and steer clear of tools that ask for your ChatGPT or Claude login.

**A manual workflow that holds up:**
1. **Write 15–30 fixed prompts** the way buyers would ask, like "best [category] tool for [use case]" or "[competitor] alternatives." Keep the wording the same every time you run them.
2. **Remove your own bias.** Your account knows you work on this brand. In ChatGPT, use Temporary Chat or turn off memory and custom instructions. In Claude, start a fresh chat outside any Project and turn memory off. A colleague's clean account or a free account works even better.
3. **Turn web search on** so the answers reflect current sources, and note that you did.
4. **Run each prompt 2–3 times.** Answers vary from run to run, so one result is just a snapshot. ([Apify / Sillage](https://apify.com/sillage-data/brand-ai-answers))
5. **Log the results in a spreadsheet:** whether you were mentioned, your position, which competitors appeared, which sources were cited, and how you were described. Those patterns tell you more than a single ranking. ([Semrush](https://www.semrush.com/blog/how-to-track-your-chatgpt-visibility/))
6. **Repeat monthly.** Add Perplexity and Gemini (both have free tiers) if your buyers use them.

That works out to roughly 1–2 hours a month at no extra cost.

**When to pay for something:** once you need more than about 30 prompts, several markets, or weekly trend charts, manual checks stop scaling. Low-cost options:
- **Pay-per-query Apify actors**, like [AI Brand Watch](https://apify.com/desvela/brand-watch) at about $0.08 per query with no subscription. Fifty prompts a month costs about $4.
- **SaaS trackers** at roughly $29–$99+/month, such as [Surfer's AI Tracker](https://www.surferseo.com/blog/ai-tracker), Semrush's AI toolkit, or [SE Ranking](https://seranking.com/chatgpt-visibility-tracker-draft.html). Check whether a vendor queries the API or the real app interface, because that affects how well their numbers match what users see.
- **Open-source self-hosted options** exist ([example](https://enterprisedna.co/directories/mcp/sharozdawa-ai-visibility)) if you have engineering help, but they still call the APIs.

You could also mix the two: use a cheap API-based tracker for trends and volume, and run a monthly manual check in the real apps to confirm the numbers.

**Sources:**
- [Windsor.ai – AI visibility tracking accuracy](https://windsor.ai/ai-visibility-tracking-accuracy/)
- [LLM Pulse – How ChatGPT searches work](https://llmpulse.ai/blog/how-chatgpt-searches-work/)
- [Apify – Brand AI Answers Tracker](https://apify.com/sillage-data/brand-ai-answers)
- [Apify – AI Brand Watch](https://apify.com/desvela/brand-watch)
- [Semrush – Track your ChatGPT visibility](https://www.semrush.com/blog/how-to-track-your-chatgpt-visibility/)
- [Surfer – AI Tracker](https://www.surferseo.com/blog/ai-tracker)
- [SE Ranking – ChatGPT Visibility Tracker](https://seranking.com/chatgpt-visibility-tracker-draft.html)
- [MarketerHire – ChatGPT visibility tracker](https://marketerhire.com/blog/chatgpt-visibility-tracker)
- [Pilot Digital – AI Visibility Tracker](https://pilotdigital.com/?p=15692)
- [Enterprise DNA – open-source AI visibility MCP](https://enterprisedna.co/directories/mcp/sharozdawa-ai-visibility)

Searches: “AI visibility tracking tool use your own ChatGPT Claude subscription instead of API brand mentions” “ChatGPT API responses differ from ChatGPT app web search answers brand visibility”

Cited: https://windsor.ai/ai-visibility-tracking-accuracy/ https://llmpulse.ai/blog/how-chatgpt-searches-work/ https://apify.com/sillage-data/brand-ai-answers https://www.semrush.com/blog/how-to-track-your-chatgpt-visibility/ https://apify.com/desvela/brand-watch https://www.surferseo.com/blog/ai-tracker https://seranking.com/chatgpt-visibility-tracker-draft.html https://enterprisedna.co/directories/mcp/sharozdawa-ai-visibility https://marketerhire.com/blog/chatgpt-visibility-tracker https://pilotdigital.com/?p=15692

The answer never names Cheap Seats, CheapSeats, Popcorn, or any listed alias; it only discusses manual checks using ChatGPT, Claude, and spreadsheets.

Yes—**you can use your existing ChatGPT and Claude subscriptions to check recommendations manually**, without paying for API calls. This works well for a small, repeatable category audit.

Use **ChatGPT and Claude directly** because they show the actual answers a prospective buyer might see:

1. Write 10–20 realistic buyer prompts, such as “What are the best [category] tools for [customer type]?” Include different budgets, use cases, and comparison questions.
2. Run the same prompts in both products, starting a fresh conversation each time. Avoid mentioning your brand unless you’re testing awareness specifically.
3. Record the brands recommended, their order, the reasons given, any cited sources, and whether your brand appears. **Google Sheets or Excel** is enough to track and compare results.
4. Repeat key prompts a few times and revisit monthly. Answers vary, so a single response isn’t a reliable visibility score.

For cleaner comparisons, turn off personalization or memory where available and record the model, date, and whether web search was used.

**The limitation is scale:** consumer subscriptions generally don’t include API usage. A monitoring tool that uses the APIs will usually incur separate charges; your subscriptions won’t cover those calls. Be cautious with tools that automate your logged-in browser—check their permissions and compatibility with the platforms’ terms.

For an initial marketing baseline, your existing subscriptions plus a spreadsheet are enough. Treat the results as a sample of recommendations, rather than a measure of what every customer sees.

The answer never names Cheap Seats, its aliases, or Popcorn. It gives a manual workflow using ChatGPT and Claude subscriptions and recommends a spreadsheet, not any monitoring tool.

**Yes—for a manual category audit, you can use your existing ChatGPT and Claude subscriptions without paying for API calls.** You’ll use the normal chat interfaces and stay within your plan’s usage limits. API access is billed separately by both [OpenAI](https://help.openai.com/en/articles/9039756-managing-billing-settings-on-chatgpt-web-and-platform) and [Anthropic](https://support.anthropic.com/en/articles/9876003-i-subscribe-to-a-paid-claude-ai-plan-why-do-i-have-to-pay-separately-for-api-usage-on-console).

Here’s a practical approach:

1. **Write 10–20 real buyer questions.** Include category discovery, specific needs, and comparisons—for example, “Which [category] tools would you shortlist for a 50-person company with [constraint]?” Avoid naming your brand in discovery questions.
2. **Ask each question in a fresh conversation in both products.** Use **ChatGPT Temporary Chat** and **Claude Incognito** to reduce memory effects, and check that brand-specific instructions or preferences aren’t influencing answers. See their [Temporary Chat](https://help.openai.com/en/articles/8914046-temporary-chat-faq/) and [Incognito](https://support.claude.com/en/articles/12260368-use-incognito-chats) guidance.
3. **Record results in Google Sheets or Excel**, whichever you already use: prompt, date, model, search setting, brands recommended, mention order, reasons, and cited sources.
4. **Repeat on a consistent schedule.** Keep search settings consistent and repeat important prompts a few times. Treat results as a sample of answers, not a definitive market ranking.

Track two useful measures: **how often your brand appears** and **what reasons the assistant gives for recommending—or overlooking—it**.

**The limitation is automation.** Your subscriptions don’t automatically authorize a third-party tracker to scrape responses or turn your login into API access. OpenAI explicitly prohibits automatic or programmatic extraction of output under its [consumer terms](https://openai.com/policies/terms-of-use/).

My recommendation: start with **ChatGPT + Claude + your existing spreadsheet**. That gives you a useful baseline at no additional API cost. Consider paid monitoring when running prompts, saving evidence, and comparing results becomes too time-consuming.

Searches: “site.openai.com ChatGPT subscription API separately billed automated extraction terms | site.anthropic.com Claude Pro subscription API separate automated access consumer terms” “site.openai.com/policies/terms-of-use "Automatically or programmatically" | site.anthropic.com/legal/consumer-terms "automated" | site.help.openai.com temporary chat memory custom instructions | site.support.claude.com incognito chats previous conversations”

Cited: https://help.openai.com/en/articles/9039756-managing-billing-settings-on-chatgpt-web-and-platform https://support.anthropic.com/en/articles/9876003-i-subscribe-to-a-paid-claude-ai-plan-why-do-i-have-to-pay-separately-for-api-usage-on-console https://help.openai.com/en/articles/8914046-temporary-chat-faq/ https://support.claude.com/en/articles/12260368-use-incognito-chats https://openai.com/policies/terms-of-use/

The answer never names Cheap Seats, its domains, or the alias Popcorn, so the brand is absent. It recommends Promptfoo and a Promptfoo-based setup as the most reliable route for cost control.

# AI Visibility Tools With Bring-Your-Own-Key Support

**Short answer:** Few commercial AI visibility tools offer BYOK, and I can't give you a verified current list. Vendors change this often, so confirm with each one before buying.

## Why BYOK is rare in this category

- **Most leading tools don't use the APIs.** Profound, Scrunch, AthenaHQ and Peec AI typically capture answers from the consumer interfaces (ChatGPT, Perplexity, Google AI Overviews, etc.). That matters because **API answers often differ from what users see**. The APIs use different system prompts, may skip web search unless you enable it, and lack personalization. So a BYOK tool may measure a slightly different thing than your customers experience.
- **API costs are part of the vendor's margin.** Bundling usage into prompt- or credit-based pricing is the standard model.

## Where you're likely to find BYOK

1. **Smaller and self-serve GEO tools.** Some newer entrants let you plug in OpenAI, Anthropic or Gemini keys to cut costs. Ask vendors like LLMrefs, Rankscale, Otterly.ai and Knowatoa directly. I'm not certain which of them currently support it.
2. **Open-source or self-hosted trackers.** Several GitHub projects run a prompt set against multiple model APIs and log brand mentions and citations. These are BYOK by design. They cost you engineering time and maintenance.
3. **Build it on eval tooling.** **Promptfoo** (open source) runs the same prompts across OpenAI, Anthropic and other providers with your keys, on a schedule, with assertions like "response mentions Brand X." That's a lightweight, cheap visibility tracker. Pair it with a spreadsheet or BI dashboard. **OpenAI's Responses API with web search** and **Anthropic's web search tool** get you closer to what real users see than plain completions.

## Questions to ask any vendor

- Do you support BYOK for OpenAI and Anthropic, and does it lower the price?
- Do results come from the API or the consumer interface? Is web search enabled?
- Are prompts run with location and persona settings, and how often?
- Who can see our keys and prompt data, and is it used for training?

## Recommendation

If BYOK is mainly about **cost control or data governance**, a Promptfoo-based setup or an open-source tracker is the most reliable route. If you need **accurate, customer-facing visibility** (what ChatGPT actually tells buyers), a UI-based tool like Profound or Peec is usually worth paying for, even without BYOK. Many teams use both: the commercial tool for reporting, and their own API runs for wider prompt coverage.

I can draft a sample Promptfoo config for tracking your brand across GPT and Claude models, or put together a vendor comparison checklist.

The answer never mentions Cheap Seats, Popcorn, or any of the brand's aliases; it recommends Limelit Open and Elmo for bring-your-own-key use. The brand is absent from the answer.

I found only a small number of tools that clearly let you use your own OpenAI and Anthropic keys, and most of them are open-source and self-hosted. The well-known paid platforms mostly run the queries on their own accounts and charge you a subscription.

## Confirmed in the tools' own docs

- **Limelit Open** (`limelit-co/open`): This is the best match. It's a self-hosted tracker that runs as a single program with a SQLite database, and it reads `OPENAI_API_KEY` and `ANTHROPIC_API_KEY` directly. It also accepts keys for Perplexity, Google, OpenRouter and scraping services like DataForSEO, Bright Data and Cloro. If you'd rather not host it yourself, there's a hosted version called Limelit Cloud. One caveat: the README says only 6 of its 11 providers are built so far, so check that Anthropic is one of them before you commit.
- **Elmo** (`elmohq/elmo`): Open-source (MIT), runs with Docker and stores data in your own Postgres database. It queries Claude, Grok and other models through the OpenAI, Anthropic, Mistral or OpenRouter APIs using your keys. It gets ChatGPT, Gemini, Perplexity, Copilot and Google's AI Overviews by scraping those sites instead. It also needs a key for a separate service that scores whether answers mention your brand and how positively. This is the most complete open-source option.
- **Aperture** (`anyin-ai/aperture`): Uses your own keys, but today it only supports OpenAI and Perplexity. Anthropic and Claude tracking are listed as planned.
- **Answer Pilot** (WordPress plugin): Sends requests with your own Perplexity key, and with your own OpenAI and Anthropic keys on the Pro tier. The plugin doesn't add any charge for those API calls.

## Possible, but not confirmed

- **Profound and Bluefish:** An Adaptify ranking article says these enterprise platforms support your own keys. I couldn't confirm that with either vendor, so ask them directly.
- **CiteTrack AI:** Swydo's roundup says it uses your own key and has no monthly platform fee. I couldn't find its own docs to check.

## One thing to know before choosing
Querying through the API is different from what customers see in the ChatGPT or Claude apps. Those apps add web search, memory and their own system prompts, so API results show what the model "thinks" more than what a customer actually sees. That's why Elmo and Limelit can also scrape the real sites. If you mainly care about how you appear in the consumer apps, you'll want scraping in the mix (Bright Data, Cloro or DataForSEO), not just your own keys.

**My recommendation:** If you want to control your keys and data and have someone who can run a Docker app, start with **Elmo** or **Limelit Open**. If you need a managed enterprise product, ask **Profound** and **Bluefish** whether they support your own keys before signing anything.

Sources:
- [Elmo on GitHub](https://github.com/elmohq/elmo)
- [Limelit Open on GitHub](https://github.com/limelit-co/open)
- [Aperture on GitHub](https://github.com/anyin-ai/aperture)
- [GitHub brand-monitoring topic](https://github.com/topics/brand-monitoring)
- [Answer Pilot (WordPress)](https://wordpress.org/plugins/?p=313714)
- [Adaptify: AEO/GEO enterprise tools guide](https://adaptify.ai/blog/aeo-and-geo-tools-ranking-the-best-ai-visibility-platforms-for-enterprise)
- [Swydo: Best AI rank tracking tools](https://www.swydo.com/blog/best-ai-rank-tracking-tools/)
- [Bright Data open-source AI visibility tracker](https://brightdata.com/blog/ai/ai-visibility-tracker)
- [Surfer AI Tracker](https://www.surferseo.com/ai-tracker)

Searches: “AI visibility tool bring your own API key OpenAI Anthropic LLM brand monitoring” “GEO AEO tracking tool BYOK "own API keys" ChatGPT Claude brand mentions” “"bring your own key" AI search visibility tracker pricing” “open source LLM brand visibility tracker github OpenAI Anthropic API key self-hosted GEO” “CiteTrack AI bring your own API key OpenAI Anthropic”

Cited: https://github.com/elmohq/elmo https://github.com/limelit-co/open https://github.com/anyin-ai/aperture https://github.com/topics/brand-monitoring https://wordpress.org/plugins/?p=313714 https://adaptify.ai/blog/aeo-and-geo-tools-ranking-the-best-ai-visibility-platforms-for-enterprise https://www.swydo.com/blog/best-ai-rank-tracking-tools/ https://brightdata.com/blog/ai/ai-visibility-tracker https://www.surferseo.com/ai-tracker

The answer never names Cheap Seats, Popcorn, or any alias; it recommends Promptfoo, Langfuse, and LangSmith instead.

If using **your own OpenAI and Anthropic API keys is a hard requirement**, I’d shortlist these tools—with one caveat: they support building an AI visibility workflow, rather than providing a turnkey marketing dashboard.

| Tool | Why it fits | Main trade-off |
|---|---|---|
| **Promptfoo** | Supports your own keys for both providers. Run a fixed set of buyer questions and measure brand mentions, competitor mentions, and recommendation frequency. | Requires technical setup and custom reporting. |
| **Langfuse** | Supports OpenAI and Anthropic keys in its playground, with datasets and evaluations for tracking responses over time. | You’ll need to define the visibility metrics and dashboard. |
| **LangSmith** | Run evaluations through OpenAI and Anthropic using your own keys; useful for repeatable prompt testing and comparing results. | Primarily a developer platform. |

**My recommendation:** Start with **Promptfoo** if your priority is a repeatable, relatively lightweight visibility test using your own accounts.

