How Google AI Overviews Choose Sources (2026 Guide)

By the Alphaa team — we build an AI agent that checks what ChatGPT, Gemini, Claude and Perplexity say about local businesses. This guide sticks to what Google has published about its own system.

Short answer: Google AI Overviews choose sources by running several related searches behind your one query, a technique Google calls query fan-out, then linking pages from those results that support the answer. To be eligible a page only has to be indexed and able to show a snippet. No special markup or file is required.

Most advice on this topic is guesswork dressed up as a ranking-factor list. Google has actually documented the outline of how it works, and the documented version is simpler and more useful than the folklore. This guide walks through it and marks clearly where Google stops explaining.

How do Google AI Overviews choose which sources to show?

AI Overviews choose sources from normal Google Search results, gathered across several searches instead of one. Google's documentation on AI features says AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources to develop a response, and that this lets them show a wider and more diverse set of helpful links than a classic results page.

In plain terms there are three stages. The system breaks the question into parts. It searches for each part using the same index and ranking systems as ordinary Search. Then a Gemini model writes the summary and attaches links to pages that back up what it wrote. Google has not published how the final links are chosen from the candidates, and anyone who gives you a precise formula for that last step is inventing it.

What is query fan-out and why does it change who gets cited?

Query fan-out is Google's name for turning one question into many searches at the same time. In its May 2025 announcement Google described it as breaking your question into subtopics and issuing a multitude of queries simultaneously on your behalf.

This matters because the page that gets cited does not have to rank for the words the person typed. Take "best family dentist near me that takes nervous patients". The system may look separately for family dentists in the area, for sedation options, for what patients say in reviews, and for opening hours. A practice with a clear page on treating anxious patients can be picked up by that one sub-search even if it is nowhere near the top for "best family dentist".

The consequence is that narrow, specific pages have more ways in than they did with ten blue links. A page that answers one sub-question completely is a better candidate than a long page that touches ten of them lightly.

What makes a page eligible to be a source?

A page is eligible if it is indexed and can be shown in Google Search with a snippet. Those are the only two technical requirements Google states, and it adds that there are no additional requirements to appear in AI Overviews or AI Mode.

• Indexed. Google has to have crawled and stored the page. A page blocked in robots.txt, marked noindex or never discovered cannot be a source.
• Snippet-eligible. Google must be allowed to show text from the page. A nosnippet rule removes the page from consideration as quoted support.
• Within Search policies. The same spam and content policies that apply to ordinary results apply here.

Eligible is not the same as chosen. Google is explicit that meeting every requirement does not mean a page will be crawled, indexed or served. Eligibility gets you into the pool; usefulness for a specific sub-question gets you picked.

Do you need schema markup or an llms.txt file to appear?

No. Google says you do not need to create new machine-readable files, AI text files or markup to appear in these features, and that there is no special schema.org structured data to add. That is a direct statement from the documentation, and it rules out a lot of what is sold as AI Overview optimization.

Structured data still has a job. In its post on performing well in AI experiences, Google asks site owners to make sure structured data matches the visible content on the page. Markup helps Google understand a page it already wants to use; it is not a ticket in. We cover what is worth adding in schema markup for AI search. An llms.txt file is a separate matter: it can help other assistants navigate your site, as described in how to create an llms.txt file, but Google does not use it for AI Overviews.

How do you make a page easier to pick as a source?

You make a page easier to pick by letting it answer one specific question completely, in text Google can read and quote. None of the steps below is a trick. They follow from fan-out and from Google's guidance on helpful, people-first content.

