AI Tools We Track
A good result in one AI tool may not show up in another. Start with shared site basics. Then test each tool to find gaps.
Each AI tool finds pages, checks what they describe, and chooses which sources to name or link. The work below covers these steps across ChatGPT, Claude, Perplexity, Gemini, and Copilot.
What We Know
This guide uses our tests and public sources. AI firms do not share every detail of how they find and link to pages. Their tools can change without notice. Treat our observations as test findings, not a full map of how each system works.
Published Bot Rules
OpenAI separates search and training controls: OAI-SearchBot finds pages for ChatGPT search, while GPTBot may collect content for training. Allow the uses you want and publish clear, accessible pages. Bot access makes a page eligible for consideration; it does not guarantee a citation.
PerplexityBot finds pages for search results and citations, not base-model training. Perplexity-User visits pages at a user's request and generally ignores robots.txt. Clear claims and supporting sources help, but access does not guarantee a citation.
Anthropic has separate bots for search, user requests, and training: Claude-SearchBot, Claude-User, and ClaudeBot. It does not publish a complete ranking formula. Accessible pages, accurate claims, and independent sources remain useful foundations.
Googlebot crawls pages for Google Search. Google-Extended controls some Gemini Apps and Vertex AI uses, including training and grounding answers. It does not affect Google Search rankings. Clear business details and search fundamentals help without guaranteeing a Gemini citation.
Microsoft does not publish a complete formula for Copilot's source selection. Keep your pages accessible to Bing and make your business, services, and credibility clear. Test buyer questions over time to see when Copilot includes you.
Shared Site Basics
In our work, clear site basics tend to help across AI tools. The tools appear to share some ways of reading pages and checking facts.
Decide which uses you allow: search, visits users request, and training or use in AI answers. Each has its own controls. Read the current bot guides and check that robots.txt allows what you intend.
Choose JSON-LD types that fit the page, such as Organization, Service, Article, and Person. Keep each entity's @id stable across pages, and use FAQPage only for questions and answers visible on that page.
Keep your name, address, phone, and canonical URL consistent across your site and profiles. Use sameAs markup to connect official profiles. Conflicting details make it harder to identify the right business.
Include key answers in server-rendered HTML so tools can read them without running JavaScript. Put a clear answer below its question heading, using lists where they make steps or definitions easier to follow.
llms.txt and llms-full.txt give tools a plain-text guide to your site. Keep them aligned with live pages, alongside valid structured data.
Backlinks, news, business directories, and government or academic sources can support your claims. GitHub and Hugging Face may also matter for some brands. These independent sources help AI verify your business.
What To Fix First
How to Track Results
Track whether each answer mentions your business or cites your site, then compare those rates with competitors over time.
AI answers vary with model settings, source updates, and model versions. Repeat each question and compare at least two model versions per tool, looking for trends across runs.
Server logs show which AI bots visit your pages, while analytics show referral clicks from AI answers. Track these separately from citations: a bot visit, a citation, and a referral click measure different things.
Official Sources
Common Questions
No. Each has its own bot rules. They do not share a full formula for how they choose sources. Start with public pages that tools can read, correct structured data, matching facts, and proof of your claims. The rules can change without notice.
Choose based on the use you want to allow. OpenAI uses OAI-SearchBot for search and GPTBot for possible training. Anthropic has Claude-SearchBot, Claude-User, and ClaudeBot for separate uses. PerplexityBot handles search; Perplexity-User handles user visits. Google-Extended covers some Gemini and Vertex AI uses, not Google Search. Access does not promise a link in an answer.
OpenAI does not share a fixed list of search sources. It says OAI-SearchBot helps ChatGPT find pages for search summaries and snippets. Let that bot read your site if you want this use. Follow good search practices. No one search index is a proven path to a ChatGPT link.
Anthropic does not share a fixed list of search sources or a full rule for links. Its guides explain Claude-SearchBot for search, Claude-User for user visits, and ClaudeBot for possible training. Allow the uses you want and test real searches more than once.
Start with Organization or LocalBusiness and link official profiles with sameAs. Add Service, Article, BreadcrumbList, and Person where they fit the page. Use FAQPage only for visible answers and keep each @id stable across pages.
It gives tools that browse the web a plain-text map of your site. It does not promise a higher rank. Add llms-full.txt for more detail. Keep both files in sync with the live pages so the facts match.
Settings, sources, user context, and model updates can all change an AI answer. Repeat questions and compare at least two model versions per tool to track patterns across runs.
Start with shared foundations: accessible pages, valid structured data, consistent business details, credible sources, and plain-text files. Then test each AI tool to find gaps that need more work.
Some tools can run JavaScript, but support varies. Put key content, headings, and structured data in the initial HTML so tools can read them without running browser code.
Google AI Overviews use search results and known facts about people, places, and brands. Clear structured data, useful text, and trusted brand facts can also help other AI tools. Track more than one tool to see the full picture.
A tool that browses may link to a page within days after it finds and lists it. A model that answers from training may need a new training cycle, with no promise it will use the page. Most change in the first 30 to 90 days comes from browsing.
We track mentions and citations across ChatGPT, Claude, Perplexity, Gemini, and Copilot on a set schedule. We repeat questions and compare at least two model versions per tool. Server logs add bot visits and AI referral traffic, showing how site access differs from use in answers.
Use one main Organization or LocalBusiness entity with a stable @id and sameAs links to verified profiles. Keep your name, address, and phone consistent across the site, structured data, and business directories.
How Canonry Approaches It
We start with the site basics that AI tools share. Our 16 checks cover those parts of your site. Then we test buyer questions across tools to see where more work may help.
More to Read