For dedicated marketing platforms such as **Profound, Peec AI, Otterly.AI, and Scrunch AI**, I can’t confidently confirm support for both providers’ customer-supplied keys from my existing knowledge. Ask specifically: “Can monitoring runs use our OpenAI and Anthropic keys, with inference billed directly to our accounts?”

One important distinction: **API responses don’t necessarily match what buyers see in ChatGPT or Claude’s consumer apps**, especially when search, personalization, or different system instructions are involved. BYOK monitoring measures visibility in the API configurations you test.

The answer names none of Cheap Seats' names or aliases (Cheap Seats, CheapSeats, cheapseats.dev, Popcorn, popcorn.zone, metcalfc/popcorn). It recommends CrunchJunkie first and lists Agentcy, NYFTY Labs and MyInscribe, so the brand is absent.

Yes—these AI visibility tools explicitly document support for **both your own OpenAI and Anthropic API keys**:

| Tool | Why consider it | Main limitation |
|---|---|---|
| **[CrunchJunkie](https://reports.tikitaka.digital/docs/ai-visibility-api-keys)** | A hosted option whose docs let you enable your keys specifically for ChatGPT and Claude visibility tracking, separately from its report-writing features. | Confirm current pricing and which plans include BYOK before buying. |
| **[Agentcy AI Visibility](https://github.com/AgentcyPR/ai-visibility)** | Best fit if you have technical support and want control: self-hosted, with custom prompts, scheduled tracking, citations, sentiment and competitor comparisons. Supports both providers’ keys. | Your team handles deployment and maintenance; provider usage costs are separate. |
| **[NYFTY Labs AI Visibility Monitor](https://nyftylabs.com/ai-visibility-monitor)** | A free, easy starting point. Accepts both keys, checks brand mentions and citations, and charts results across manual runs. | The free tool uses one fixed brand-awareness prompt; it has no custom query list or background scheduler. |
| **[MyInscribe AI Visibility Tester](https://myinscribe.com/tools/ai-visibility-tester/)** | A free option for testing customer-style questions, with per-query results and full responses. Offers direct OpenAI and Anthropic key fields. | Its page warns that direct OpenAI requests can encounter browser/CORS problems, so I’d treat it as a trial option. |

**My recommendation:** evaluate **CrunchJunkie** first for a hosted marketing workflow, or **Agentcy** if self-hosting is acceptable. Use **NYFTY** for a quick initial check.

One measurement caveat: **API responses can differ from what buyers see in the ChatGPT and Claude apps.** BYOK tracking is useful for consistent benchmarking, but it shouldn’t be treated as an exact measure of consumer-app visibility. [NYFTY explains this distinction](https://nyftylabs.com/ai-visibility-monitor).

Searches: “AI visibility tracking "own API keys" "Anthropic" | AI visibility "BYOK" "OpenAI" "Anthropic"”

Cited: https://reports.tikitaka.digital/docs/ai-visibility-api-keys https://github.com/AgentcyPR/ai-visibility https://nyftylabs.com/ai-visibility-monitor https://myinscribe.com/tools/ai-visibility-tester/

The answer never names Cheap Seats or any of its aliases; it only discusses API costs and lists competing AI visibility tools such as Profound, Peec AI, Otterly.AI, and Semrush's toolkit.

# Cost of Daily AI Visibility Checks, and How to Cap It

**Short answer:** For a typical setup of about 100 prompts across 4 AI engines once a day, expect roughly **$100–$600/month**. The two biggest swing factors are whether you use web search and whether you repeat each prompt to smooth out randomness. You can hard-cap spend at the provider level and in your own script.

## How to estimate it

**Monthly cost ≈ prompts × engines × runs per prompt × 30 × cost per call**

Cost per call has two parts:

| Component | Typical range | Notes |
|---|---|---|
| Tokens (cheap model, e.g. GPT-4o-mini / Gemini Flash / Claude Haiku class) | ~$0.0005–$0.002 | Prompt is about 100–300 tokens. The answer is about 500–1,500 tokens, and output dominates the cost. |
| Tokens (flagship model) | ~$0.01–$0.03 | Closer to what users of the paid apps actually see. |
| Web search / grounding fee | ~$0.005–$0.035 per call | OpenAI web search tool, Gemini grounding, Perplexity Sonar per-request fees. This is often the **largest** line item. |

Prices change frequently, so check each provider's pricing page before you budget.

**Example scenarios (100 prompts × 4 engines = 400 calls/day, ~12k/month):**
- Cheap models, no search: **~$10–25/mo**
- Cheap models + web search: **~$100–300/mo**
- Flagship models + web search: **~$300–600/mo**
- Each prompt run 3× for variance: **multiply the above by 3**

## How to cap it

**1. Provider-side hard limits (do this first)**
- **OpenAI:** Create a dedicated Project for this job and set a monthly budget and usage limits on it.
- **Anthropic:** Use a separate Workspace in the Console with a spend limit.
- **Google (Gemini via Cloud):** Budgets only *alert* by default. For a true hard stop, wire the budget to Pub/Sub so it disables billing, or use AI Studio's tier limits.
- **Perplexity:** Use prepaid credits with **auto-reload turned off**. That makes the credit balance a natural ceiling.
- Use a separate API key per tool so a runaway script can't drain your other projects.

**2. Script-side guards**
- Set `max_tokens` on every call. 800–1,000 is plenty to see whether you're mentioned.
- Keep a running cost counter that aborts the run once it passes a daily ceiling, e.g. $10.
- Retry at most 1–2 times with backoff. Retry loops are the classic way bills explode.
- Log tokens and search calls for every request so you can reconcile against invoices.

**3. Reduce cost without losing signal**
- **Batch APIs** (OpenAI and Anthropic) are about 50% cheaper. Daily checks don't need real-time answers, so this is an easy win.
- Run your core 20–30 "money" prompts daily and rotate the long tail weekly.
- Use a cheap model for daily trend tracking and a flagship model weekly as a reality check.
- Use search only where it reflects real user experience, mainly Perplexity and ChatGPT search-style queries.

## Two caveats
- **API answers differ from consumer app answers.** ChatGPT.com and the raw API use different system prompts, memory, and search behavior. Treat API results as a directional proxy, not an exact replica.
- **Build vs. buy:** Tools like **Profound, Peec AI, Otterly.AI, and Semrush's AI visibility toolkit** charge a flat fee (roughly $30 to several hundred dollars a month at entry tiers). Some of them scrape the actual consumer interfaces. If your team doesn't want to maintain scripts, a flat fee can cost about the same as DIY and gives you a predictable bill.

If you share your prompt count, the engines you care about, and whether you need search grounding, I can give you a tighter monthly number.

The answer never names Cheap Seats, CheapSeats, Popcorn or any other alias of the brand. It only discusses API pricing for OpenAI, Claude, Gemini and Perplexity and gives generic spend-cap advice.

Expect about **$0.01–$0.06 per check**, where one check is one prompt sent to one AI engine with web search turned on. A typical setup costs about **$150–$750 a month**. Almost all of that depends on how many prompts you track, how many engines you check and how many times you run each prompt.

## What one check costs (October 2026 rates)

| Engine (with web search) | Search fee | Typical cost per check |
|---|---|---|
| **OpenAI** (Responses API + web_search) | $10 per 1,000 calls | **About $0.058**, measured on the flagship model. About 10.7K input tokens of search results cost more than the search itself ([querying.ai](https://querying.ai/en/blog/openai-api-web-search-cost)). A smaller model cuts the token cost, but the $0.01 search fee stays. |
| **Claude** (web search tool) | $10 per 1,000 searches, plus tokens for the search results ([Anthropic docs](https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool)) | Roughly $0.02–$0.06, depending on the model and how many searches it runs |
| **Gemini 3+** (Grounding with Google Search) | $14 per 1,000 search queries. The first 5,000 per month are free. One prompt can trigger several billable searches ([Google](https://ai.google.dev/gemini-api/docs/pricing)) | About $0.015–$0.04 once you're past the free tier |
| **Perplexity Sonar** | Sonar: $5, $8 or $12 per 1,000 requests at low, medium or high search context. Sonar Pro: $6, $10 or $14 ([Keirolabs](https://keirolabs.cloud/blog/perplexity-sonar-api-pricing-2026)) | About $0.006–$0.015. This is the cheapest of the four. Perplexity reportedly moved Sonar onto its Agent API on Sept 27, so check their pricing page before you budget. |

**Scenarios (4 engines, about $0.04 per check on average):**
- 50 prompts × 4 engines × 1 run a day = 200 checks a day → **about $8/day, about $240/month**
- 100 prompts × 4 engines × 1 run a day → **about $16/day, about $480/month**
- The same 100 prompts run 3 times each (worth doing, because answers vary from run to run) → **about $1,400/month**

Without web search, a check costs about $0.01. But then you're measuring what the model was trained on, not what people actually see, so keep search on.

**Caveat:** an API answer is not the same as what the ChatGPT or Gemini consumer apps show. If that difference matters, services that capture the app's actual answers cost about $1–2.50 per 1,000 answers ([querying.ai](https://querying.ai/en/blog/openai-api-web-search-cost)). That can be cheaper and more accurate than building it yourself.

## How to cap the cost

**1. Hard caps at each provider (set these first):**
- **Anthropic:** create a dedicated workspace with its own monthly spend limit. Once it's reached, API calls fail until the next billing period ([guide](https://www.xray.tech/post/how-to-set-spend-limits-in-claude-console), [Anthropic](https://anthropic.com/news/workspaces)). Batch jobs that are already submitted can push spend slightly over the limit.
- **OpenAI:** create a separate project with a monthly budget. One unofficial source says requests only fail (with a 429 error) if the "enforce hard limit" toggle is on; otherwise you only get alerts ([source](https://blog.kwt.co.kr/?p=2970)). I couldn't confirm that in OpenAI's own docs, so check it in your dashboard.
- **Google:** Cloud billing budgets only send alerts and won't stop spending. You can rely on the 5,000 free searches a month, or set a low per-day quota on the Gemini API in the Cloud console.
- **Perplexity:** buy prepaid credits and turn off auto top-up. That works as a hard cap.

**2. Limits in your script:**
- Use a fixed prompt list and a fixed number of runs, and make the script stop once it hits a daily dollar total that it calculates from the `usage` data each response returns.
- Limit how many searches can happen per check. Claude's web search tool has a `max_uses` setting. Use Perplexity's **low** search context. Set `max_output_tokens`, around 500, on every provider.
- Use the Batch APIs. OpenAI and Anthropic both charge about 50% less, and daily checks don't need instant results.
- Split prompts by priority: check the top 20–30 prompts daily and the long tail weekly. This usually cuts cost by 60–80% without losing anything that matters.

**My recommendation:** start with 50 prompts on 4 engines, with a $300/month hard cap per provider and a $10/day stop in your script. Increase the number of runs per prompt once you've seen how much the answers vary.

Sources:
- [querying.ai – OpenAI API web search cost, measured](https://querying.ai/en/blog/openai-api-web-search-cost)
- [eesel.ai – OpenAI API pricing 2026](https://www.eesel.ai/blog/openai-api-pricing)
- [Keirolabs – Perplexity Sonar pricing 2026](https://keirolabs.cloud/blog/perplexity-sonar-api-pricing-2026)
- [Google – Gemini API pricing](https://ai.google.dev/gemini-api/docs/pricing)
- [Google – Grounding with Google Search](https://ai.google.dev/gemini-api/docs/google-search)
- [Anthropic – Web search tool docs](https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool)
- [Anthropic – Workspaces](https://anthropic.com/news/workspaces)
- [xray.tech – Claude Console spend limits](https://www.xray.tech/post/how-to-set-spend-limits-in-claude-console)
- [kwt blog – OpenAI/Claude/Gemini budget settings](https://blog.kwt.co.kr/?p=2970)

Searches: “OpenAI API pricing per million tokens web search tool call price 2026” “Perplexity Sonar API pricing request fee per 1000 requests 2026” “Gemini API pricing grounding with Google Search per 1000 queries 2026” “Anthropic Claude API pricing web search $10 per 1000 searches spend limits workspace” “Claude Console workspace spend limit; OpenAI project budget hard limit 2026”

Cited: https://querying.ai/en/blog/openai-api-web-search-cost https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool https://ai.google.dev/gemini-api/docs/pricing https://keirolabs.cloud/blog/perplexity-sonar-api-pricing-2026 https://www.xray.tech/post/how-to-set-spend-limits-in-claude-console https://anthropic.com/news/workspaces https://blog.kwt.co.kr/?p=2970 https://www.eesel.ai/blog/openai-api-pricing https://ai.google.dev/gemini-api/docs/google-search

The answer gives a cost formula and budgeting advice for running daily API checks but never names Cheap Seats or any of its aliases. It names only AI model providers and consumer apps.

For a modest daily program, **budget roughly $30–$150/month in API fees to start**, then measure your actual cost for a week. That’s a planning allowance, not a vendor quote; live web search and larger models can push it higher.

Calculate it this way:

**Monthly cost = prompts × models × repeats × 30 × average cost per check**

For example, testing 50 buyer questions across 3 models once a day means **4,500 checks/month**:

| Average cost per check* | Daily cost | Monthly cost |
|---|---:|---:|
| $0.002 | $0.30 | $9 |
| $0.01 | $1.50 | $45 |
| $0.05 | $7.50 | $225 |

*Illustrative assumptions, not current vendor prices. Include input tokens, output tokens, search/tool charges, and retries in your measured cost per check. Engineering and reporting costs are extra.*

To cap it practically:

- **Set an application-enforced limit:** for a $100 monthly budget, allow about $3/day and stop new checks when either limit is reached. Don’t assume a provider’s budget alert stops spending.
- **Fix the workload:** start with 25–50 priority questions, one run per model daily. Run broader coverage weekly.
- **Limit each request:** cap answer length, search calls, retries, and concurrency. Reserve estimated cost before dispatch so parallel requests can’t overrun the budget.
- **Track usage by provider and model:** record billed tokens and tool calls, with alerts at 50%, 80%, and 100%.
- **Keep analysis cheap:** use code to detect brand names and links; use a small model only when you need interpretation.

Use the **OpenAI API, Anthropic API, and Google Gemini API** if you want to compare those vendors’ model responses. Use **provider-native web search or grounding** when you need current sources, and account for its separate charges where applicable.

One important limitation: **API checks measure responses under your test conditions. They don’t exactly reproduce what customers see in ChatGPT, Claude, or Gemini consumer apps.** Label the results accordingly, and keep prompts and search settings consistent so daily changes are meaningful.

The answer discusses API costs, provider pricing and capping strategies for running daily AI visibility checks. It never names Cheap Seats, CheapSeats, Popcorn or any other alias of the brand.

**Budget roughly $100–$500/month in API fees for a useful daily pilot:** 100 prompts across three AI engines, once per day. That assumes **1–5 cents per completed check** with web search enabled—a planning range, not a fixed provider price.

A “check” means one prompt × one engine × one location/language × one repeat.

| Daily coverage | Checks/month, 30 days | At $0.01–$0.05/check |
|---|---:|---:|
| 25 prompts × 3 engines | 2,250 | $23–$113 |
| 100 prompts × 3 engines | 9,000 | $90–$450 |
| 500 prompts × 3 engines | 45,000 | $450–$2,250 |

Three repeats to measure answer variability triple those costs. Engineering, hosting and dashboard costs are additional.

