Quick wins

# Publish cheapseats.dev/docs/costs with multi-run measured spend and budget config

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
- Cheap Seats vs Profound for tracking whether AI assistants recommend us: which should I pick?
- Cheap Seats vs Otterly.AI for a startup that wants to track AI visibility cheaply?
- What does it cost to run Cheap Seats every day, and how do I keep the spend under control?

## What the assistant said
P11 claude search objection: "numbers from one measured run, not an actual bill"; the codex search answer was absent.

## Who wins the question today
https://docs.litellm.ai/docs/proxy/users. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at https://cheapseats.dev/docs/costs 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
Publish the cost of one measured run (prompt count × 2 tracks × models) on the API path, next to the near-zero marginal cost on Claude Code and Codex subscriptions (noting it counts against subscription limits). Include the daily budget config snippet. Claude search currently cites only site PR 16's '$114/mo from one run', so confirm that figure across several runs first. popcorn.zone/docs/costs (PR 35) is on the old domain; port it to cheapseats.dev rather than wait on it.

## Done when
Re-running prompt 11, codex-sol search names Cheap Seats and cites cheapseats.dev/docs/costs.

## 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 cheapseats.dev/docs/grading with tracks, grade rules, sample answers, accuracy

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
- Cheap Seats or Semrush AI Toolkit for tracking our brand in ChatGPT and Claude?
- How does Cheap Seats decide whether an AI assistant recommended a brand?
- Is Cheap Seats a good fit for a developer-tools company checking whether ChatGPT and Claude recommend it?

## What the assistant said
P8 codex search: "advises the user to request Cheap Seats' classification rules and sample scored answers".

## Who wins the question today
https://thestory.is/en/journal/how-to-measure-brand-visibility-in-chatgpt/. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at https://cheapseats.dev/docs/grading 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 each question is asked on a search track (web search first) and a recall track (no tools), that answers are graded recommended, mentioned or absent, and what else is recorded: position, top pick, every product named, and objections. Show 2–3 real scored answers with their kept citations, so readers see that raw answers sit behind every number and not just a single score. Measure grader agreement against hand labels on a sample and publish the figure; the popcorn.zone version (PR 8) was never crawled.

## Done when
Re-running prompt 8, at least 2 of 4 answers describe the three grades and two tracks, and a search answer cites the 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.

# Publish cheapseats.dev/docs/self-hosting: trade-offs, ops burden, data handling

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 Cheap Seats a good open-source alternative to paid AI visibility tools?
- What are the downsides of self-hosting Cheap Seats to measure AI search visibility?
- Is Cheap Seats a good fit for a developer-tools company checking whether ChatGPT and Claude recommend it?

## What the assistant said
P7 codex search: "could not verify Cheap Seats' architecture, coverage, or features" and listed generic downsides from unrelated GitHub pages.

## Who wins the question today
https://semrush.com/blog/llm-monitoring-tools. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at https://cheapseats.dev/docs/self-hosting 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
List the real downsides honestly: someone must run the container, upgrade it and own security, and results are noisy so trends need repeated runs. Pair each with its mitigation: single Docker container with SQLite, or Helm or an exe.dev VM; the daily spend cap; prompts and answers go only to the providers you configure; every answer and citation is kept for audit. Address 'API differs from the consumer app' by explaining the search track. Name who it is not for (teams without an engineer) so the page reads as credible.

## Done when
Re-running prompt 7, codex-sol search cites the page and lists Cheap Seats-specific downsides instead of generic ones.

## 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 cheapseats.dev/docs/agencies on multi-client setup; confirm workspace support

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 Cheap Seats a good fit for an agency tracking AI visibility for many clients?

## What the assistant said
P9 codex search: GetIntel "explicitly documents separate client tracking, branded scheduled reports".

## Who wins the question today
https://getintel.ai/solutions/agencies/. Read it and make sure our version answers everything it does, better.

## Your task
Write or update the page at https://cheapseats.dev/docs/agencies 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 an agency models clients: one prompt set per client in TOML, multiple sites per brand, fix-list owners, and per-prompt-set rankings exported via the JSON API or MCP for reporting. First confirm whether separate client workspaces, per-client budgets and branded or white-label reports exist. If they do not, file them as a product gap, because GetIntel and LLM Pulse win on exactly those.

## Done when
Re-running prompt 9, at least one search answer cites the page and states how Cheap Seats separates clients.

## 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.