1. Confirm the page is indexed. Use URL Inspection in Search Console. If it is not indexed, nothing else on this list matters yet.
2. Check you have not blocked snippets. Look for nosnippet or a very low max-snippet value in your meta robots tag. Some themes and SEO plugins set these without telling you.
3. List the sub-questions behind your main query. For a service business these are usually price, area served, availability, qualifications and what happens during the job.
4. Give each sub-question its own clear answer. A heading that states the question and a first sentence that answers it. Put real figures, places and conditions in the answer.
5. Keep the answer in the HTML. Text that only appears after a script runs, inside an image or behind a tab is harder for any system to quote.
6. Say something only you can say. Google's advice is to focus on unique, non-commodity content rather than restating what every other page says. Your own prices, process and local detail are that content.
7. Keep business facts consistent. For local queries Google also draws on your Business Profile, so hours, services and address should match your site. See how your Business Profile feeds AI answers.

The writing side of this is covered in detail in how to write content AI quotes.

How is AI Mode different from AI Overviews?

AI Mode uses the same fan-out approach as AI Overviews but runs it further, in a separate conversational tab. Google says both are powered by a custom version of Gemini 2.5, and that its Deep Search feature in AI Mode can issue hundreds of searches to build a fully cited report.

For a site owner the practical point is that you do not optimize for them separately. One eligible, clearly written page serves both.

How can you see whether AI Overviews cite your site?

You can see it in Google Search Console, with limits. Google's generative AI performance report shows how many times links to your site were shown in AI Overviews and AI Mode, and which pages earned those impressions. It reports impressions only. It does not give clicks, position or the queries involved, and Google says not every site has access yet.

Traffic from these features is also included in the main Performance report under the Web search type, mixed in with ordinary results. For anything more specific you have to run the searches yourself and note which pages are linked, repeating it over several days because the answer and its links change. Our guide to measuring AI visibility sets out a method.

One honest note about our own product: Alphaa checks ChatGPT, Gemini, Claude and Perplexity. It does not track Google AI Overviews or AI Mode, so for those two Search Console and manual checks are the tools.

Can you stop AI Overviews using your content?

Yes, with the same controls that limit snippets in Search. Google lists nosnippet, data-nosnippet, max-snippet and noindex as the ways to restrict what is shown from your pages in AI features.

There is no switch that removes you from AI Overviews while keeping full snippets in ordinary results. Google points to its separate Google-Extended control for limiting AI training and grounding in some of its other systems, which is not a control for AI features in Search. For most businesses opting out is the wrong move anyway. AI Overviews now appear in up to about 48% of commercial-intent searches (BrightEdge, 2026), and a link there is one of the few ways left to be seen on those pages. The wider picture of who each crawler is and what blocking it costs is in AI crawlers and robots.txt.

What should a local business do first?

Start by checking that your main service pages are indexed and not blocking snippets, then write one clear answer for each question a customer asks before booking. That covers the two published requirements and the one thing fan-out rewards.

After that the work is the same as for any AI assistant: accurate public facts, specific pages, consistent listings. Our guide to showing up in AI Overviews as a local business goes through the local signals, and the free 60-second check shows what ChatGPT, Gemini, Claude and Perplexity currently say about you.

What else do people ask about AI Overview sources?

Do AI Overviews only cite pages that rank on page one?

No, not necessarily. Because of query fan-out, a page can be picked up by one of the related searches Google runs even when it does not rank well for the words the person actually typed.

Does Google use different ranking for AI Overviews?

Google says AI Overviews draw on its core Search systems and that normal SEO best practices still apply. It has not published a separate set of ranking factors for choosing the links inside an overview.

How many sources does an AI Overview show?

It varies by query and Google has not published a fixed number. The links shown can also differ between two people asking the same thing, or for the same person a day apart.

Will being cited in an AI Overview bring traffic?

Sometimes, though many people read the summary and stop. Roughly two-thirds of Google searches now end without a click to a website (SparkToro/Similarweb, 2026), so treat a citation as visibility first and visits second.

Does paying for Google Ads help you appear in AI Overviews?

No. The links in an AI Overview come from organic Search systems, and advertising spend is not an input. We go through the evidence in whether paid ads affect AI recommendations.