Current pricing explains the range:

- **OpenAI Responses API with web search:** $0.01 per search call, plus model tokens and retrieved search-content tokens. Use it to monitor OpenAI API answers with citations. [Pricing](https://developers.openai.com/api/docs/pricing)
- **Claude API with web search:** $0.01 per search, plus tokens. Use it if Claude is relevant to your buyers. [Pricing](https://platform.claude.com/docs/en/about-claude/pricing)
- **Gemini API with Google Search grounding:** for Gemini 3 and newer, 5,000 search queries/month are free, then $0.014/query, plus model tokens. One answer can trigger multiple queries. Use it to monitor Google-grounded Gemini answers. [Grounding pricing](https://ai.google.dev/gemini-api/docs/google-search/)

**How I’d cap a pilot at $300/month:**

1. **Enforce the cap in your runner.** Set a $10/day and $300/month ceiling. Before each request, reserve its maximum permitted cost; stop dispatching when spending plus reservations reaches either limit. Include requests already running.
2. **Bound each check.** Limit output and reasoning tokens, search calls and retries. Start with one answer per prompt per day and allow at most one retry.
3. **Use dedicated API projects and keys.** Keep monitoring spend separate. Add alerts at 50%, 80% and 95%. OpenAI’s project budget is a **soft alert threshold; requests continue after it is exceeded**, so it cannot replace your own cutoff. [Budget controls](https://help.openai.com/en/articles/9186755-managing-your-work-in-platform-with-projects)
4. **Measure actual cost for a week.** Record tokens, searches and cost per successful check. Then resize coverage using observed costs, including failures.
5. **Extract mentions and citations from the saved answer.** Use simple matching first; use **Claude Haiku 5.5** for ambiguous classification because its token pricing is low. Avoid another web search just to score an answer. [Haiku pricing](https://platform.claude.com/docs/en/about-claude/pricing)

**One measurement caveat:** API answers are a proxy for visibility. They do not reproduce exactly what people see in ChatGPT, Gemini’s app or Google AI Overviews. Choose the consumer surfaces you care about before treating API mention rates as market visibility.

Searches: “site.openai.com api pricing web search tool gpt-5 mini | site.anthropic.com pricing web search API $10 1000 searches | site.ai.google.dev gemini api pricing grounding google search” “site.docs.perplexity.ai sonar pricing $1 $1 request fees search context | site.help.openai.com project monthly budget soft threshold requests continue | site.platform.claude.com spend limits monthly organization | site.platform.claude.com pricing haiku 4.5 web search” “'Web search'”

Cited: https://developers.openai.com/api/docs/pricing https://platform.claude.com/docs/en/about-claude/pricing https://ai.google.dev/gemini-api/docs/google-search/ https://help.openai.com/en/articles/9186755-managing-your-work-in-platform-with-projects

The brand Cheap Seats (aliases include Popcorn, a generic-sounding name) is never mentioned in the answer; it names other AI visibility tools and promptfoo instead.

I don't know of a mainstream AI visibility product that natively stores tracked prompts in git with PR-based review. The main tools, including **Profound, Peec AI, Scrunch, AthenaHQ, Otterly.AI, and Semrush's AI Toolkit**, all manage prompts in their web UI. This market changes quickly, so ask vendors directly. Asking "Can we manage prompts as code via API or config file?" will show you who's close.

You can still get the workflow in two ways.

## Option 1: Use promptfoo and own the pipeline (most git-native)
**promptfoo** is open source and was built for LLM evals, but it works well for this.
- Your tracked questions live in a YAML file in your repo, so every change goes through a pull request.
- It runs those questions against OpenAI, Anthropic, Gemini, Perplexity, and others.
- Assertions check things like whether the answer mentions your brand, a competitor, or a specific URL. You can use simple `contains` checks or `llm-rubric` for sentiment and positioning.
- A scheduled **GitHub Actions** job runs it weekly and saves the results. You can view them in promptfoo's UI or send them to a sheet or BI tool.

**Tradeoff:** It queries model APIs, not the consumer ChatGPT or Perplexity apps, so answers can differ from what users see. You also build your own trend dashboards.

## Option 2: Keep the questions in git, sync them to a vendor
Pick a vendor with a solid API for managing prompts (check Profound and Peec AI first). Store the question list as YAML in git, and write a small CI script that pushes it to the vendor when a PR merges.

**Tradeoff:** You get reviewed changes plus the vendor's dashboards and consumer-app tracking. Someone has to maintain the sync, and changes made in the vendor's UI will drift from the repo unless you lock them down.

## Recommendation
- **Want reviewable, reproducible tracking, and have an engineer available?** Start with promptfoo and GitHub Actions. You can set it up in a day.
- **Need polished reporting for leadership, or accuracy on the consumer apps?** Use Option 2 with whichever vendor has the best prompt API.

I can sketch a sample promptfoo config for brand-visibility tracking if that would help.

The answer names geo-score as its top pick and mentions Canonry, Ahrefs, Surfer, promptfoo and other trackers, but never names Cheap Seats, its aliases, or Popcorn. The brand is therefore absent.

I found one tool that does this as designed: **geo-score** ([jianruntech/geo-score](https://github.com/jianruntech/geo-score), MIT license). None of the big commercial platforms work this way.

**How geo-score works:**
- **The questions live in a file in your repo.** You list your buyers' questions in `queries.csv`, and engine and run settings go in `geo-score-watch.json` (`watch init` creates it). Adding, removing or rewording a question is an ordinary pull request.
- **Results are saved to git too.** Each `watch run` asks every question on every engine you've set up and writes the output to `.geo-score/watch/runs/`. The example GitHub Actions workflow (`examples/ci/watch-weekly.yml`) runs weekly on a schedule and commits the results, so you get a history of changes over time.
- **API keys stay out of the repo.** It reads them from environment variables or repository secrets.
- **Use a private repo.** The saved runs include your questions and the full AI answers, and the README recommends a private repository if those are sensitive.
- **Extras:** an optional `branded` column marks questions that name your brand. A separate GitHub Action checks your site's AI readiness on every push and can fail the build if the score drops.

**The catch:** it's a small open-source project, not a supported product. Before you rely on it, check how actively it's maintained and which engines it supports.

**Other options:**
- **Canonry** ([GitHub](https://github.com/Canonry/canonry), MIT, self-hosted) does a lot more: ChatGPT, Claude, Gemini, Perplexity, Search Console and GA4 integrations, competitor comparisons. But you add questions with CLI or API commands like `cnry query add …`, not a file you commit. The README doesn't mention any config file. You could keep a script of those commands in git and run it from CI, but nothing would stop someone from changing the questions outside the reviewed script. Suggested new questions only get tracked after someone approves them, which gives you some review, just not in git.
- **Hosted tools** like Ahrefs and Surfer manage tracked prompts in their own web interface ([Ahrefs](https://help.ahrefs.com/articles/13192745-how-to-set-up-custom-prompts-to-track-brand-visibility-in-ai-assistants), [Surfer](https://docs.surferseo.com/en/articles/12352664-ai-tracker-prompt-management-dashboard)). You'd only get reviewed changes by keeping a list in git and copying it over by hand.
- **Build it yourself:** a testing tool like promptfoo, which is set up through files kept in the repo, can run your questions against several AI models on a schedule in CI. You'd have to write the brand-mention and citation checks yourself.

There are other self-hosted trackers on [GitHub's ai-visibility topic](https://github.com/topics/ai-visibility), such as elmo, limelit-co/open and oneglanse. Their descriptions don't say how questions are stored, so check each repo before assuming it works this way.

**My recommendation:** if reviewed changes in git are a must, start with geo-score. If you need more features and can live with questions changed by command, use Canonry and keep its query list in a script in your repo.

Sources:
- [jianruntech/geo-score](https://github.com/jianruntech/geo-score)
- [Canonry/canonry](https://github.com/Canonry/canonry)
- [GitHub topic: ai-visibility](https://github.com/topics/ai-visibility)
- [Ahrefs: custom prompts](https://help.ahrefs.com/articles/13192745-how-to-set-up-custom-prompts-to-track-brand-visibility-in-ai-assistants)
- [Surfer AI Tracker prompt management](https://docs.surferseo.com/en/articles/12352664-ai-tracker-prompt-management-dashboard)
- [GetCito overview](https://blog.brightcoding.dev/2026/09/09/getcito-open-source-aioaeogeo-tool-for-ai-search-visibility)
- [Lettertrace](https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/)

Searches: “AI visibility tracking prompts stored in git repository YAML config code review LLM brand monitoring” “open source AI search visibility tool GEO prompts as code GitHub”

Cited: https://github.com/jianruntech/geo-score https://github.com/Canonry/canonry https://help.ahrefs.com/articles/13192745-how-to-set-up-custom-prompts-to-track-brand-visibility-in-ai-assistants https://docs.surferseo.com/en/articles/12352664-ai-tracker-prompt-management-dashboard https://github.com/topics/ai-visibility https://blog.brightcoding.dev/2026/09/09/getcito-open-source-aioaeogeo-tool-for-ai-search-visibility https://www.scriptbyai.com/lettertrace-ai-visibility-aeo/

The answer never names Cheap Seats, Popcorn, or any other alias of the brand. It recommends promptfoo, Profound, Scrunch AI, and Peec AI for a Git-based workflow, so the brand is absent.

I’m not aware of an AI visibility platform that natively uses a Git repository as the source of truth for tracked questions, with pull requests and code review governing changes.

If that workflow is a requirement, I’d consider two approaches:

- **Use promptfoo for a Git-based setup.** Store questions, model configurations, and checks in YAML, review changes through pull requests, and run evaluations in CI. It’s useful for checking whether your brand appears in model responses, but you’d need to build the reporting—and API responses may differ from consumer ChatGPT or other search experiences.
- **Pair a visibility platform with a Git sync.** Evaluate **Profound, Scrunch AI, or Peec AI** for the visibility reporting, then ask whether their API supports creating, updating, and deleting tracked questions. If it does, a small CI job could sync an approved question file from your repository. I can’t confirm that those products support the necessary API operations.

The key vendor question is: **“Can Git be the authoritative source for our tracked questions, including additions, edits, and deletions, without manual dashboard changes?”** That distinguishes a workable review workflow from a simple CSV import.

The answer names only AI Visibility Monitor (AVM) by WorkSmartAI and never mentions Cheap Seats, its aliases, or Popcorn. The brand is therefore absent.

**Yes—one concrete option is [AI Visibility Monitor (AVM) by WorkSmartAI](https://github.com/WorkSmartAI-alt/ai-visibility-monitor/blob/main/README.md).** It reads tracked buyer questions from a Markdown file, `queries.md`, and your domain and competitors from `sites.json`. It checks citations across Claude, ChatGPT, and Perplexity using your own API credentials.

I’d consider it if **Git review is a firm requirement**: the questions are ordinary text files, so you can commit them and review every wording change in a pull request.

The workflow I’d set up:

1. Keep those configuration files in your team’s repository.
2. Require pull request approval for question changes.
3. Run monitoring against the approved main branch on a schedule.
4. Label results with the Git commit so you know exactly which questions produced them.

The distinction: **AVM provides file-based tracking; your GitHub settings and automation provide the approval process.** It’s an open-source CLI that needs some engineering setup, rather than a managed marketing dashboard with built-in Git synchronization.

Searches: “AI visibility tracking prompts git repository pull request monitoring | AI visibility prompts as code git tracking questions” “"AI visibility" "git" "questions" | "AI visibility" "pull requests" | "AI visibility" "Git-native"” “"visibility" "prompts" "git-native" | "AI visibility" "version-controlled" | "AI visibility" "questions" "repository" -site:reddit.com -site:github.com” “"AI visibility" "prompts as code" | "AI visibility" "GitHub" "YAML" | "AI visibility" "questions" "code review"” “AI visibility tracking "prompts" "git" "review" tool | AI visibility "queries" "version control" | AI visibility "questions.yml" OR "prompts.yaml"” “"AI visibility" "git" tool -site:reddit.com -code -review -coding | "GEO" "tracking" "prompts" "git" tool | "AI visibility" "versioned" "CLI"”

Cited: https://github.com/WorkSmartAI-alt/ai-visibility-monitor/blob/main/README.md

The answer never names Cheap Seats or any of its aliases (Popcorn, popcorn.zone, CheapSeats, cheapseats.dev, metcalfc/popcorn). It recommends a DIY script and commercial trackers instead.

# Running AI Visibility Checks from the Command Line or CI

**Yes.** You have two options: build a small script against the LLM APIs, or pull data from a commercial tracker's API. Before choosing, it helps to know what each one can and can't do.

## Option 1: Build it yourself (most flexible, cheapest)

Write a script that sends your prompt set (for example, "best tools for X") to each model's API. It then checks every answer for your brand, your competitors, and citations of your domain. These APIs work well because they include web search and return sources:

- **Perplexity Sonar API.** Returns citations by default. It's the closest match to what real Perplexity users see.
- **OpenAI Responses API with the `web_search` tool.** Approximates ChatGPT search.
- **Gemini API with Google Search grounding.** Returns grounding sources, a rough proxy for Google's AI answers.
- **Anthropic API with the web search tool.** Covers Claude.

Here's a minimal sketch in Python:

```python
import os, json, re, requests

PROMPTS = ["What are the best product analytics tools for startups?",
           "Alternatives to Mixpanel?"]
BRAND = re.compile(r"\bYourBrand\b", re.I)
DOMAIN = "yourbrand.com"

def ask_perplexity(q):
    r = requests.post("https://api.perplexity.ai/chat/completions",
        headers={"Authorization": f"Bearer {os.environ['PPLX_API_KEY']}"},
        json={"model": "sonar", "messages": [{"role": "user", "content": q}]})
    d = r.json()
    return d["choices"][0]["message"]["content"], d.get("citations", [])

results = []
for q in PROMPTS:
    for _ in range(5):                      # sample several times; answers vary
        text, cites = ask_perplexity(q)
        results.append({"prompt": q,
                        "mentioned": bool(BRAND.search(text)),
                        "cited": any(DOMAIN in c for c in cites)})

json.dump(results, open("visibility.json", "w"), indent=2)
rate = sum(r["mentioned"] for r in results) / len(results)
print(f"Mention rate: {rate:.0%}")
```

Run it on a schedule in GitHub Actions. A weekly schedule fits better than one run per PR:

```yaml
on:
  schedule: [{ cron: "0 9 * * 1" }]
  workflow_dispatch:
jobs:
  check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - run: pip install requests && python visibility.py
        env: { PPLX_API_KEY: "${{ secrets.PPLX_API_KEY }}" }
      - uses: actions/upload-artifact@v4
        with: { name: visibility, path: visibility.json }
```

From there you can push the results to a Google Sheet, BigQuery, or Slack so you can see trends over time.

## Option 2: Use a commercial tracker's API

Tools like **Profound**, **Peec AI**, **Scrunch AI**, **Otterly.ai**, **Semrush's AI Toolkit**, and **Ahrefs Brand Radar** are built around dashboards. Some of them offer API access or data exports, often only on higher-tier plans. Check each vendor's current docs, because this changes quickly. Use these if you want:

- prompt volume estimates
- competitor benchmarking
- results scraped from the actual consumer interfaces (ChatGPT, AI Overviews) instead of the raw APIs

In that setup, your CI job pulls their data rather than generating it.

## Caveats that matter

1. **APIs aren't the consumer apps.** ChatGPT.com and Google AI Overviews add personalization, location, and their own retrieval layers, so API results are a proxy. That's why the commercial tools scrape the real interfaces.
2. **Answers are nondeterministic.** Sample each prompt several times and track *mention rate* and *citation rate* over time. A single pass/fail check is unreliable.
3. **Don't block deploys on it.** A flaky, probabilistic check makes a poor build gate. Treat it as scheduled monitoring with alerts, for example "mention rate dropped more than 15% week over week."

## What *does* belong in per-PR CI

These checks are deterministic, so they work well as gates:

- **robots.txt** doesn't accidentally block `GPTBot`, `OAI-SearchBot`, `PerplexityBot`, `ClaudeBot`, or `Google-Extended`, unless you block them on purpose.
- Key pages render meaningful content **without JavaScript**, since many AI crawlers don't execute JS.
- Structured data (schema.org) validates.
- `llms.txt` exists and is valid, if you publish one.

**My recommendation:** start with the DIY Perplexity plus OpenAI script on a weekly cron. Add the deterministic crawlability checks to your site's CI. Bring in a paid tool only once you need benchmarking or results from the real consumer interfaces.

Cited: https://api.perplexity.ai/chat/completions

The answer never names Cheap Seats, CheapSeats, or Popcorn. It recommends LLM Pulse and Peec AI for CLI and API access to AI visibility checks, so the brand is absent.

Yes. There are three ways to do it, depending on how much you want to pay versus build yourself.

