Handbook for AI agents: how a website exists for AI apps
This handbook is everything the Parallel World lab knows about how AI apps (ChatGPT, Claude, Gemini, Perplexity, Copilot) and their bots find, read, trust and cite a website: the principles, and under each the claims they rest on. Each claim says how sure we are, where it comes from, what it does not prove and what a site owner can do. Hypotheses and refuted claims are included and labelled. Give this page to your AI agent and let it check your own site.
How to use it with your AI agent
Copy the prompt below into your agent (ChatGPT, Claude, Gemini, or a coding agent) and replace the site address. Every page of this handbook is also served as Markdown: add .md to the address, or send the header "Accept: text/markdown".
I want AI apps and agents (ChatGPT, Claude, Gemini, Perplexity, Copilot and their crawlers) to find, read and correctly describe my website: <MY SITE URL>.
1. Read the handbook at https://pw.migurei.ai/en/handbook and every page it links to (send "Accept: text/markdown" or add .md for a Markdown copy).
2. Check my site against each principle. Use only what anyone can read on public pages: do not log in, submit forms or buy anything.
3. For each change you propose, name the principle and the claim it rests on, with its evidence level. Treat hypotheses as experiments to test, not rules. Do not act on refuted claims.
4. Start with what blocks the earliest step of the chain: found, fetched, indexed, retrieved, trusted, cited.
5. Tell me what you could not check, and what each change will not prove.Evidence levels
- External sourceDocumentation, a standard, a survey or a measurement published by someone else. We read it; we did not measure it.
- Our lab observationWhat we saw in our own server logs or in answers of AI apps to our test questions, without a control group.
- Our lab measurementAn experiment on our own sites, graded against a plan written before it ran.
- HypothesisWhat we believe but have not shown. Nothing to do yet, except test it.
- RefutedA claim our data contradicted. It stays here so nobody relies on it again.
Contents
- 8 principles, 75 claimsA. Bots and AI appsHow a page gets from your server into an AI answer: which bot fetches it, which index the app reads, how a bot learns a URL exists, and what makes a page readable and trusted.
- 7 principles, 20 claimsB. Japanese buyersHow business buyers in Japan use AI apps to find and choose a service, what they check on the official site, and which sources they trust.
- 7 principles, 15 claimsC. Open and closedThe other side of group A. Group A asks whether a site exists for AI; this group asks who the site is open to and what it gets in return. Each door opened for AI is open to others too; a good site is open in the right places and closed in the right places.
- 9 principles, 10 claimsD. The agent worldHow the world of AI agents is changing: several clocks for one company, entities instead of pages, indexes with no door for site owners, identity by signature, and buyers who take AI answers into their own documents. Most of these are hypotheses.
- 80 claimsMore claims, not yet under a principleClaims in our knowledge base that no principle names yet. Same format as the principle pages.
What this handbook does not prove
- Most lab observations come from a few sites of our own (pw.migurei.ai and migurei.ai), in Japanese and English, over weeks, not years; some come from scans of public Japanese sites and from one partner site's logs. They show what happened there, not what will happen on your site.
- External sources are listed as we read them on the date shown; vendors change their crawlers and documents often. Each claim has a date by which we re-check it.
- No claim here is a score. Following every "what to do" does not guarantee that an AI app will name or cite your site.
- The claims are written in Vietnamese in our knowledge base; this English text is a translation, and the Japanese text is machine-translated from the English.