How ChatGPT, Perplexity and Google's AI Overviews choose what to cite

More people now ask ChatGPT, Perplexity, Google's AI Overviews or Microsoft Copilot a question instead of scrolling through a list of links. When these tools answer, they often cite a few websites as sources. Business owners reasonably want to know how those sources get chosen. The honest answer is that nobody outside these companies knows the exact details. But each company has published a fair amount about how its systems find and use web pages, and that's enough to act on sensibly.

Two different things that get called "AI"

It helps to separate two ways AI tools can know about your business:

What the model learned in training. Large language models are trained on large amounts of text, which may include public web pages. That knowledge is fixed at a point in time and the model doesn't cite where it came from. You can't directly control it.
What the tool looks up live. When a tool searches the web to answer a question, it fetches current pages and usually links to them. This is "AI search", and it's where citations come from.

This section focuses mostly on the second, because it's the part you can influence and measure.

The general pattern (as far as is publicly known)

Most AI search features work in a broadly similar way, though each company's details differ and change:

The question is interpreted. The tool may rewrite it or break it into several searches.

Pages are retrieved from a search index. That might be the company's own index, a partner's, or both.

The model reads the retrieved content and writes an answer.

Sources are attached. Some of the pages used are shown as citations or links.

The important implication: if your page isn't in the index the tool searches, it can't be cited. Being findable by ordinary search engines is still the foundation.

Every citation starts in the index. If your page isn't in the index the tool searches, it can't be cited.

Read this as text

The question is interpreted. The tool may rewrite it or break it into several searches. Google calls this query fan-out. The diagram shows three searches as an example; the real number varies.
Pages are retrieved from a search index.That might be the company’s own index, a partner’s, or both.
The model reads the retrieved content and writes an answer.
Sources are attached. Some of the pages used are shown as citations or links.

Google: AI Overviews and AI Mode

Google has published guidance on how its AI features relate to websites. The key points from Google Search Central's documentation on AI features:

AI Overviews and AI Mode draw on Google's normal search index. To be eligible to appear as a supporting link, a page must be indexed and eligible to be shown in Google Search with a snippet.
Google says there are no additional technical requirements and no special optimization needed to appear in these features. Its advice is to follow the same fundamentals it recommends for Search in general.
The usual controls apply. Snippet controls such as nosnippet and max-snippet affect how your content can be used in these features too.
Google says AI features may use a technique it calls "query fan-out", running several related searches to find a wider range of supporting pages than a single search would.

Google also has a separate robots.txt token called Google-Extended. Google's crawler documentation says it lets sites control whether content is used to help train and ground Gemini models, and that it doesn't affect a site's inclusion or ranking in Google Search. Blocking it is not a way to opt out of AI Overviews.

OpenAI: ChatGPT search

OpenAI documents several web crawlers, each with a different purpose:

OAI-SearchBot is used to surface websites in ChatGPT's search features. OpenAI's documentation says sites that block it won't be shown in ChatGPT search answers, though they may still appear as plain navigational links.
GPTBot is used to crawl content that may be used to train OpenAI's models. Blocking it is about training, not search.
ChatGPT-User is used when a user's request causes ChatGPT to visit a page.

OpenAI has also said ChatGPT search uses third-party search providers as well as its own crawling. So being well indexed in major search engines, including Bing, is sensible.

Perplexity

Perplexity publishes details of its crawlers too. PerplexityBot is used to surface and link websites in Perplexity's results. A separate user agent handles fetches triggered by user requests. Perplexity shows numbered citations for most answers, so being crawlable by PerplexityBot is the main eligibility step.

Microsoft Copilot

Copilot's web answers are grounded in Bing search results. That makes Bing Webmaster Tools worth setting up. It's free, and lets you submit your sitemap and see how Bing indexes your site.

The bots side by side:

Company	Used for search answers	Used for model training
Google	Google's normal search index	Google-Extended (blocking it doesn't affect Search or AI Overviews)
OpenAI	OAI-SearchBot (ChatGPT-User visits pages when a user asks)	GPTBot
Perplexity	PerplexityBot	Not covered here
Microsoft Copilot	Bing search results	Not covered here

So what decides which page gets cited?

None of these companies publishes a scoring formula. Based on what they do say, and on how search systems generally work, these factors are reasonable to focus on:

Eligibility: crawlable by the relevant bots, indexed, not blocked from snippets.
Relevance: the page directly addresses the question asked.
Clarity: the answer is stated plainly and is easy to extract, not buried in marketing language. See Writing content AI assistants can quote.
Trust and reputation: Google's long-standing guidance on helpful, reliable content applies to its AI features too. Other tools likely weigh source reputation as well, and what others say about you matters. See Reviews, directories and mentions.
Freshness, for questions where recent information matters.

Results also vary a lot. The same question asked twice, or by two people, can produce different answers and different citations. Treat any single result as a snapshot, not a ranking.

What this means for you

Check your robots.txt. Make sure you're not accidentally blocking the search bots above. Decide deliberately whether to allow training bots like GPTBot. See Sitemaps & robots.txt explained.
Get your SEO fundamentals right. Indexed pages, clear titles, good structure, fast and mobile-friendly. See Is SEO dead?.
Set up Bing Webmaster Tools as well as Google Search Console.
Write pages that answer real questions directly.
Build your reputation off your own site: reviews, directory listings, mentions.
Check what AI tools say about you periodically. See How to check whether AI tools mention your business.

Frequently asked questions

Can I pay to be cited in AI answers?

Not in the organic answers, according to these companies' public statements. Some platforms are experimenting with or running ads in or near AI answers, which are labeled as ads. Be wary of services that promise guaranteed AI citations.

Should I block AI bots to protect my content?

It's a business decision. Blocking search-related bots (like OAI-SearchBot or PerplexityBot) can remove you from those tools' answers. Blocking training bots (like GPTBot or Google-Extended) affects training but, per the companies' documentation, not search visibility. Many businesses that want to be found allow the search bots and decide separately about training.