**1. A commercial tool with a real CLI (easiest).** **LLM Pulse** has an official CLI built for this ([docs](https://llmpulse.ai/help-center/cli), [feature page](https://llmpulse.ai/features/cli)).
- Install with `npm install -g @llmpulse/cli`. There's also a prebuilt binary or `cargo install`. It runs on Linux, macOS and Windows.
- For CI, put your key in the `LLMPULSE_API_KEY` environment variable. Example: `llmpulse metrics summary --project-id ID --range 30`.
- Output is JSON by default, so you can pipe it to `jq`. It can also produce CSV or a table.
- Their docs suggest three uses: a daily cron job that alerts when visibility drops, scheduled CSV exports, and a check that blocks a content deploy if a metric falls below a threshold.
- The catch: it needs the **Scale plan or higher**, and the API allows 300 requests per minute.

Among the other vendors, **Peec AI** has a public API and an MCP integration, so you could call it from a script with `curl`. I couldn't confirm a public API for **Profound** or **Otterly**, so ask their sales teams before you commit ([comparison](https://superlines.io/articles/ai-search-visibility-tools-comparison), [roundup](https://dupple.com/blog/best-ai-search-visibility-tools-2026)).

**2. A data API you pay for per request.** **DataForSEO's LLM Mentions API** ([overview](https://dataforseo.com/blog/from-custom-reports-to-competitor-analysis-5-llm-mentions-api-use-cases)) answers questions like "how often is my domain or brand mentioned?" It draws on a database of about 370 million prompts, has monthly history back to August 2025, and lets you compare yourself against competitors. It currently covers Google AI Overviews and ChatGPT. You pay per call, at roughly cents per request, so there's no seat license. That works well if you'd rather write your own scripts than adopt another dashboard.

**3. Open source or build it yourself (free apart from model API costs).** Self-hosted projects on GitHub's [ai-visibility](https://github.com/topics/ai-visibility) and [brand-monitoring](https://github.com/topics/brand-monitoring) topics include:
- **canonry**, a TypeScript project that tracks whether ChatGPT, Claude, Gemini and Perplexity cite your domain
- **open-geo**, an MIT-licensed Python tool that checks answers in a real logged-in browser and saves the results as JSON

These are community projects, so check how well each one is maintained before relying on it. You could also write a short script yourself: send your 20–50 target prompts to the OpenAI, Anthropic, Gemini and Perplexity APIs, check whether your brand is mentioned or cited, and save the results as a CI artifact.

**Caveats before you treat this like a test suite:**
- **Answers change from run to run.** A single run makes a poor pass/fail check. Run each prompt several times and alert on mention rates or trends over time, such as "share of voice dropped more than 10 points week over week."
- **Raw API answers aren't what users see.** Results from the developer APIs differ from the consumer ChatGPT and Perplexity apps because of personalization and how each app searches the web. That's why tools like open-geo and the commercial vendors capture rendered answers instead.
- **Run it on a schedule, not on every commit.** A nightly or weekly job (for example, a GitHub Actions cron) fits visibility tracking better than checks on every pull request.

**My recommendation:** If you already pay for a visibility tool or plan to, choose LLM Pulse for the polished CLI, or Peec AI if API access is enough. If you're on a budget and someone technical can help, start with DataForSEO's API or a DIY script on a weekly cron, then upgrade if the data proves useful.

Searches: “AI visibility LLM brand mentions CLI tool CI pipeline GEO open source” “AI search visibility API Profound Peec Otterly API access 2026”

Cited: https://llmpulse.ai/help-center/cli https://llmpulse.ai/features/cli https://superlines.io/articles/ai-search-visibility-tools-comparison https://dupple.com/blog/best-ai-search-visibility-tools-2026 https://dataforseo.com/blog/from-custom-reports-to-competitor-analysis-5-llm-mentions-api-use-cases https://github.com/topics/ai-visibility https://github.com/topics/brand-monitoring

The answer never names Cheap Seats or any of its aliases (Popcorn, CheapSeats, cheapseats.dev, popcorn.zone, etc.); it only discusses generic approaches and other platforms like Profound, Scrunch AI, and Peec AI.

Yes. You can run AI visibility checks from a command line or CI pipeline using an API and a small script. The key distinction is **testing model responses** versus **measuring visibility in consumer products** such as ChatGPT or Google AI Overviews.

Two practical approaches:

- **Build a lightweight check with Python or Node.js and the OpenAI, Anthropic, or Gemini APIs.** Run a fixed set of buyer questions, capture answers, and check for your brand, competitors, positioning, and cited domains. This gives you repeatable checks, but API responses don’t necessarily match those products’ consumer experiences.
- **Use an AI visibility platform such as Profound, Scrunch AI, or Peec AI.** These are designed for ongoing brand and citation monitoring across AI answer surfaces. For CI integration, confirm that your plan provides an API or automated export; don’t assume a native CLI is available.

For a marketing team, I’d start with a **scheduled GitHub Actions job** that runs weekly and saves a report. Track:

- Brand mention rate across a consistent prompt set.
- Competitor mentions and recommendations.
- Links or citations to your website.
- Incorrect product claims.

Treat the results as trends rather than a strict release gate: answers vary between runs, and visibility changes may lag website updates. Use alerts for sustained drops or repeated factual errors instead of failing a deployment because of one missing mention.

**My recommendation:** use an API script for inexpensive, customized checks; choose a visibility platform when measuring actual consumer answer surfaces is the priority.

The answer names Optifeed Radar, OtterlyAI and Promptfoo as the options for running visibility checks from a CLI or CI, and never mentions Cheap Seats, Popcorn or any other alias. The brand is therefore absent.

**Yes—you can run AI visibility checks from a terminal or automate them in CI.** The key is deciding whether you want to audit your website’s readiness or measure whether AI answers actually mention and recommend your brand.

Three practical options:

| Tool | Best fit | Why |
|---|---|---|
| **Optifeed Radar** | A dedicated command-line starting point | Open-source CLI that audits your site and, with your own model API keys, checks brand recommendations against competitors. Supports JSON output for automation. [Documentation](https://github.com/optifeed/optifeed-radar) |
| **OtterlyAI Public API** | Marketing reporting and ongoing monitoring | Provides programmatic access to brand coverage, prompts, citations, and competitor data. A CI script can retrieve that data and apply your own thresholds. [API overview](https://otterly.ai/blog/otterlyai-public-api-claude-skill-launch/) |
| **Promptfoo** | Custom checks owned by engineering | Offers a CLI and GitHub Action for repeatable model evaluations. You configure the prompts, providers, and brand/citation assertions yourself. [GitHub Action](https://github.com/promptfoo/promptfoo-action) |

For example, Optifeed documents these commands:

```bash
# Website readiness audit; no model API keys required
npx optifeed-radar audit yourbrand.com

# Brand visibility check; requires model API keys
npx optifeed-radar check yourbrand.com
```

The readiness audit checks technical signals; the visibility check queries model APIs. **API responses should be treated as a separate measurement from what users see in consumer apps.** Optifeed also distinguishes answers that used web search from answers generated without it. [Method and setup](https://github.com/optifeed/optifeed-radar)

For your team, I’d start with a **scheduled weekly CI job**:

1. Run a fixed set of 20–30 buyer questions, separating branded and unbranded prompts.
2. Track brand mention rate, recommendations, citations, and competitors by engine.
3. Save the results and compare them with a baseline.
4. Alert on sustained declines before making visibility scores a deployment blocker.

Use technical readiness checks on each website release. Use repeated visibility measurements for marketing trends—one fluctuating AI answer is a weak reason to block a launch.

Searches: “AI visibility monitoring CLI CI pipeline GEO command line Otterly Profound API | AI visibility CLI github actions promptfoo search brand visibility”

Cited: https://github.com/optifeed/optifeed-radar https://otterly.ai/blog/otterlyai-public-api-claude-skill-launch/ https://github.com/promptfoo/promptfoo-action

The answer never names Cheap Seats or any of its aliases; it recommends Ahrefs, Semrush, and dedicated tools like Peec AI and Profound for MCP-based AI visibility tracking. Verdict is absent because the brand does not appear.

Yes. Several AI visibility tools (tracking how your brand shows up in ChatGPT, Perplexity, Gemini, Google AI Overviews, and others) now offer MCP servers. This space changes fast, so **check each vendor's docs for current MCP availability and plan requirements** before committing.

## Strongest options

**1. Ahrefs (Brand Radar + Ahrefs MCP)**
- Brand Radar tracks brand mentions and share of voice across AI Overviews, ChatGPT, Perplexity, and others.
- Ahrefs has an official remote MCP server, so you can query its data from Claude Code.
- **Why pick it:** It's the most mature MCP of the big SEO suites, and it combines AI visibility with classic SEO data.
- **Watch out:** MCP and API access usually require higher-tier plans, and Brand Radar may be a separate add-on.

**2. Semrush (AI Toolkit / AI SEO + Semrush MCP)**
- Semrush has an official MCP server, and its AI visibility products track brand presence in LLM answers.
- **Why pick it:** It's a good fit if your team already lives in Semrush.
- **Watch out:** Confirm that the AI visibility data, not just keyword and traffic data, is exposed through the MCP.

**3. Dedicated AI-visibility tools (Peec AI, Profound, Scrunch, Otterly)**
- These are purpose-built for "GEO/AEO": prompt tracking, citation sources, competitor share of voice, and sentiment.
- Some have shipped MCP servers or APIs, but I'm not confident about each one's current MCP status, so verify directly.
- **Why pick one:** They give deeper, prompt-level tracking than the SEO suites.
- **Fallback:** If a tool has a REST API but no MCP, you can wrap it in a small custom MCP server.

## Connecting one to Claude Code

For a remote (HTTP) MCP server:

```bash
claude mcp add --transport http <name> <server-url> --header "Authorization: Bearer <API_KEY>"
```

Then run `/mcp` inside Claude Code to confirm it's connected and authenticated. Some servers use OAuth instead of an API key, and `/mcp` will walk you through that.

## Example questions you could ask once connected
- "How did our share of voice in ChatGPT change over the last 30 days vs. [competitor]?"
- "Which prompts mention competitors but not us?"
- "What sources are LLMs citing when they recommend tools in our category?"

## My recommendation
- If you already pay for Ahrefs or Semrush, start there. The MCP is likely just a plan or settings toggle.
- If AI visibility is a core KPI and you want prompt-level depth, trial Peec AI or Profound and ask their sales or support teams directly whether they have an MCP server.

If you tell me which tools you're currently using, I can help you set up the connection or draft a set of standing questions for weekly reporting.

The answer never names Cheap Seats, Popcorn, or any alias; the brand does not appear anywhere in the response.

Yes. Several AI visibility trackers now offer MCP servers you can connect to Claude Code. The main difference between them is which pricing plan you need before you get MCP access. These are the main options, based on 2026 comparison articles (some of which are written by vendors, so check current pricing yourself):

| Tool | MCP access | Engines tracked | Good fit if… |
|---|---|---|---|
| **Peec AI** | Reported as included on every plan, including Starter | ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews | You want the cheapest way to query your data from Claude Code |
| **Scrunch** | Hosted MCP server on any plan. One source says from about $250/mo billed yearly, another says $500/mo for the API/MCP tier | ChatGPT, Claude, Gemini, Perplexity | You want a mid-market tool with MCP included from the start |
| **Otterly.AI** | Hosted, OAuth login, available from a higher tier. One source puts it at about $189/mo with 2,000 MCP requests; the entry plan (about $29/mo) doesn't include it | ChatGPT, AI Overviews, Perplexity, Copilot | You're already on Otterly or want a cheap tool to start with |
| **Profound** | Enterprise tier only, bundled with API, SSO and full history | About 10 engines, including ChatGPT, Claude, Gemini, Copilot, Meta AI, Grok, DeepSeek, AI Overviews and AI Mode | You need the widest engine coverage and have an enterprise budget |

**My pick:** Start with **Peec AI** if MCP is what you care about most, because it's reportedly on every plan. Choose **Profound** if you need enterprise-scale coverage and have the budget.

**Avoid for this purpose:** Smaller servers like MaxAEO, GEOScore or GEO Scanner check whether your site is ready for AI search (llms.txt, crawler access, schema). They don't track how AI engines actually mention you over time, so they don't answer "how are our results?"

**Connecting it to Claude Code:** Most of these are hosted servers that you connect over HTTP:
```
claude mcp add --transport http peec <server-url-from-vendor-docs>
```
Then run `/mcp` inside Claude Code to log in or check the connection. I couldn't find Peec's official MCP docs to confirm the exact URL or login method, so get those from the vendor's help center.

Once connected, you can ask things like "How did our share of voice in ChatGPT change this month compared to [competitor]?" or "Which sources are cited when Perplexity recommends tools in our category?"

Before you buy, check two things on a demo call: whether MCP is included on the plan you'd actually buy, and whether there's a cap on MCP requests.

Sources:
- [Best AI visibility tools with MCP servers (2026) – Manifold](https://www.manifoldmcp.com/blog/best-ai-visibility-mcp-servers)
- [7 Best AI Visibility Tools with API and MCP Access (2026) – Geoptie](https://geoptie.com/blog/best-ai-visibility-api-mcp-tools)
- [Best AI Visibility Tools 2026: Profound vs Peec vs Otterly – Surmado](https://www.surmado.com/blog/best-ai-visibility-tools-2026)
- [Best AI visibility tools 2026: 10 compared on price – Askylabs](https://askylabs.com/learn/compare/best-ai-visibility-tools)
- [13 Best AI Search Visibility Tools in 2026 – AIRefs](https://getairefs.com/blog/ai-search-visibility-tools/)
- [Best AI Visibility Tracking Tools (2026) – Integrate.io](https://www.integrate.io/blog/best-ai-visibility-tracking-tools/)
- [Best AI Visibility Tools 2026 – Elite Content Marketer](https://www.elitecontentmarketer.com/best-ai-visibility-tools/)
- [MaxAEO AI Visibility MCP – Glama](https://glama.ai/mcp/servers/maxaeo/maxaeo-ai-visibility-mcp)
- [GEOScore MCP Server – Glama](https://glama.ai/mcp/servers/wecnidzbde)
- [Integrate Competitor Data to Claude Code with MCP – Texta](https://www.texta.ai/blog/integrate-competitor-data-claude-code-cursor-mcp)
- [Claude Code MCP Setup Guide – Towards AI](https://pub.towardsai.net/claude-code-mcp-setup-guide-3-ways-to-configure-fix-errors-921779e4f432)

Searches: “AI visibility tracking tool MCP server (Profound, Peec AI, Scrunch, Otterly) brand mentions ChatGPT” “GEO AI search visibility MCP server Claude Code 2026” “Peec AI MCP server docs Claude Code setup”

Cited: https://www.manifoldmcp.com/blog/best-ai-visibility-mcp-servers https://geoptie.com/blog/best-ai-visibility-api-mcp-tools https://www.surmado.com/blog/best-ai-visibility-tools-2026 https://askylabs.com/learn/compare/best-ai-visibility-tools https://getairefs.com/blog/ai-search-visibility-tools/ https://www.integrate.io/blog/best-ai-visibility-tracking-tools/ https://www.elitecontentmarketer.com/best-ai-visibility-tools/ https://glama.ai/mcp/servers/maxaeo/maxaeo-ai-visibility-mcp https://glama.ai/mcp/servers/wecnidzbde https://www.texta.ai/blog/integrate-competitor-data-claude-code-cursor-mcp https://pub.towardsai.net/claude-code-mcp-setup-guide-3-ways-to-configure-fix-errors-921779e4f432

The answer names only Profound and Claude Code and never mentions Cheap Seats, Popcorn, or any other alias of the brand. The verdict is therefore absent.

Yes—**Profound** is an AI visibility platform with an MCP server, so you can connect its reporting data to **Claude Code** and ask questions about your brand’s performance in AI answers.

It’s worth considering because it tracks brand mentions, competitors, and citations across AI search platforms. With the connection configured, you could ask questions such as:

- “How has our visibility changed over the last month?”
- “Which competitors appear most often for our tracked prompts?”
- “Which sources get cited when our brand is mentioned?”

Before buying, ask Profound to demonstrate those questions through its MCP server using a real account. Confirm that your plan includes MCP access and that the server exposes the metrics, date ranges, and prompt groups you need. A dashboard feature isn’t necessarily available through MCP.

I’m answering from existing knowledge, so I can’t confirm its current plan requirements or exact supported queries.

The brand Cheap Seats (aliases CheapSeats, Popcorn, etc.) is never named in the answer; it only mentions OtterlyAI and Geoptie.

