How does ChatGPT choose its sources?

When it searches the web, ChatGPT turns the question into one or more searches, uses third-party search providers including Bing and its own crawler, reads a few pages and cites the ones it takes information from. To be picked, a page must be readable by OAI-SearchBot, present in the indexes and have a passage that answers directly.

How does Perplexity choose its sources?

Perplexity searches the web for almost every question, in its own index built with PerplexityBot, and shows numbered sources next to the answer. Passage relevance, page freshness and the bot being able to read the page are what count.

How does SEO for Copilot work?

Copilot uses Bing's index, so the starting point is SEO for Bing: the site verified in Bing Webmaster Tools, the sitemap submitted, pages indexed. The AI Performance report in Bing Webmaster Tools shows how often your pages are cited in Microsoft's AI answers and for which searches.

Does Gemini use Google Search to answer?

Yes, when a question needs current information Gemini searches with Google and shows the sources. Google Search rules apply: the page must be indexed, readable by Googlebot and useful for the question.

How do you get cited by AI?

Three things: engine search bots free to read the site, pages with passages that answer customers' questions directly, and a brand described consistently on third-party sites too. Then check with repeated questions which sources get cited and work on the missing ones.

Does blocking GPTBot or Google-Extended remove the site from AI answers?

Not from search. GPTBot collects data to train OpenAI's models, while ChatGPT search relies on OAI-SearchBot. Google-Extended decides whether content can be used for Gemini and its models, but it does not change presence in Google Search, AI Overviews and AI Mode included.

GeoSuite

Back to Home

Guide · Sources of AI answers

How ChatGPT, Perplexity, Gemini and Copilot choose their sources

When an AI engine searches the web to answer, it turns the question into a few searches, takes results from an index, picks the passages that answer best and cites the pages it took them from. To be a source you have to be in the right index, readable by the right bot and have the passage that answers.

TL;DR

Each engine searches a different index: ChatGPT also relies on Bing, Copilot on Bing, Gemini and AI Overviews on Google, Perplexity on its own. The choice happens passage by passage: the page that answers the sub-question clearly, is up to date when it matters and agrees with other sources wins. When the engine does not search, it answers from what it learned in training, and there what counts is the brand's mentions across the web over time.

The two paths of an answer

An AI engine answers in two ways, and different things matter for each.

From the model's memory

The model answers from what it learned in training, without searching. It cites no links and changes only with new model versions: the brand's mentions built up over the years are what count.

With a web search

The engine searches in real time, reads a few pages and cites the ones it takes information from. This is called grounding, and it is the path a page can count on as soon as it is indexed.

Where the main engines get their sources

The bot to allow and the place to look at data change from engine to engine.

Where the main engines get their sources
Engine	Where it searches	Bot to allow	Where to see the data
ChatGPT	Third-party search providers, Bing among them, and OpenAI's crawler	OAI-SearchBot, and ChatGPT-User for pages opened on request	Visits from chatgpt.com in analytics
Microsoft Copilot	Bing's index	Bingbot	The AI Performance report in Bing Webmaster Tools
Perplexity	Its own index	PerplexityBot	Visits from perplexity.ai in analytics
Gemini	Google Search	Googlebot	Visits from gemini.google.com in analytics
AI Overviews and AI Mode	The Google Search index	Googlebot	Search Console, together with web search

What weighs in the choice of a source

No engine publishes the formula. These are the factors the engines themselves describe and that show up when you repeat the same questions.

Accessibility

The bot must be able to read the page: no block in robots.txt, text present in the HTML rather than loaded later, reasonable response times.

Passage relevance

The engine picks a piece of a page, not the whole page. The paragraph that answers the sub-question in plain words, without detours, wins.

Freshness

For prices, versions, regulations and rankings engines prefer up-to-date sources. A visible and truthful last-updated date helps.

Agreement between sources

A fact confirmed by several independent sites is easier to reuse than one that appears only on the brand's own site.

Authority

The signals that matter for search, such as links and domain reputation, are still a filter: an unknown source on a sensitive topic is rarely picked.

How to become a source

Five steps, from the most technical to what happens off the site.

1. Let search bots in
In robots.txt allow OAI-SearchBot, PerplexityBot, Bingbot and Googlebot. Blocking the bots that collect data to train models, such as GPTBot, is a separate choice and does not remove the site from search.
2. Get indexed on Bing too
ChatGPT and Copilot also draw on Bing. Verify the site in Bing Webmaster Tools and submit the sitemap; IndexNow gets new pages there right away.
3. Write the quotable passage
For every important question, a two-sentence answer at the start of a section, with names, figures and conditions spelled out.
4. Get named off the site
Reviews, trade comparisons, forums and press are the sources engines cross-check for comparison and recommendation questions.
5. Check which sources get cited
Ask the engines your customers' questions, note the cited sources and work on those: sometimes it pays to get cited by a third-party page instead of writing a new one.

Frequently asked questions

How does ChatGPT choose its sources?: When it searches the web, ChatGPT turns the question into one or more searches, uses third-party search providers including Bing and its own crawler, reads a few pages and cites the ones it takes information from. To be picked, a page must be readable by OAI-SearchBot, present in the indexes and have a passage that answers directly.
How does Perplexity choose its sources?: Perplexity searches the web for almost every question, in its own index built with PerplexityBot, and shows numbered sources next to the answer. Passage relevance, page freshness and the bot being able to read the page are what count.
How does SEO for Copilot work?: Copilot uses Bing's index, so the starting point is SEO for Bing: the site verified in Bing Webmaster Tools, the sitemap submitted, pages indexed. The AI Performance report in Bing Webmaster Tools shows how often your pages are cited in Microsoft's AI answers and for which searches.
Does Gemini use Google Search to answer?: Yes, when a question needs current information Gemini searches with Google and shows the sources. Google Search rules apply: the page must be indexed, readable by Googlebot and useful for the question.
How do you get cited by AI?: Three things: engine search bots free to read the site, pages with passages that answer customers' questions directly, and a brand described consistently on third-party sites too. Then check with repeated questions which sources get cited and work on the missing ones.
Does blocking GPTBot or Google-Extended remove the site from AI answers?: Not from search. GPTBot collects data to train OpenAI's models, while ChatGPT search relies on OAI-SearchBot. Google-Extended decides whether content can be used for Gemini and its models, but it does not change presence in Google Search, AI Overviews and AI Mode included.