Free toolNo signup~4 min read

AI Search Simulator

How ChatGPT, Google AI Overviews, AI Mode, Gemini, Perplexity and Claude find and cite sources, engine by engine, and what decides whether they can use your site.

How AI search engines find and cite sources

Every AI engine answers in one of two ways: from what its model learned in training, or by searching the web first and writing from the pages it found. Only the second can see your site this week. Which one you get depends on the engine and, for some, on the question. This guide sets the six engines side by side and ends with what actually decides whether an engine can use your pages.

Engine by engine

EngineWhen it searchesThe crawler that fetches pagesWhat decides whether it can use you
ChatGPT The model decides per question. When it searches, it rewrites your question into its own queries and links the sources it used. OAI-SearchBot (search). GPTBot only collects training data. OAI-SearchBot allowed in robots.txt and let through by your host or CDN. OpenAI treats the two settings as independent.
Perplexity Every answer runs a web search and lists its sources. PerplexityBot PerplexityBot allowed and able to fetch the page, with your facts in the HTML.
Google AI Overviews Built on Google Search. Google shows an overview on some searches and not others. Googlebot Indexed and eligible for a snippet. Google says there are no additional requirements and no special optimizations.
Google AI Mode Answers every search from Google's index. Google says it may use different models and techniques from AI Overviews. Googlebot The same gates as AI Overviews: indexed, snippet-eligible, no nosnippet on the text that matters.
Gemini Grounds answers in Google Search when the model decides to. Googlebot. Google-Extended controls training and grounding use, not Search. Indexed in Google, with the facts readable without JavaScript.
Claude Searches when search is on; in the Claude apps that is a setting the user controls. Claude-SearchBot (search). ClaudeBot collects training data. Claude-SearchBot allowed and able to fetch your pages. With search off, only training data reaches the answer.

What the evidence supports

  • Access. A crawler that robots.txt blocks, or that your firewall turns away with a 403, cannot read the page, so the engine cannot cite it. This is the one lever with evidence behind it, and it only ever takes things away.
  • Facts in the HTML. OpenAI's and Anthropic's crawlers fetched JavaScript files but did not run them in Vercel's crawler study, so a price or a feature that only appears after JavaScript runs may never be read.
  • Index and snippet eligibility for the Google surfaces: no noindex, no nosnippet or max-snippet:0 on the text that matters.

What it does not

No markup, phrasing or page structure has credible evidence of getting a page cited. Google says structured data is not required for its AI features. Lists, tables and FAQ blocks may help a reader; there is no measured effect on AI citation. Our research hub sets out which claims hold up and which do not, with sources.

Check your own site

More free AI visibility tools

All 10 tools →
Free

GEO Readiness Quiz

15 questions, 3 minutes: a 0–100 readiness score and the fixes that matter first.

~3 minOpen
PopularFree

AI Visibility Checker

See whether ChatGPT and Google AI Overviews name your brand, what they say, and who they cite instead.

~30 secRun free
NewFree

Google AI Mode Checker

Read Google AI Mode's answer for any search, see every site it cites, and whether yours is one.

~15 secRun free

Want this on every prompt, on all 6 engines, on a schedule? See GEO tool pricing or read how the AI visibility audit works.