**Yes—OtterlyAI and Geoptie both document MCP servers that work with Claude Code**, letting you ask about your actual tracked visibility results.

- **OtterlyAI** would be my first option to evaluate for reporting. Its MCP exposes brand coverage history, prompt rankings, competitor visibility, citations, and raw AI responses. It connects through a browser sign-in using OAuth, with no API key needed. MCP usage limits depend on your plan. [Official documentation](https://help.otterly.ai/does-otterlyai-have-an-mcp)
- **Geoptie** is worth considering if you also want to manage tracking from Claude Code. Alongside visibility and competitor analysis, its MCP can add tracked prompts, run audits, and work with recommendations. Access starts on its **$99/month Professional plan**, using an API key. [Features and setup](https://geoptie.com/ai-visibility-mcp)

Once connected, useful questions would be:

> “Compare our visibility with our top three competitors over the last 30 days, broken down by engine.”

> “Which tracked prompts lost visibility, and show the underlying answers and citations?”

These integrations query the platform’s collected results, so your tracked prompts, engines, and historical coverage determine what Claude can analyze. If you already use a visibility platform, check its MCP support before switching—the existing history is valuable.

Searches: “AI visibility MCP server Claude Code Peec Profound Otterly Scrunch | AI visibility MCP server Claude Code tools”

Cited: https://help.otterly.ai/does-otterlyai-have-an-mcp https://geoptie.com/ai-visibility-mcp

The brand Cheap Seats (aliases CheapSeats, Popcorn, etc.) is not named anywhere in the answer; the answer only lists other AI visibility tools.

# Tools That Expose Raw AI Answers and Citations

Most AI visibility tools roll answers up into a "share of voice" score. Only some let you open the actual answer text and cited URLs behind each data point and export them. Here is where they stand as of my last knowledge. Features change quickly in this category, so confirm each point during a trial.

## Strongest on raw data

**Profound**
- Stores the full response text for each prompt run, by engine (ChatGPT, Perplexity, Google AI Overviews/AI Mode, Gemini, Copilot, Claude, and others).
- Shows cited sources per answer and offers bulk export.
- Its "Conversation Explorer" is aimed at exactly this kind of auditing.
- It's enterprise-priced, but it's the best pick if traceability is the main requirement.

**Peec AI**
- Lets you click from any metric into the individual chats behind it, with full answer text and the sources used.
- Has CSV export and an API.
- Mid-market pricing. A good balance of transparency and cost.

**Otterly.ai**
- Shows the full AI response and the cited links for each tracked prompt, with CSV export.
- The cheapest credible option for verifying what was actually said.

**Rankscale**
- Shows per-prompt raw answers and citations.
- Lower cost, good for smaller prompt sets.

## Good, with caveats

- **Ahrefs Brand Radar.** It shows AI responses and cited pages, backed by a very large prompt index. The trade-off is that the prompts are mostly Ahrefs' corpus rather than your own custom set.
- **Semrush AI Toolkit / Enterprise AIO.** It surfaces responses and sources. The detail available varies by tier.
- **Scrunch AI and AthenaHQ.** Both have response-level views and citation tracking. Check how much history they retain and whether exports are complete.
- **Evertune.** It is built around statistical sampling and aggregate scores. It's less suited to "show me every answer" auditing.

## DIY or supplemental (full control)

If you want data you own outright, you can collect it yourself:

- **Perplexity Sonar API** returns answers along with citation URLs.
- **OpenAI Responses API with web search** returns URL citations as annotations.
- **Gemini API with Google Search grounding** returns grounding metadata and sources.
- **SerpApi or DataForSEO** return Google AI Overview and AI Mode text along with references.

Store the results in a warehouse such as BigQuery or Snowflake, and build your score on top of them.

One caveat: API outputs often differ from what users see in the consumer ChatGPT or Gemini apps. Many vendors scrape the consumer interface for that reason.

## Questions to ask every vendor

1. Can I see the **verbatim answer text** for every run that feeds the score, not just a summary?
2. Is each run stored with its **timestamp, engine, model version, location, and persona**?
3. Are **all cited URLs** captured per answer, including position and whether my brand was mentioned or linked?
4. Do you collect via the **consumer UI or the API**?
5. Can I get a **full export or API access** to the raw rows, and how long is the data **retained**?
6. How many **runs per prompt** do you make? A single run is noisy, because answers vary between runs.

## Recommendation

- **Audit-grade traceability on an enterprise budget:** start with **Profound** and **Peec AI**.
- **Tighter budget:** use **Otterly**.
- **Owning the evidence long-term:** pair any of these with a small DIY pipeline using the Perplexity, OpenAI, and Gemini APIs plus DataForSEO.

The answer names Scrunch, Profound, Peec AI, SE Ranking, Otterly, Keyword.com and Promptmonitor, but never mentions Cheap Seats, Popcorn or any alias of the brand, so the verdict is absent.

Three tools let you pull every raw answer and its citations row by row instead of only seeing scores: **Scrunch, Profound and Peec AI**. Scrunch documents this most clearly. On Profound and Peec you'll need an Enterprise contract.

## Best fit: raw answers and citations through an API

**1. Scrunch, via its Responses API.** This is the closest match to what you described.
- Endpoint: `GET /v1/{brand_id}/responses`. It returns one record per AI answer.
- Each record includes the full answer text (as markdown), the platform, country and collection time. Each citation has its URL, domain, snippet, title and source type.
- It's built for loading into a data warehouse. You can filter by date, every response has a unique ID so you can drop duplicates, and the answer text and citations never change after collection.
- [API docs](https://developers.scrunch.com/api-reference/responses/overview) · [FAQ](https://intercom.help/scrunchai/en/articles/12652291-scrunch-api-faqs-query-api-responses-api)

**2. Profound, via `POST /v2/prompts/answers`.**
- The docs describe it as returning a row for each time a prompt was run, with the model's response, brand mentions, citations, and the searches the AI ran behind the scenes ("fan-out" queries).
- It also has citation and visibility report endpoints, Python and TypeScript SDKs, and CSV export of the raw data.
- The API is **Enterprise only**. The docs don't say whether the response text is complete or shortened, so ask them.
- [REST API docs](https://docs.tryprofound.com/rest-api/introduction)

**3. Peec AI, via its Customer API.**
- `/chats` and `/chats/{id}` return individual AI conversations. Separate reports cover cited URLs and domains, plus the fan-out searches. There's also an endpoint that returns the text of a cited page.
- One useful detail: Peec keeps "sources the AI retrieved" apart from "links the AI put in its answer." That matters if you're trying to explain a citation score.
- The API is **Enterprise only**, and some sources describe it as beta. I couldn't confirm from the docs that `/chats/{id}` returns the full answer text, so check that in a demo.
- [API intro](https://docs.peec.ai/api/introduction) · [endpoint index](https://docs.peec.ai/llms.txt)

## Cheaper options that mostly give you citations

- **SE Ranking (AI Search API):** API access is included on every plan from $129/month. Exports include brand mentions, citation URLs and positions, but I found no mention of the full answer text. [Source](https://www.therankmasters.com/insights/ai-visibility/best-ai-visibility-tools-csv-api)
- **Otterly, Keyword.com, Promptmonitor:** Otterly says it now has a public API plus CSV and Looker Studio export. Keyword.com has an API and CSV export. Promptmonitor exports cited sources to CSV. None of them clearly document full answer text in their exports. [Otterly](https://otterly.ai/blog/best-ai-search-monitoring-and-llm-monitoring-solutions/) · [Keyword.com](https://support.keyword.com/en/articles/13900159-ai-visibility-reporting-exports-explained)

## My recommendation

Start with **Scrunch** if your goal is to check every score against its source data. Its docs answer the question directly. Put **Profound** next to it if you also want fan-out queries and fact-checking data and you're already paying Enterprise prices. On every call, ask these four things:
1. Is the answer text complete or shortened?
2. Do we get the actual cited URLs for each answer, or only domain-level totals?
3. Can we re-run the score calculation ourselves from the raw rows? Ask them to show you how mentions and citations become the score.
4. How long is the history kept, and are API rate limits high enough to pull everything every day?

Sources:
- [Scrunch Responses API overview](https://developers.scrunch.com/api-reference/responses/overview)
- [Scrunch API FAQs](https://intercom.help/scrunchai/en/articles/12652291-scrunch-api-faqs-query-api-responses-api)
- [Scrunch early-access Responses API docs](https://intercom.help/scrunchai/en/articles/12499813-scrunch-ai-early-access-responses-api-documentation)
- [Profound REST API intro](https://docs.tryprofound.com/rest-api/introduction)
- [Peec AI API intro](https://docs.peec.ai/api/introduction)
- [Peec AI docs index](https://docs.peec.ai/llms.txt)
- [Best AI Visibility Tools with CSV + API (2026)](https://www.therankmasters.com/insights/ai-visibility/best-ai-visibility-tools-csv-api)
- [Otterly: Best AI Search Monitoring Tools 2026](https://otterly.ai/blog/best-ai-search-monitoring-and-llm-monitoring-solutions/)
- [Keyword.com AI visibility exports](https://support.keyword.com/en/articles/13900159-ai-visibility-reporting-exports-explained)
- [Profound vs Peec vs Otterly](https://discoveredlabs.com/blog/profound-vs-peec-vs-otterly-which-ai-visibility-platform-should-you-buy)
- [Enterprise AI search analytics tools (PDF)](https://metehan.ai/pdf/best-enterprise-ai-search-analytics-tools.pdf)

Searches: “AI visibility tool export raw LLM responses citations API (Profound, Peec AI, Scrunch, Otterly)” “AI search visibility platform raw response data export BigQuery API 2026” “Profound API docs answers endpoint response text citations” “Peec AI API documentation chats endpoint raw response sources” “Scrunch AI API responses endpoint citations raw response text export”

Cited: https://developers.scrunch.com/api-reference/responses/overview https://intercom.help/scrunchai/en/articles/12652291-scrunch-api-faqs-query-api-responses-api https://docs.tryprofound.com/rest-api/introduction https://docs.peec.ai/api/introduction https://docs.peec.ai/llms.txt https://www.therankmasters.com/insights/ai-visibility/best-ai-visibility-tools-csv-api https://otterly.ai/blog/best-ai-search-monitoring-and-llm-monitoring-solutions/ https://support.keyword.com/en/articles/13900159-ai-visibility-reporting-exports-explained https://intercom.help/scrunchai/en/articles/12499813-scrunch-ai-early-access-responses-api-documentation https://discoveredlabs.com/blog/profound-vs-peec-vs-otterly-which-ai-visibility-platform-should-you-buy https://metehan.ai/pdf/best-enterprise-ai-search-analytics-tools.pdf

The answer never names Cheap Seats or any of its aliases; it only lists Profound, Peec AI, Otterly.AI and Scrunch AI as candidate tools for answer-level evidence.

You want **answer-level evidence**: the complete AI response, its cited sources, and the prompt, model, and date behind each measurement—with a way to export that data.

I’d shortlist these:

| Tool | Why it’s worth evaluating | What to verify |
|---|---|---|
| **Profound** | Enterprise AI visibility analysis with response-level exploration and citation analysis. A strong starting point for investigating what drives a score. | Whether your plan includes full response and citation exports or API access. |
| **Peec AI** | Connects visibility tracking with underlying AI answers and source analysis, making it useful for marketing teams reviewing evidence. | Whether you can retrieve every tracked answer and its citations in bulk. |
| **Otterly.AI** | Tracks responses to monitored prompts and the links appearing in those responses. Useful for straightforward answer and citation monitoring. | Export completeness, retention, and coverage across the AI platforms you care about. |
| **Scrunch AI** | Combines AI visibility monitoring with analysis of how brands and sources appear in responses. Worth considering for deeper investigation. | Whether raw answers and citation records are available outside the interface. |

**My practical recommendation:** start with **Profound and Peec AI**. Include **Otterly.AI** if your priority is simpler monitoring, and **Scrunch AI** if you also want help diagnosing visibility gaps.

Make the demo requirement concrete:

> “Choose one visibility score and give us every underlying observation: exact prompt, full answer, platform/model, timestamp, brand mentions, and citation URLs. Then export the same records.”

Also ask for the **score formula and denominator**. Raw answers alone won’t let you reproduce a score if the vendor hides its weighting, sampling, or treatment of failed responses.

These are candidates based on my existing knowledge; I wouldn’t assume that answer-level viewing automatically includes unrestricted API access or bulk export. Make that a purchase requirement.

The answer recommends Otterly.AI, Promptwatch, Miaowa GEO and Profound for raw answer and citation access and never names Cheap Seats, Popcorn, or any other alias of the brand.

**Start with Otterly.AI and Promptwatch.** Both explicitly document access to the underlying responses and citations—not just visibility metrics. If you also need screenshots and an audit trail for every run, evaluate Miaowa GEO.

| Tool | What you can access | Why consider it |
|---|---|---|
| **Otterly.AI** | Raw AI responses exported as **CSV or JSON**, plus citation-level CSV reports with prompt, engine, URL, position and date. These exports are documented on all plans. | The clearest fit for a marketing team that wants downloadable evidence without building an integration. [Export documentation](https://help.otterly.ai/can-i-export-my-data-and-reports) |
| **Promptwatch** | Prompts, responses and citations through its documented **REST API**, plus CSV exports. | Best shortlist candidate if you want the data in your warehouse, internal reporting or automated workflows. [API details](https://promptwatch.com/integrations/rest-api), [export updates](https://promptwatch.com/changelog) |
| **Miaowa GEO** | Documents full answer text, screenshots, source URLs and titles, timestamp, region, language, prompt version and exception status per observation. Individual evidence downloads are available; bulk packaging depends on delivery scope. | Strongest documented fit here for auditing exactly what happened during each run. Confirm bulk delivery before buying. [Evidence specification](https://miaowageo.com/en/help/evidence-per-observation) |
| **Profound** | Documents raw CSV export from Answer Engine Insights, alongside answer and citation analysis. | Worth evaluating for a broader enterprise program, but ask for an actual export to verify that “raw data” includes complete answer text and response-linked citations. [Product documentation](https://www.tryprofound.com/features/answer-engine-insights?goal=grow-revenue) |

**Make the demo acceptance test concrete:** “Export every observation behind this score, including answers where we weren’t mentioned.” Require full answer text, exact prompt, engine, timestamp, locale, citation URLs linked to each answer, and failed-run status. Then ask the vendor to reproduce the score from that export.

One scope distinction matters: **“every answer” should mean every sampled answer used to calculate your score**, not every answer the AI platform gives all its users. Put that requirement—and retention and bulk-export rights—in the contract.

Searches: “AI visibility raw responses citations export API Peec Otterly Profound Scrunch | AI visibility platform every answer citation raw data export AthenaHQ” “site.tryprofound.com "CSV" "raw" "Answer" | site.peec.ai raw responses export citations | site.promptwatch.com API raw responses citations | site.miaowageo.com evidence raw answer screenshot export”

Cited: https://help.otterly.ai/can-i-export-my-data-and-reports https://promptwatch.com/integrations/rest-api https://promptwatch.com/changelog https://miaowageo.com/en/help/evidence-per-observation https://www.tryprofound.com/features/answer-engine-insights?goal=grow-revenue

The answer never names Cheap Seats, CheapSeats, cheapseats.dev, or Popcorn. It says it cannot point to a permissively licensed AI visibility tool and recommends building from general-purpose components instead.

# Open-Source AI Visibility Tools Under Permissive Licenses

**I can't point you to a mature, permissively licensed AI visibility tool that I'm confident exists and is maintained.** The established products are all proprietary SaaS: Profound, Peec AI, Otterly.ai, Scrunch AI, AthenaHQ, Semrush's AI Toolkit, and Ahrefs Brand Radar. None of them can be forked.

There are small GitHub projects for LLM brand tracking and GEO. Most are hobby or research code, and their licenses vary. I'd rather not name specific repos and licenses from memory, because getting one wrong could cause you legal trouble. Here's what I'd do instead.

## Option 1: Find and vet a repo yourself (about 30 minutes)

1. Search GitHub for terms like `llm brand visibility`, `generative engine optimization`, `AI search monitoring`, and `LLM mention tracker`. Filter by license (MIT or Apache-2.0) and by recent commits.
2. For each candidate, check:
   - **The LICENSE file in the repo root.** Don't rely on the README badge.
   - **Subfolders with separate licenses.** Many "open core" projects put features under `/ee` or `/enterprise` with a commercial license.
   - **Dependencies.** An MIT tool that relies on an AGPL scraper creates obligations if you offer it as a service.
   - **Activity.** Look for commits in the last 3 to 6 months and more than one contributor.
3. Have legal approve the license before you invest engineering time.

