How ChatGPT, Perplexity, and Gemini Decide Which Sites to Cite
Updated 11 October 2026Free check: can AI read your site?
See your AI score before you read on
About 90 seconds · No email required · Free
Below is how AI engines return their sources, judging by their API documentation. Which pages they pick the vendors do not disclose, and the mechanisms of ChatGPT, Perplexity and Gemini differ noticeably.
Classic SEO is about position in a list. GEO is about letting AI select your page as a source. Each AI search engine has its own selection mechanism, and it helps to understand it before changing anything on your site.
The Three Pipelines
ChatGPT, Perplexity and Gemini all surface web content in their answers, but the path from your site to a citation differs for each. Here is how each one works.
ChatGPT: Responses API + web_search
According to OpenAI’s documentation, the Responses API has a built-in web_search tool: the model decides whether to search the web. The older web_search_preview remains for legacy integrations. OpenAI does not name the search service behind it; third-party descriptions mention Bing.
What it looks like in the API:
- User asks a question
- The model decides whether web search is needed; the response contains a
web_search_callitem (a search, opening a page, or searching within a page) - Then comes a
messageitem with the answer text - By default the answer carries inline citations for URLs from the search results as
url_citationannotations
What This Means for Optimization
- → Don't shut the search bot out. OpenAI says OAI-SearchBot surfaces sites in ChatGPT search; a site that blocks it in robots.txt will not appear in search answers.
- → Check Bing too. If third-party descriptions are right and ChatGPT draws on Bing, presence in Bing’s index helps as well; OpenAI does not confirm this.
- → Page load speed. This is our assumption: slow pages may be skipped during the fetch phase. There is no documentation on it.
- → Clear, quotable paragraphs matter. Content made of distinct, standalone statements is easier to quote accurately.
Perplexity: Sonar with Direct Citations
Perplexity is built specifically for web search and shows its sources with every answer. The Sonar API (Perplexity's model family) returns, per its documentation, a citations array (a list of URLs) and a search_results array with the title, URL, date and snippet of each result with every response.
Perplexity's documentation does not say how it picks pages or how many it uses. The answer carries numbered markers such as [1][3] that point to the search results.
From this comes our assumption, which Perplexity has not confirmed: content that is clearly segmented (headers, short paragraphs, bullet points) is easier to quote accurately.
What This Means for Optimization
- → Content segmentation. Use H2/H3 headers to create quotable sections. Each section should answer one question.
- → Freshness. Perplexity’s
search_resultscarry date fields (date,last_updated); whether freshness affects selection is not published. Update key pages and show a true date. - → FAQ schema. Question-and-answer pairs in FAQPage JSON-LD are easier for a machine to parse.
- → Domain authority. Our assumption: it may be considered in the initial retrieval phase as well.
Gemini: Google Search Grounding and AI Overviews
The Gemini API has a "grounding" feature that ties answers to Google Search. For current models it is enabled with the google_search tool; per Google's documentation google_search_retrieval is for older models. In the Interactions API (Google documentation, checked October 11, 2026) sources come back as url_citation annotations tied to a text span; other Gemini API response formats use different field names, so check the documentation.
Google AI Overviews is the summary at the top of Google search results. Google says AI Overviews and AI Mode work over the regular Search index; it does not describe how Gemini API answers relate to AI Overviews.
The difference from ChatGPT: Gemini searches Google. Google describes optimizing for generative AI search as optimizing for the search experience, and thus still SEO.
What This Means for Optimization
- → Google SEO remains the foundation. Gemini and AI Overviews build on Google Search. Classic SEO still matters.
- → E-E-A-T signals. Experience, Expertise, Authoritativeness, Trust — quality criteria from Google’s guidance; how far AI engines rely on them is not confirmed.
- → FAQPage schema. It helps a machine parse the page but does not guarantee inclusion in AI Overviews.
- → Page experience. Google advises a good page experience: slow pages are worse for users.
Yandex: Search with Neuro and source links
In Yandex, answers come from "Search with Neuro." Per Yandex's help, algorithms pick materials from the search results, a generative neural network analyzes them in real time, and the final answer comes with links to the materials it was assembled from. To check Yandex, look at whether your site is among those links.
Yandex Webmaster has a markup validator. Yandex supports not all Schema.org types and does not promise the data will appear in results, so follow the current list of supported types in its help.
The Common Signals Across All Three
Despite different architectures, AI search engines share a set of signals that make a site easier to read:
How to Check Your Citations
To know whether AI engines cite you, you have to ask them. For each engine that looks like this:
- ChatGPT: Use the Responses API with the
web_searchtool and checkurl_citationannotations in the output - Perplexity: Use the Sonar API and check the
citationsandsearch_resultsarrays in the response - Gemini: Use grounding via the
google_searchtool and check the annotations in the response (in the Interactions API these areurl_citation; confirm field names in the current Gemini API documentation) - Yandex: In "Search with Neuro" check whether your site is among the source links
Causabi monitoring on the paid plans does the same: every week it asks ChatGPT, Gemini, Grok and Claude your questions through their official APIs and the report shows whether you are mentioned.
See your AI score
The check takes about 90 seconds: Causabi rates how readable your site is to AI and shows what is getting in the way.
Check your AI score →Other ways to check
- Ahrefs Site Audit — a check of technical SEO signals (we have not tested it) that affect crawlability (not AI-citation-specific, but foundational)
- Surfer SEO — content optimization for search signals (not tested by us)
- Frase — content briefs and NLP optimization (not tested by us)
- Manual monitoring — run your target queries directly in ChatGPT/Perplexity/Gemini weekly and track if your domain appears. Time-consuming but free.
Related GEO guides
FAQ schema and AI citations
How FAQ markup relates to citations
The Complete llms.txt Guide
A proposed format for AI agents: what it is, how to add it, what is known about its use
GEO optimization for Wix
Platform-specific signals that affect your citation rate
GEO optimization for Tilda
Platform-specific signals that affect your citation rate
Apply this to your site — free, no signup
Check your site’s AI score
About 90 seconds · No email required · Free