AI Hub · Foundations

How Google AI Overviews choose their sources

By the FlyRank team, Manchester · Published 15 October 2026

TL;DR: Google says AI Overviews and AI Mode link to pages that support the response, found through related searches run across subtopics, a process it calls query fan-out. A page must be indexed and eligible to show a snippet. Google states that no special markup or AI-specific file is required. Beyond that, the exact selection method is not published, so treat any detailed claim about “ranking factors” with caution.

What are AI Overviews and AI Mode?

AI Overviews are generated summaries that appear at the top of some Google results, and AI Mode is a conversational search experience. Both write an answer and show links to supporting websites.

Google notes that the two systems can use different models and techniques, so the pages they show may differ. It also says AI Overviews appear only when its systems judge them useful beyond classic results, which means many searches never show one.

For a business, the practical point is that these features sit on top of Google’s index. The foundation is still search, which is why the shared work described in AEO vs SEO carries over directly.

What is query fan-out?

Query fan-out is Google’s technique of running several related searches across subtopics and data sources to gather material for one response. A single question can therefore draw on pages that answer parts of it rather than the full wording.

Google describes this in its documentation on AI features and your website (Google Search Central). The system issues multiple related searches, then identifies additional supporting pages while it generates the answer.

Consider a customer asking how to choose a web design agency. The system might look separately at pricing, timescales, portfolio quality, and platform choices. A page that answers one of those sub-questions well can be used even if it does not target the headline question.

This has a direct writing consequence. Covering the connected questions around a topic gives you more routes into the answer. Our article on topical authority explains how to plan that coverage.

Key takeaway: Google answers one question by searching for many. Pages that answer the sub-questions clearly have more ways to be selected.

What must a page have to be eligible?

According to Google, a page must be indexed and eligible to appear in Search with a snippet. There are no additional technical requirements for appearing as a supporting link in AI features.

That gives a short checklist to verify for every important page.

Google also states that meeting every requirement does not ensure a page will be crawled, indexed or served. Eligibility is a minimum, not a promise.

What makes a page more likely to be used?

Google does not publish a list of selection signals for AI features, but its guidance points to standard fundamentals: helpful content, accessible text, strong internal links and a good page experience. Writing that answers a sub-question directly is easiest to use.

Because the exact method is not public, it is better to rely on principles that also help ordinary search. These are practical steps we recommend, based on how passages are retrieved rather than on any confirmed ranking factor.

• Put the answer first. Start each section with a one or two sentence response to the heading question.
• Keep paragraphs standalone. Name the subject in the paragraph so it still makes sense when quoted alone.
• Cite your sources. Link to primary documentation for any factual or technical claim.
• Use tables and lists where a comparison or sequence is clearer than prose.
• Cover related questions on the same page or in connected pages.
• Be accurate and current. Review and date important pages so facts are not out of step with each other.

Be cautious with statistics circulating in the industry about citation rates, word counts or ranking positions. Many come from third-party studies of limited scope, and they describe correlations, not Google’s method. Use them as hypotheses to test on your own site.

You can apply the same discipline to your own content using the AI content quality checklist. If you want a team to run it across your key pages, our answer engine optimisation services cover the audit and rewrite.

What does Google say you do not need?

Google says you do not need special schema.org structured data, AI-specific machine-readable files, AI text files or new markup to appear in AI Overviews or AI Mode. Standard SEO best practice applies.

This is a useful corrective to a lot of advice online. Adding markup is fine, and structured data should match the visible text on the page, but it is not a gate to entry. We explain what schema is genuinely good for in schema markup for AI search.

The same goes for files such as llms.txt. The proposal is documented at llmstxt.org, but Google states that AI-specific files are not needed for its AI features. Our llms.txt guide looks at where it may still be a cheap addition.

If you want to restrict how your content is used, Google offers controls: nosnippet, data-nosnippet, max-snippet and noindex limit what appears, and the Google-Extended token addresses AI training and grounding in other Google systems. Changes need time to be recrawled.

How can you see AI feature traffic?

Google says clicks from AI features are counted in the Search Console Performance report under the Web search type. They are not reported as a separate segment, so you cannot isolate them directly in that report.

This limits what you can prove. You can see overall impressions and clicks for pages and queries, and you can compare changes after updating content. You cannot confirm that a particular visit came from an AI Overview using Search Console alone.

Many teams combine several signals: manual checks of target queries, tracking of cited pages over time, and analytics on referral sources from other AI platforms. See measuring AI search visibility for a practical approach, including how to record results without overstating them.

Key takeaway: make key pages indexable and snippet-eligible, answer the sub-questions directly, cite sources, and measure carefully. There is no confirmed shortcut.