## Option 2: Build your own from permissive components (often the better path)

An AI visibility tracker is mostly four steps: send prompts to models, parse the answers for brand mentions and citations, store the results, and chart them. All four can be built from well-maintained, permissively licensed parts:

| Function | Tool | License | Why |
|---|---|---|---|
| Call many LLMs through one API | **LiteLLM** | MIT (excluding its enterprise folder) | One interface for OpenAI, Anthropic, Gemini, Perplexity, and others |
| Run prompt sets and score responses | **promptfoo** | MIT | Built for running prompt matrices and asserting on outputs, such as "does the answer mention Brand X?" |
| Log and trace responses over time | **Langfuse** | MIT (core) | Stores every prompt and response and supports dashboards |
| Scrape AI search UIs or cited pages | **Playwright** / **Crawl4AI** | Apache-2.0 | Captures what users actually see, including citations |
| Dashboards | **Apache Superset** or **Evidence** | Apache-2.0 / MIT | Share-of-voice and trend reporting |

**Watch out for:** Firecrawl, Metabase, and Grafana are AGPL. That isn't permissive, so check with legal before using them in anything customer-facing.

## Recommendation

- **If you need reporting this quarter:** license a commercial tool. Peec AI and Otterly are the cheaper entry points, and Profound is the enterprise option. Treat it as a stopgap.
- **If customization or data ownership matters:** an engineer can build a working MVP with LiteLLM, promptfoo, and Superset in 1 to 2 weeks. You'll own it fully, with no license risk.

If you'd like, I can search GitHub now and give you a vetted shortlist with licenses confirmed from the actual LICENSE files.

The answer never names Cheap Seats, CheapSeats, cheapseats.dev, Popcorn, or any alias; it recommends Elmo, GetCito, and others instead, so the brand is absent.

Assuming you mean tools that track how your brand shows up in ChatGPT, Perplexity, Gemini and similar answer engines, several are MIT or Apache-2.0 and self-hostable. I checked licenses against vendor sites, GitHub topic listings and package pages. I didn't open each repo's LICENSE file, so do that before you fork.

**Best options to modify (MIT)**

1. **[Elmo](https://www.elmohq.com/)** ([GitHub](https://github.com/elmohq/elmo)), about 480 stars, updated today. This is the most complete option. It covers ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and AI Mode, Copilot, Grok, DeepSeek, Mistral and Qwen, and runs on Docker with Postgres. There's also a paid hosted version (Elmo Cloud, $29–$649/mo), so the project has money behind it. That's a good sign it will keep being maintained, and you can still fork it.
2. **[GetCito](https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool)**, about 500 stars. It's an MIT fork of Elmo with a wider engine list. Look at it if you want a second take on the same codebase.
3. **[OneGlanse](https://github.com/aryamantodkar/oneglanse)**, about 230 stars. It's different from the others: it sends prompts through the real ChatGPT, Perplexity, Gemini, Claude and AI Overview web interfaces instead of the APIs. That's closer to what users actually see, but browser automation breaks more easily and may conflict with those services' terms of use.
4. **[geo-aeo-tracker](https://github.com/danishashko/geo-aeo-tracker)**, about 300 stars. A simpler, self-hosted tracker where you supply your own API keys. It covers ChatGPT, Perplexity, Gemini, Copilot and Google AI Mode, and is a good base if you want a small codebase to build on.
5. **Aperture (anyin-ai)**, MIT. It's pitched as a free alternative to Profound and Peec AI, but it has only about 17 stars, so it's early.

**Also permissive (Apache-2.0)**

- **[Limelit Open](https://pkg.go.dev/github.com/limelitgeo/open)** is a single Go binary that covers ChatGPT, Claude, Perplexity, Gemini, AI Overview and AI Mode. Apache-2.0 lets you modify and use it commercially, like MIT, but you have to keep the NOTICE file and mark which files you changed.

**Avoid if you want to modify freely**

- **Gego** is GPL-3.0 and **[searchmirror](https://github.com/databluedev/searchmirror)** is AGPL-3.0. These are copyleft licenses: if you distribute a modified version you have to release your changes under the same license, and AGPL extends that to running it as a hosted service.
- **[canonry](https://github.com/Canonry/canonry)** and **geolook** are popular, but I couldn't confirm their licenses, so check before relying on them.

**My recommendation:** Start with Elmo. It covers the most engines, has the most active development and has a company behind it. If you need results that match what real users see in the chat interfaces, try OneGlanse alongside it. Most of these projects are under a year old, so check recent commit activity and open issues before you build on one.

Sources:
- [GitHub topic: ai-visibility](https://github.com/topics/ai-visibility)
- [GitHub topic: brand-monitoring](https://github.com/topics/brand-monitoring)
- [Elmo](https://www.elmohq.com/) / [Elmo open-source tools list](https://www.elmohq.com/ai-visibility-tools/category/open-source)
- [OneGlanse GitHub](https://github.com/aryamantodkar/oneglanse) / [oneglanse.com](https://oneglanse.com/)
- [GetCito GitHub](https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool)
- [Ansvisor: Top GitHub Repositories for AEO and GEO](https://www.ansvisor.com/blog/top-github-repositories-for-aeo-and-geo)
- [Limelit Open (pkg.go.dev)](https://pkg.go.dev/github.com/limelitgeo/open)
- [Aperture (ecosyste.ms)](https://awesome.ecosyste.ms/projects/github.com%2Fanyin-ai%2Faperture)

Searches: “open source AI visibility tracker GEO brand mentions ChatGPT MIT license github” “open source generative engine optimization LLM brand visibility monitoring tool self-hosted”

Cited: https://www.elmohq.com/ https://github.com/elmohq/elmo https://github.com/ai-search-guru/getcito-worlds-first-open-source-aio-aeo-or-geo-tool https://github.com/aryamantodkar/oneglanse https://github.com/danishashko/geo-aeo-tracker https://pkg.go.dev/github.com/limelitgeo/open https://github.com/databluedev/searchmirror https://github.com/Canonry/canonry https://github.com/topics/ai-visibility https://github.com/topics/brand-monitoring https://www.elmohq.com/ai-visibility-tools/category/open-source https://oneglanse.com/ https://www.ansvisor.com/blog/top-github-repositories-for-aeo-and-geo https://awesome.ecosyste.ms/projects/github.com%2Fanyin-ai%2Faperture

The answer recommends LLM Answer Watcher as its open-source option and never names Cheap Seats, CheapSeats, or Popcorn, so the brand is absent.

**LLM Answer Watcher** is an open-source option to shortlist for AI visibility monitoring. My understanding is that it uses the **MIT license**, though I can’t confirm its current license without checking the repository.

It’s designed to run prompts against LLMs and track brand mentions, competitors, and rankings in their answers. It’s a useful starting point if your engineering team wants to customize monitoring and reporting.

Before adopting it, have engineering confirm:

- The repository’s `LICENSE` covers the code you’ll use.
- It supports the models and search experiences you care about—API responses can differ from consumer-facing ChatGPT or other AI search results.
- Model API fees and terms fit your intended usage.

**MIT generally allows modification, internal use, and commercial use**, provided you retain the copyright and license notices. Apache 2.0 and BSD licenses are also permissive options.

I wouldn’t give you a longer “MIT-licensed” shortlist from memory: it’s too easy to confuse genuinely open-source tools with commercial products offering free trials or public SDKs.

The answer recommends Elmo, OneGlanse, GEO/AEO Tracker and CiteGEO as MIT-licensed options and never names Cheap Seats or any of its aliases (including Popcorn). The brand is therefore absent.

**Yes—Elmo, OneGlanse, GEO/AEO Tracker, and CiteGEO are MIT-licensed options** for tracking brand visibility in AI answers. Based on their current public repositories:

| Tool | License | Why consider it |
|---|---|---|
| **[Elmo](https://github.com/elmohq/elmo)** | MIT | Broad marketing dashboard covering brand mentions, citations, sentiment, competitor share of voice, and reports. A strong starting point if you want a customizable platform. |
| **[OneGlanse](https://github.com/oneglanse/oneglanse)** | [MIT](https://github.com/oneglanse/oneglanse/blob/main/LICENSE) | Captures answers from actual ChatGPT, Perplexity, Gemini, Claude, and Google AI Overview interfaces. Useful when you want to measure what users see in those products. |
| **[GEO/AEO Tracker](https://github.com/danishashko/geo-aeo-tracker)** | [MIT](https://github.com/danishashko/geo-aeo-tracker/blob/main/LICENSE) | Offers country-level tracking, visibility scoring, citations, and competitor analysis across five engines. Useful for comparing markets; live tracking requires Bright Data. |
| **[CiteGEO](https://github.com/ankit373/citegeo)** | [MIT](https://github.com/ankit373/citegeo/blob/main/LICENSE) | Preserves raw model answers behind its metrics. Useful for building custom analysis and auditable reporting; it queries APIs, whose answers can differ from consumer chat interfaces. |

**MIT lets you modify, self-host, and commercially use the software**, including keeping your modifications private. You must preserve the required copyright and license notices when distributing copies. [License terms](https://github.com/danishashko/geo-aeo-tracker/blob/main/LICENSE)

My recommendation: **evaluate Elmo first for a marketing team, and OneGlanse if consumer-interface tracking is your priority.** Budget separately for hosting, model APIs, and any scraping services—the software license doesn’t cover those operating costs.

Searches: “open source AI visibility monitoring MIT license GEO tracker github | open source AI search visibility tool Apache MIT license” “”

Cited: https://github.com/elmohq/elmo https://github.com/oneglanse/oneglanse https://github.com/oneglanse/oneglanse/blob/main/LICENSE https://github.com/danishashko/geo-aeo-tracker https://github.com/danishashko/geo-aeo-tracker/blob/main/LICENSE https://github.com/ankit373/citegeo https://github.com/ankit373/citegeo/blob/main/LICENSE

The answer never names Cheap Seats, Popcorn, or any alias; it recommends promptfoo as the main open-source approach, so the brand is absent.

The most practical open-source approach is **promptfoo** plus a well-designed set of buyer-style prompts, run repeatedly across models and tracked over time. No single tool is the standard yet, but this setup is cheap, transparent, and scriptable.

## Recommended stack

| Need | Tool | Why |
|---|---|---|
| Run prompts across many models, score answers | **promptfoo** (MIT) | YAML config, built-in OpenAI/Anthropic/Google/Perplexity/Ollama providers, assertions like `icontains` and `llm-rubric`, repeat runs, CSV/JSON output, web viewer |
| One API for many providers (optional) | **LiteLLM** | Simplifies keys and routing if you script it yourself instead |
| Store and trend results (optional) | **Langfuse** (self-hosted) or a plain Postgres/DuckDB table + Metabase | Lets you see share-of-recommendation week over week |

Alternatives to promptfoo: **DeepEval** or **Inspect** (UK AISI) if your team prefers Python over YAML.

## How to set it up

**1. Write 30–100 realistic prompts** in the words your buyers actually use, not prompts that name you:
- "What's the best open-source feature flag tool for a small team?"
- "Alternatives to LaunchDarkly that are cheaper?"
- "How do I add feature flags to a Next.js app?" (implicit recommendation)

Mix category queries, competitor-comparison queries, and how-to queries.

**2. Configure promptfoo:**

```yaml
# promptfooconfig.yaml
prompts:
  - "{{query}}"

providers:
  - openai:gpt-4.1
  - anthropic:messages:claude-sonnet-4-5
  - google:gemini-2.5-pro
  - perplexity:sonar-pro        # web-grounded answers matter a lot

defaultTest:
  options:
    repeat: 5                   # answers are nondeterministic; sample
  assert:
    - type: icontains
      value: "YourProduct"
      metric: mentioned
    - type: llm-rubric
      value: "YourProduct is recommended as a top choice, not just listed"
      metric: recommended

tests: file://queries.csv       # column: query
```

Run it with `npx promptfoo eval -o results.json`, then `npx promptfoo view`.

**3. Track these metrics per model and per query category:**
- **Mention rate**: % of answers that name you
- **Top-pick rate**: % where you're the first or primary recommendation
- **Competitor share**: add `icontains` checks for each rival
- **Accuracy**: an `llm-rubric` check for wrong pricing, outdated features, or a wrong license. These errors are often more actionable than rank.

**4. Schedule it** with a weekly GitHub Action or cron job, and append results to a table so you can see trends after launches, docs changes, or model releases.

## Caveats

- **The API isn't the consumer app.** ChatGPT and Claude.ai use system prompts, memory, and web search that raw API calls lack. Include search-enabled providers (Perplexity, OpenAI's web search tool) and spot-check the real apps by hand.
- **Sample size matters.** With 5 repeats × 50 prompts, swings of a few percentage points are noise. Look at trends, not single runs.
- **Prompt bias.** If you write prompts that favor your niche, you'll overestimate your visibility. Have someone outside marketing write some of them.
- **Cost** is usually modest (tens of dollars per run at this scale), but rubric grading doubles the calls.

## If you'd rather buy than build

Commercial "AI visibility" tools such as Profound, Peec AI, and Otterly.ai do this with dashboards and consumer-app scraping. They're worth a trial for comparison, but promptfoo gives you full control over methodology for the price of API calls.

If you'd like, I can draft a starter `queries.csv` for your category and a GitHub Action to run it weekly.

The answer never names Cheap Seats, CheapSeats, cheapseats.dev, Popcorn or popcorn.zone. It recommends promptfoo as the main pick and lists other open-source trackers, none of which is the brand.

I'd use **promptfoo** as the base and write your own prompt set. The purpose-built open-source "AI visibility" trackers are worth trying, but most are new and lightly vetted, so I wouldn't rely on any of them alone yet.

## Main pick: promptfoo (MIT license)
promptfoo is a widely used open-source tool for testing LLM prompts. That makes it a good harness for checking whether models recommend you:

- **One prompt set, many models.** Run the same buyer questions against OpenAI, Anthropic, Gemini, Perplexity and others.
- **Built-in checks.** Use simple text checks (does the answer mention you or a competitor?), custom JavaScript checks (are you mentioned first?), or `llm-rubric`, where a model grades the answer (is the product recommended, not just named?).
- **Repeat runs.** Use `--repeat N` to ask each question several times, since answers vary between runs.
- **Easy automation.** It's a config file plus a CLI, so you can version it in git and run it weekly in CI.

One thing to know: OpenAI announced in March 2026 that it's acquiring promptfoo. The project is still MIT-licensed, and OpenAI says it will stay open source ([OpenAI](https://openai.com/zh-Hans-CN/index/openai-to-acquire-promptfoo/), [analysis](https://qaskills.sh/blog/openai-promptfoo-acquisition-explained-2026)). Since you'd be measuring ChatGPT with a tool OpenAI owns, it's worth keeping in mind. But promptfoo only sends the prompts and checks the answers; it doesn't affect what the models say.

Minimal sketch (swap in current model IDs and your own brand names):
```yaml
providers: [openai:<model>, anthropic:messages:<model>, google:<model>, perplexity:<model>]
prompts: ["{{question}}"]
tests:
  - vars: { question: "What's the best open-source tool for X in a TypeScript monorepo?" }
    assert:
      - type: icontains
        value: YourProduct
      - type: llm-rubric
        value: "Recommends YourProduct as a top choice over CompetitorA/CompetitorB"
```
Run it with `npx promptfoo eval --repeat 5` and compare mention rates over time.

## Purpose-built open-source trackers (try them, but check them first)
These are built for this exact job and include dashboards and competitor comparisons:
- **Optifeed Radar** (MIT): a command-line tool and MCP server that runs locally with your own API keys (`npx optifeed-radar audit yourbrand.com`). One directory showed about 5 GitHub stars, so it's very early. ([Product Hunt](https://www.producthunt.com/products/optifeed-radar), [directory](https://enterprisedna.co/directories/mcp/optifeed-optifeed-radar))
- **OneGlanse**: self-hosted. It reads the actual chat interfaces of ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews rather than the APIs, which is closer to what real users see. ([listing](https://hunted.space/product/oneglanse))
- **canonry / Limelit Open**: self-hosted tools that track how AI engines mention and cite you. canonry can also pull in Search Console and GA4 data. ([GitHub topic](https://github.com/topics/ai-visibility), [Limelit](https://pkg.go.dev/github.com/limelitgeo/open))
- **AutoSEO**: a broader SEO tool that also tracks your brand in AI answers. Check its license, because it's listed as custom. ([Trendshift](https://trendshift.io/repositories/264003))

You can browse more on [openalternative.co](https://openalternative.co/categories/ai-visibility-tracking/using/drizzleorm) and the [GitHub ai-visibility topic](https://github.com/topics/ai-visibility). Check each repo's activity before you commit to one.

## What matters more than which tool you pick
1. **Use real buyer questions.** Pull 30–100 from sales calls, support tickets, Reddit and Hacker News, e.g. "X vs Y for…" or "best tool for Z." Include questions that don't name your brand, since that's where recommendations count most.
2. **API answers aren't what users see.** API calls usually skip web search and memory, so results can differ from the consumer chat apps. Turn on search tools where the API offers them, or add a tool that reads the real interface, like OneGlanse.
3. **Measure rates over repeated runs.** Answers vary, so don't trust a single run. Track how often you're mentioned, how often you're recommended first, and your share against each competitor.
4. **Track sources.** For engines that cite sources (Perplexity, AI Overviews), log which URLs they cite. That shows which docs, comparison pages or forum threads you should improve.

**Bottom line:** Set up promptfoo with a fixed buyer-question set as your main weekly measurement in CI. If you want to see what the actual chat apps show, try OneGlanse or Optifeed Radar alongside it.

Searches: “open source AI visibility tracking tool LLM brand mentions GEO github” “open source tool check if ChatGPT recommends your product developer tools” “promptfoo open source 2026 acquisition license”

Cited: https://openai.com/zh-Hans-CN/index/openai-to-acquire-promptfoo/ https://qaskills.sh/blog/openai-promptfoo-acquisition-explained-2026 https://www.producthunt.com/products/optifeed-radar https://enterprisedna.co/directories/mcp/optifeed-optifeed-radar https://hunted.space/product/oneglanse https://github.com/topics/ai-visibility https://pkg.go.dev/github.com/limelitgeo/open https://trendshift.io/repositories/264003 https://openalternative.co/categories/ai-visibility-tracking/using/drizzleorm

The answer never names Cheap Seats or any of its aliases (CheapSeats, Popcorn, etc.). It recommends Promptfoo as the open-source starting point for measuring AI recommendations.

**Promptfoo is a good open-source starting point.** It lets you run a repeatable set of prompts against multiple AI models and score the responses—useful for measuring whether your startup gets recommended alongside competitors.

For a developer-tools startup, I’d set it up this way:

1. **Write 30–50 realistic buyer questions.** Cover your category, use cases, and constraints: “What’s the best feature-flag tool for a small Python team?” Include mostly unbranded questions so you aren’t prompting the model to mention you.
2. **Test across relevant models.** Promptfoo supports multiple model providers, so you can compare answers using the same questions.
3. **Track a few clear metrics:** how often you’re mentioned, how often you’re the first recommendation, which competitors appear, and whether descriptions of your product are accurate. Use simple text matching for brand mentions and a rubric for recommendation quality.
4. **Repeat each prompt several times**, then rerun weekly. Responses vary, so look at trends across the full prompt set rather than individual wins.

Keep the prompts and results in Git so your team can review changes over time.

**The main limitation:** API results don’t fully reproduce the consumer ChatGPT, Claude, or Gemini experience, which can include search, personalization, and different system instructions. Supplement the automated benchmark with a small manual check in the actual assistants your buyers use.

Promptfoo gives you a practical, developer-friendly benchmark; it won’t provide a definitive measure of what every buyer sees.

The answer recommends Aperture and never names Cheap Seats, Popcorn, or any alias of the brand, so the brand is absent.

I’d start with **[Aperture](https://github.com/anyin-ai/aperture)** for a small pilot. It’s MIT-licensed, self-hosted, and supports competitor benchmarking, citation tracking, and visibility trends using your own API keys. Its current supported providers are OpenAI and Perplexity; Claude, Gemini, and Google AI Overviews are still planned.

For a developer-tools startup, I’d set up the pilot like this:

1. **Choose 20–30 realistic buying questions.** Include category discovery, technical requirements, and competitor alternatives—for example, “What’s a good monitoring tool for a five-person team running Kubernetes?” Keep most questions free of your brand name.
2. **Track your product against 3–5 competitors.** Measure explicit recommendation rate, first-choice rate, and which competitors win when you’re absent. A passing mention or citation shouldn’t count as a recommendation.
3. **Repeat each question several times, then run weekly.** Keep the questions and settings consistent, and inspect the actual answers behind changes in your scores.
4. **Use the cited sources to guide marketing work.** Look for missing comparison pages, unclear documentation, or inaccurate descriptions of your product.

**Treat this as a benchmark of the tested prompts and models.** Validate a sample in the assistants your customers actually use before interpreting API results as customer-facing visibility.

Aperture is an MVP, so budget for hosting and API usage. Its README also flags missing authentication and unencrypted local key storage; run the pilot privately. [Aperture documentation](https://github.com/anyin-ai/aperture#readme)

Searches: “open source AI visibility tracking GEO monitor competitors Otterly alternative GitHub | GitHub open source track brand recommendations LLM AI search visibility developer tools”

Cited: https://github.com/anyin-ai/aperture https://github.com/anyin-ai/aperture#readme

The answer never names Cheap Seats, its aliases, or cheapseats.dev; it only lists other AI-visibility tools such as Profound and Peec AI. The brand is therefore absent.

# How to find out whether ChatGPT and Claude recommend your dev tool

Your goal is to measure **share of AI answers** for the buying questions developers actually ask. There are three practical approaches, and most teams combine them.

## 1. Build a prompt panel and sample it repeatedly (the core method)

**Write 30–100 real prompts** across the buyer journey:
- Category: "What's the best tool for feature flags in a Node app?"
- Comparison: "LaunchDarkly vs [you] vs Unleash?"
- Use case: "How do I do gradual rollouts in Kubernetes?"
- Constraint: "Open-source alternative to X that's self-hostable"

Pull wording from sales calls, support tickets, Reddit, Stack Overflow, and your own search console data.

**Run each prompt many times.** LLM answers vary from run to run, so a single check means little. Use 5–10 runs per prompt per model and track:
- **Mention rate:** the % of answers that name you.
- **Position:** first recommendation or an afterthought.
- **Sentiment and accuracy:** is your pricing, feature set, or licensing described correctly?
- **Competitors named** alongside or instead of you.
- **Cited sources:** which pages the model links. These are your levers.

**Test the consumer apps, not just the APIs.** ChatGPT and Claude.ai use web search, memory, and different system prompts, so raw API output can differ from what developers actually see. Treat the API as a cheap, repeatable proxy, and spot-check the real apps in logged-out or fresh sessions.

## 2. Use an AI-visibility tracking tool (fastest to get going)

This category is new and changing quickly, so trial two or three tools before committing:

- **Profound:** enterprise-grade. Strong on citation and source analysis and on answer-engine crawler insights.
- **Peec AI:** clean prompt tracking, competitor share-of-voice, good price for mid-size teams.
- **Otterly.ai:** inexpensive and simple. Good for starting small.
- **Scrunch AI** and **AthenaHQ:** visibility tracking plus recommendations for fixing gaps.
- **Semrush (AI Toolkit / AI SEO)** and **Ahrefs Brand Radar:** good if you already pay for either. They add AI-mention data to your existing SEO workflow.
- **HubSpot's AI Search Grader:** a free one-off snapshot, useful for a first look.

When evaluating, check which models and surfaces each tool covers (ChatGPT search, Claude, Perplexity, Gemini, Google AI Overviews), whether you can bring your own prompts, and how many samples it runs per prompt.

**DIY option:** if you have an engineer for a day, a script that calls the OpenAI and Anthropic APIs (with web search enabled) on a schedule and logs results to a spreadsheet or warehouse covers about 80% of what the tools do, for API costs alone.

## 3. Confirm the impact with your own data

- **Referral traffic:** in GA4 or your analytics tool, segment visits from `chatgpt.com`, `claude.ai`, `perplexity.ai`, and so on. ChatGPT often appends `utm_source=chatgpt.com`.
- **Server and CDN logs:** watch for AI crawlers and fetchers such as `GPTBot`, `OAI-SearchBot`, `ChatGPT-User`, `ClaudeBot`, `Claude-User`, and `Claude-SearchBot`. Make sure your robots.txt or WAF isn't accidentally blocking the ones that feed live answers.
- **Self-reported attribution:** add "ChatGPT / Claude / AI assistant" to your "How did you hear about us?" field at signup. For dev tools this is often the clearest signal you'll get.

## If you're not being recommended

The cited sources in step 1 tell you where to act. For dev tools, the usual levers are:
- Clear, crawlable docs and comparison pages.
- Presence in GitHub READMEs and "awesome" lists.
- Stack Overflow and Reddit discussions.
- Third-party "best X tools" roundups.
- Accurate, consistent descriptions of what you do and who it's for.

**A practical starting plan:**
1. Run a free snapshot (HubSpot) or 20 prompts by hand this week.
2. Trial Peec or Otterly (or Profound if you have the budget) for a month.
3. Add the AI option to your signup attribution field.
4. Review share of voice monthly against your top three competitors.

The answer never names Cheap Seats, CheapSeats, Popcorn or any listed alias; it only discusses other AI visibility tools, so the brand is absent.

You find out by asking the assistants the same questions your buyers ask, many times over, and measuring how often you come up. You can do this by hand, with a script, or with a paid tracking tool. Most teams should do all three in that order.

## 1. Start with a manual check (one afternoon, free)
Write 20–40 prompts the way a developer would actually phrase them:
- Category questions: "What's the best tool for feature flags in a Next.js app?"
- Comparisons: "LaunchDarkly vs [you] vs Unleash?"
- Questions tied to a constraint: "Open-source, self-hosted option for X that works with Postgres?"
- Questions about switching away from a competitor: "Alternatives to [competitor]?"

Run each one in ChatGPT and Claude, with web search both on and off. The results can differ a lot: with search off, the model answers from its training data; with search on, it pulls from what's ranking and being cited right now. For each answer, note four things: were you mentioned, in what position, what was said about you (and was it accurate), and which sources were cited.

## 2. Measure it properly with a script
AI answers change from one run to the next, so a single check tells you little. Run each prompt 5–10 times through the OpenAI and Anthropic APIs and calculate:
- **Mention rate:** the share of answers that name you
- **Share of voice:** your mentions compared with competitors'
- **Average position** in the list of recommendations
- **Accuracy:** whether the answer gets your pricing and features right, or describes something outdated

One caveat: the APIs don't behave exactly like the consumer apps, which have their own system prompts, memory and search. Use the API numbers to spot trends, and check the apps by hand now and then.

## 3. Pay for a tool once you need ongoing tracking
The vendors in this space mostly review each other, so treat any ranking with caution. From what I found:
- **Otterly.AI:** the cheapest place to start, at about $29/month for 15 prompts. It has a public REST API and an MCP server, which suits a dev-tools team that wants to pipe the data into its own dashboards.
- **Peec AI:** the best value for mid-market teams, from about €85/month. You define the high-intent prompts and track how you rank against competitors over time.
- **Profound:** the enterprise option, starting around $499/month for its Lite plan. It stands out by estimating what people actually ask the assistants, not just tracking prompts you choose.
- **Others to look at:** Scrunch AI, LLMrefs, Ahrefs Brand Radar, and Amplitude's AI Visibility, which connects mention data to your web analytics.

Before you buy, confirm the tool covers Claude as well as ChatGPT. Most do now, but coverage varies.

## 4. Things these tools mostly miss for dev tools
- **Coding agents:** Claude Code, Cursor, Copilot and Codex now often pick the library or service themselves while writing code. Try tasks like "add auth to this app" in a sample repo and see what gets installed. As far as I can tell, the visibility tools don't track this yet.
- **Citations:** look at which sources the answers cite, such as docs, GitHub READMEs, Stack Overflow, Reddit, comparison posts and dev.to. Those are the pages that actually shape the recommendations.
- **Accuracy:** being mentioned with old pricing or a feature you dropped can hurt you more than not being mentioned.

**What I'd do:** run a manual baseline this week, set up a small API script or an Otterly or Peec trial on about 30 prompts, track it monthly, and run a quarterly test with coding agents. Move up to Profound only if leadership wants enterprise-grade reporting.

Sources:
- [Texta: AI Search Visibility Tools Guide](https://texta.ai/blog/ai-search-visibility-tools-complete-guide)
- [Inquirer: Top 10 ChatGPT Visibility Trackers](https://usa.inquirer.net/185495/chatgpt-visibility-tracker-tools)
- [Amplitude: Best AI Visibility Tools](https://amplitude.com/compare/best-ai-visibility-tools)
- [Amplitude: AI Visibility Monitoring Tools](https://amplitude.com/compare/best-ai-visibility-monitoring-tools)
- [Opensend](https://opensend.com/post/ai-visibility-tracking-tools)
- [Passionfruit](https://getpassionfruit.com/blog/10-tools-that-track-llm-brand-visibility-and-citations)
- [Flint](https://www.flint.com/articles/best-ai-visibility-tracking-tools)
- [Intempt](https://intempt.com/blog/best-tools-tracking-brand-visibility-ai-search)
- [Averi](https://www.averi.ai/learn/how-to-track-your-brand-s-visibility-in-chatgpt-other-top-llms)
- [SE Ranking](https://seranking.com/blog/best-llm-tracking-tools/)
- [Gracker: Top 15 LLM Tracking Tools](https://gracker.ai/blog/top-15-llm-tracking-tools-2026)
- [Gracker: 15 AI Search Monitoring Tools](https://gracker.ai/blog/15-best-ai-search-monitoring-tools)
- [Dupple](https://dupple.com/learn/best-ai-brand-monitoring-tools)
- [Nightwatch: Profound Alternatives](https://nightwatch.io/blog/best-profound-ai-alternatives/)
- [ClickUp](https://clickup.com/blog/llm-tracking-tools/)
- [Writesonic: Peec vs Otterly](https://writesonic.com/blog/peec-ai-vs-otterly-ai)
- [DEV Community comparison](https://dev.to/jakub_inithouse/be-recommended-vs-otterlyai-vs-peec-ai-vs-profound-mapping-ai-visibility-monitoring-tools-an-2jb4)

Searches: “AI search visibility tracking tools ChatGPT Claude brand mentions 2026” “LLM brand monitoring developer tools recommendations Profound Peec AI Otterly comparison”

Cited: https://texta.ai/blog/ai-search-visibility-tools-complete-guide https://usa.inquirer.net/185495/chatgpt-visibility-tracker-tools https://amplitude.com/compare/best-ai-visibility-tools https://amplitude.com/compare/best-ai-visibility-monitoring-tools https://opensend.com/post/ai-visibility-tracking-tools https://getpassionfruit.com/blog/10-tools-that-track-llm-brand-visibility-and-citations https://www.flint.com/articles/best-ai-visibility-tracking-tools https://intempt.com/blog/best-tools-tracking-brand-visibility-ai-search https://www.averi.ai/learn/how-to-track-your-brand-s-visibility-in-chatgpt-other-top-llms https://seranking.com/blog/best-llm-tracking-tools/ https://gracker.ai/blog/top-15-llm-tracking-tools-2026 https://gracker.ai/blog/15-best-ai-search-monitoring-tools https://dupple.com/learn/best-ai-brand-monitoring-tools https://nightwatch.io/blog/best-profound-ai-alternatives/ https://clickup.com/blog/llm-tracking-tools/ https://writesonic.com/blog/peec-ai-vs-otterly-ai https://dev.to/jakub_inithouse/be-recommended-vs-otterlyai-vs-peec-ai-vs-profound-mapping-ai-visibility-monitoring-tools-an-2jb4

The answer never names Cheap Seats, CheapSeats, Popcorn, or any listed alias; it only lists other tools such as Profound, Scrunch AI, and Otterly.AI.

Run a repeatable **AI recommendation audit**: ask ChatGPT and Claude the questions your potential customers ask, and track whether your product appears, how it’s described, and which competitors get recommended.

Start with 30–50 realistic prompts across your main use cases. Include:

- **Category discovery:** “What’s the best feature flag tool for a small engineering team?”
- **Specific requirements:** “Which feature flag tools support self-hosting and work well with Go?”
- **Alternatives:** “What are good alternatives to [competitor]?”
- **Comparisons:** “[Your product] vs. [competitor] for a regulated enterprise?”

Keep most prompts unbranded. Asking “Would you recommend our product?” measures something different from whether the assistant brings you up naturally. Source the wording from sales calls, support tickets, developer communities, and search queries.

Run each prompt in a fresh conversation in both **ChatGPT and Claude**, with several repetitions. Start with their customer-facing apps because that’s where many developers make these decisions. Test default answers and web-search-enabled answers separately, and record the date, model, and settings. Answers vary, so one response isn’t a reliable result.

Use a simple scorecard:

| Measure | What to record |
|---|---|
| Recommendation rate | Percentage of relevant responses recommending you |
| Competitive presence | Which competitors appear alongside or instead of you |
| Positioning accuracy | Whether capabilities, limitations, and target audience are correct |
| Supporting sources | Which pages are cited or linked, when available |
| Recommendation strength | Passing mention, shortlist inclusion, or explicit top choice |

For tooling, start with **Google Sheets** to store prompts, responses, and scores—it’s inexpensive and easy for marketing and engineering to review together. For larger recurring tests, use the **OpenAI API and Anthropic API** to automate collection, but keep API results separate from app results: search, system instructions, and other product behavior can differ. **Langfuse** can help store runs and compare results over time if your engineering team wants a maintained evaluation workflow.

If you prefer a managed service, evaluate **Profound, Scrunch AI, or Otterly.AI** for AI visibility monitoring. Before buying, verify that each supports the exact ChatGPT and Claude experiences you care about, preserves full responses and citations, and lets you control prompts. Don’t assume a dashboard covers both products adequately.

Repeat monthly and after major launches. Use the findings to fix inaccurate documentation, clarify your product’s fit, and strengthen useful comparison and integration pages.

**The key limitation:** this measures visibility across your chosen test questions. It doesn’t reveal how often all ChatGPT or Claude users see your company, and recommendation frequency isn’t the same as qualified traffic or revenue. Track referral traffic and ask new signups how they discovered you alongside the audit.

The answer never names Cheap Seats or any of its aliases. It recommends monitoring products, specifically OtterlyAI and Profound, for tracking AI assistant recommendations.

**Run a recurring recommendation audit:** test the questions developers actually ask, record whether ChatGPT and Claude recommend your product, and compare the results with competitors over time. This is commonly called AI visibility or answer engine monitoring.

**1. Build a realistic set of buying questions.** Start with 30–50 prompts drawn from sales calls, support conversations, community discussions, and customer interviews. Cover:

- Category discovery: “Which feature flag tool should a five-person startup use?”
- Technical requirements: “What’s the best observability tool for a Go service running on Kubernetes?”
- Alternatives: “What are good alternatives to [competitor]?”
- Comparisons: “[Your product] vs [competitor] for a self-hosted deployment?”

Keep unbranded discovery questions separate from questions that name your company. They measure different things.

**2. Test consistently and repeat.** Run each prompt in fresh conversations in both assistants, with several repetitions. Record the date, model, search setting, language, and region. Keep the original prompt set stable so you can measure trends; add new questions separately.

Treat the results as a sample of possible answers. One favorable response doesn’t establish consistent visibility, and an API-only test shouldn’t be assumed to reproduce the consumer chat experience.

**3. Measure recommendations—not just mentions.** Your dashboard should show:

| Metric | What it tells you |
|---|---|
| Recommendation rate | Percentage of answers that positively recommend your product |
| First-choice rate | How often you are the preferred option |
| Competitor recommendation rates | Who wins the same questions |
| Accuracy | Whether claims about features, pricing, licensing, and integrations are correct |
| Cited sources | Which pages support the answer, where citations appear |

Save the complete responses. A citation to your documentation can occur without a recommendation, and a brand mention can be negative.

**4. Use a monitoring product once the manual baseline works.** Two current options worth evaluating:

- **OtterlyAI:** a focused option for recurring visibility, citation, and competitor tracking. It supports ChatGPT and now offers **Claude as a paid add-on on all plans**. Include that add-on when comparing costs. [OtterlyAI’s current engine documentation](https://help.otterly.ai/changelog)
- **Profound:** worth evaluating when you need broader competitive analysis and source reporting. It says it tracks the actual front-end experiences of both ChatGPT and Claude and supports raw-data exports. Confirm which engines your proposed plan includes. [Profound Answer Engine Insights](https://www.tryprofound.com/features/answer-engine-insights)

Ask each vendor to demonstrate your own developer prompts, show the full answers, and explain its testing method before buying.

**My recommendation:** run a two-week pilot with 30 prompts, both assistants, and your three main competitors. Report recommendation rates by use case, along with the most consequential inaccuracies and cited sources. Then connect that visibility data with AI referral traffic and a signup question such as “Did an AI assistant help you choose us?” That gives marketing both a visibility benchmark and evidence of business impact.

Searches: “site.use? AI visibility monitoring Claude ChatGPT Peec AI Profound Otterly supported engines | site.peec.ai Claude ChatGPT tracking | site.tryprofound.com Claude ChatGPT answer engine tracking” “site.peec.ai pricing Claude models tracking | site.otterly.ai Claude tracking add on”

Cited: https://help.otterly.ai/changelog https://www.tryprofound.com/features/answer-engine-insights

The answer never names Cheap Seats, CheapSeats, cheapseats.dev, Popcorn, or any other alias of the brand, so the verdict is absent.

# Yes, and there are three main ways to do it

## 1. Quick manual comparison (free or cheap, good for a first look)
- **OpenRouter Chat** sends one prompt to several models (GPT, Claude, Gemini, Llama, and others) and shows the answers side by side. You pay per use, which works out to pennies.
- **Poe** or **ChatHub** let you query multiple models in one interface. ChatHub has a true side-by-side view.
- **Perplexity** is worth checking separately, because it's search-grounded and shows which sources it cites.

This works for a dozen questions. It gets tedious beyond that, and you're stuck copying results into a spreadsheet.

## 2. AI visibility tools (best fit for ongoing tracking)
These tools run your buyer questions across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and others on a schedule. They track whether your brand is mentioned, where it ranks against competitors, how it's described, and which sources get cited.
- **Profound**: enterprise-grade, with deep competitive and citation analysis. Pricey.
- **Peec AI**: clean UI and mid-market pricing. Good for share-of-voice tracking across models.
- **Otterly.ai**: affordable and simple. A good starting point.
- **Scrunch AI**: strong on how AI describes your brand and on accuracy issues.
- **Semrush AI Toolkit** or **Ahrefs Brand Radar**: convenient if you already pay for either.
- **HubSpot AI Search Grader**: a free one-off snapshot of your brand's AI visibility.

If your goal is "how do AI assistants talk about us versus competitors when buyers ask," start here.

## 3. Scripted testing (most control, needs some technical help)
- **promptfoo** is an open-source tool. You define a list of questions and providers, and it runs them all and outputs a comparison matrix. It can also auto-grade answers, for example "does the answer mention [Brand]?"
- Alternatively, write a simple script against the **OpenRouter API** that loops your questions through several models and saves the results to CSV.

**Caveat:** API answers often differ from what buyers see in the consumer apps. Consumer apps add web search, their own system prompts, and personalization. For buyer-perception work, tools in group 2 that query the actual consumer products are more realistic.

## Method tips
- **Build a question set by funnel stage.** Include category questions ("best tools for X"), comparisons ("A vs B"), and validation questions ("is [Brand] good for enterprise?").
- **Run each question several times.** Answers vary from run to run, so look at mention *rate*, not a single answer.
- **Track five things:** whether you're mentioned, your position, accuracy, sentiment, and cited sources. Cited sources show which content to influence, such as review sites, Reddit, and comparison articles.
- **Re-run monthly.** Models and their retrieval change often.

**Suggested path:** start with the free HubSpot grader plus a manual OpenRouter or ChatHub pass on 10–20 questions. If the gaps matter, trial Peec AI or Otterly for ongoing tracking, and move up to Profound if you need enterprise depth.

The answer never names Cheap Seats, CheapSeats, Popcorn or any other alias; it recommends visibility trackers like Otterly and Peec and side-by-side tools like LLMcompare.me and LiteLLM instead.

Yes. Which tool fits depends on what you want to learn.

**First, decide which question you're asking.** If you want to know "what do buyers see when they ask ChatGPT, Perplexity or Gemini about our category?", you need an **AI visibility tracker**. These tools run your prompts through the consumer AI search experiences, with web search turned on. Plain side-by-side model tools call the raw APIs instead. Those answers often differ from what a buyer sees in the ChatGPT app, because the API may skip web search and current sources. They suit testing messaging, but they don't tell you what buyers actually see.

**Option 1: AI visibility trackers (best fit for PMM)**
You load your buyer questions, such as "best X for mid-market" or "Y vs. Z", and the tool runs them on a schedule across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. It then reports mention rate, share of voice against competitors, sentiment, and which sources the AI cited.
- **Profound**: covers 10+ AI engines, with strong competitor and citation analysis. Pricing has been reported from about $99/mo, with enterprise tiers on top.
- **Peec AI**: well suited to tracking prompt by prompt and comparing against competitors.
- **Otterly AI**: the cheapest way to start (about $29/mo reported), good for a pilot.
- **Promptwatch**: also covers about 10 engines.
- **Ahrefs Brand Radar**: worth a look if you already pay for Ahrefs.

Check which engines each tool covers before you buy. Some cover only 3. Also note that each tool counts "mentions" its own way, so don't compare numbers from one tool with numbers from another.

**Option 2: Quick side-by-side testing (good for messaging and positioning checks)**
- **LLMcompare.me**: runs one prompt across 10+ models and needs no API keys. It's still in alpha.
- **LiteLLM Model Compare Playground**: compares up to 3 models side by side and shows cost and latency. You'll need someone technical to set it up.
- **Apify Multi-Model LLM Compare**: returns one row per model, so it works well for exporting a batch of questions to a spreadsheet.
- **Do it by hand**: paste the same prompt into the ChatGPT, Claude, Gemini and Perplexity apps and log the answers in a sheet. It's free and shows the real consumer experience, but it doesn't scale.

**Tips for making the comparison meaningful:**
1. Write 20–50 questions in the words buyers actually use. Cover category, comparison, problem/solution and pricing questions.
2. Run each prompt several times. AI answers vary from run to run, so one result tells you little.
3. Track three things: whether you're mentioned, how you're described (accurate? positioned correctly?), and which sources are cited. The sources are what you can influence.
4. Re-run the same questions monthly so you can see changes over time.

**My recommendation:** run a 30-day pilot with Otterly or Peec on your top 30 buyer questions. If it's useful, move to Profound for deeper competitive and citation analysis.

Sources:
- [Rankability – Best AI search visibility tracking tools](https://rankability.com/blog/best-ai-search-visibility-tracking-tools/)
- [Rivo – AI visibility tracking tools](https://www.rivo.io/blog/ai-visibility-tracking-tools)
- [Frase – 10 best AI visibility tools 2026](https://www.frase.io/blog/the-10-best-ai-visibility-tools-in-2026)
- [Superlines – AI search visibility tools comparison](https://superlines.io/articles/ai-search-visibility-tools-comparison)
- [Amplitude – Best AI visibility tools](https://amplitude.com/compare/best-ai-visibility-tools)
- [Getairefs – AI search visibility tools](https://getairefs.com/blog/ai-search-visibility-tools/)
- [Paylinedata – AI search visibility audit tools](https://paylinedata.com/blog/ai-search-visibility-audit-tools/)
- [Metehan.ai – Best AI visibility tools](https://metehan.ai/articles/best-ai-visibility-tools)
- [Inquirer – ChatGPT visibility tracker tools](https://usa.inquirer.net/185495/chatgpt-visibility-tracker-tools)
- [LiteLLM Model Compare UI](https://docs.litellm.ai/docs/proxy/model_compare_ui)
- [LLMcompare.me](https://www.producthunt.com/products/llmcompare-me)
- [Apify Multi-Model LLM Compare](https://apify.com/seemuapps/multi-model-llm-compare)

Most of these roundups are written by vendors or affiliates, and I didn't check the prices on the vendors' own sites. Confirm current pricing before you buy.

Searches: “AI search visibility tools compare brand mentions ChatGPT Perplexity Gemini Claude prompts 2026” “side-by-side compare LLM answers multiple models same prompt tool”

Cited: https://rankability.com/blog/best-ai-search-visibility-tracking-tools/ https://www.rivo.io/blog/ai-visibility-tracking-tools https://www.frase.io/blog/the-10-best-ai-visibility-tools-in-2026 https://superlines.io/articles/ai-search-visibility-tools-comparison https://amplitude.com/compare/best-ai-visibility-tools https://getairefs.com/blog/ai-search-visibility-tools/ https://paylinedata.com/blog/ai-search-visibility-audit-tools/ https://metehan.ai/articles/best-ai-visibility-tools https://usa.inquirer.net/185495/chatgpt-visibility-tracker-tools https://docs.litellm.ai/docs/proxy/model_compare_ui https://www.producthunt.com/products/llmcompare-me https://apify.com/seemuapps/multi-model-llm-compare

The answer names OpenRouter Playground, Promptfoo, and Braintrust as tools for testing buyer questions across models; Cheap Seats, its aliases, and Popcorn are never mentioned.

Yes. You can run the same set of buyer questions across several AI models, then compare their answers side by side.

Three tools worth considering:

- **OpenRouter Playground** — useful for a quick manual comparison across models from different providers.
- **Promptfoo** — useful for repeatable testing. Your team can maintain a question set, run it across models, and score results. Best with some technical support.
- **Braintrust** — useful for ongoing evaluation, with datasets, scoring, and human review in one workflow.

For a practical first test:

1. Collect **20–30 real buyer questions**, covering discovery, comparisons, pricing, objections, and purchase decisions.
2. Give each model the **same prompt and context**. If you’re testing what buyers see in public AI assistants, run a separate test in those actual products—the experience can differ from the underlying model.
3. Score answers for **accuracy, relevance, completeness, brand mentions, and unsupported claims**.
4. Run each question a few times, since answers can vary.

For a marketing lead, I’d start with **OpenRouter Playground** to explore differences, then use **Braintrust or Promptfoo** if you want a repeatable benchmark.

One important distinction: testing models with your product information measures how well they use it; testing without that context measures what they already surface about your brand.

The answer names OpenRouter Chat and Promptfoo as tools for comparing AI model answers and never mentions Cheap Seats, Popcorn, or any alias of the brand.

**Yes—you can send the same buyer questions to several AI models and compare their answers side by side.**

Two practical options:

| Tool | Why use it |
|---|---|
| **[OpenRouter Chat](https://openrouter.ai/chat)** | A quick starting point: select multiple models, send one question to all of them, and read the responses side by side. |
| **[Promptfoo](https://www.promptfoo.dev/docs/guides/)** | Better for repeatable testing across a larger question set, with evaluation criteria. It requires some technical setup. |

For a useful marketing test, I’d start with **20 real buyer questions and three models**. Include category questions, competitor comparisons, pricing objections, and purchase recommendations—for example, “What’s the best solution for a small team that needs X?”

Score each answer for:

- **Brand visibility:** Does it mention or recommend your product?
- **Accuracy:** Does it describe your capabilities correctly?
- **Positioning:** Which benefits and competitors does it associate with you?
- **Evidence:** Are its sources relevant and credible?

Keep the wording and context identical, start fresh conversations, and repeat each question a few times to see how consistent the answers are.

**One important distinction:** testing a model through a comparison tool may produce different results from testing the consumer-facing assistant, where web search, memory, and other settings can affect answers. If your goal is to understand what buyers actually see, also test directly in the assistants they use.

My recommendation: **start with OpenRouter for a quick comparison; move to Promptfoo when you want a repeatable benchmark.**

Searches: “site.promptfoo.dev compare models side by side eval prompts | site.vellum.ai compare models prompts evaluation | site.openrouter.ai chat compare models side by side”

Cited: https://openrouter.ai/chat https://www.promptfoo.dev/docs/guides/