How Does Google AI Mode Pick Sources? Query Fan-Out Explained

Everything below comes from three Google publications: the May 20, 2025 announcement that introduced AI Mode to everyone in the US, the Search Central guidance published the next day, and the AI features documentation that governs eligibility. Where Google is silent, this article says so instead of guessing.

What Google says fan-out is

The AI Mode announcement describes the mechanism in one sentence: AI Mode “uses our query fan-out technique, breaking down your question into subtopics” and issuing “a multitude of queries simultaneously on your behalf.” Google frames the payoff as reach, letting Search “dive deeper into the web than a traditional search on Google.”

The same post describes Deep Search as the heavy version: it “uses the same query fan-out technique but taken to the next level” and “can issue hundreds of searches, reason across disparate pieces of information” to produce a cited report.

Read as an engineer, the implication is simple. Each sub-query is a retrieval event against the normal index. A page gets pulled into the answer because it was a good result for one of those sub-queries, not because it was a good result for the sentence the user typed.

A worked example of one question fanning out

Google does not publish the sub-queries it generates, so the list below is an illustration of the shape, not Google data. Take a question a homeowner might ask AI Mode: “Is a kitchen remodel in Roseville worth it, and what should I expect?” A plausible fan-out looks like this:

1. What does a kitchen remodel cost in Roseville, CA?
2. How long does a kitchen remodel take?
3. Do kitchen remodels need a permit in Placer County?
4. What is the difference between cabinet refacing and replacement?
5. Which kitchen remodelers in Roseville have reviews and a license?
6. How do I budget for a kitchen remodel?

Six searches, six chances to be the retrieved page, and six places where a competitor with a plain, direct answer can take the slot. A remodeling site whose only page is a glossy “Kitchens” page with a gallery and a contact form is eligible for all six and a strong candidate for none. A site with a cost page, a timeline page, a permit explainer and a refacing-versus-replacement comparison is a strong candidate for four.

This is the same discipline we describe as content chunking: each sub-question gets a section that can stand alone when it is lifted out.

Eligibility is ordinary indexing, nothing more

Google’s AI features documentation says a page must be indexed and eligible to show with a snippet, and then: “There are no additional technical requirements.” On markup: “There’s also no special schema.org structured data that you need to add.”

The Search Central guidance spells out what the technical requirements mean in practice: “Googlebot isn’t blocked, the page works (Google receives an HTTP 200 (success) status code), and the page has indexable content.” If your pages fail those three, nothing about fan-out matters yet. That is why we treat index coverage as the prerequisite for every AI visibility project.

One nuance on structured data: Google asks that you make “sure that all the content in your markup is also visible on your web page.” Markup that mirrors the page helps Google parse it. Markup that invents facts the page does not show is a guideline violation, not a shortcut.

How to shape a page so sub-queries can find it

Google’s own advice for AI experiences is to “focus on making unique, non-commodity content,” and it notes that in AI search users “are asking longer and more specific questions, as well as follow-up questions to dig even deeper.” Longer, more specific questions are sub-queries by another name. The table below maps common sub-question types to the section that satisfies them.

Three habits cover most of the table:

• One sub-answer per H2. Phrase the heading as the question where it reads naturally, and answer it in the first sentence or two beneath it.
• State facts as facts. Prices, timelines, areas served and requirements go in plain sentences, lists or tables, in the server-rendered HTML.
• Keep FAQs visible. A question answered on the page can be retrieved. A question that exists only inside a schema block cannot be shown, and Google asks that markup match visible content anyway.

Preview controls limit you, by design

Google gives site owners four levers for what appears in listings, including AI formats, and the Search Central post names them: “Make use of nosnippet, data-nosnippet, max-snippet, or noindex to set your display preferences.” Then the sentence that matters for visibility: “More restrictive permissions will limit how your content is featured in our AI experiences.”

Two practical consequences. First, a max-snippet set years ago to protect headlines from aggregators now caps how much of your page can be quoted in an answer. Second, the lever people reach for by mistake, the Google-Extended token, does nothing here. It controls Gemini training and grounding, not Search, as we cover in a separate post. If you want to be quoted, audit your snippet directives and leave Google-Extended out of the decision.

Reading the effect in Search Console

There is no AI Mode filter. Google’s documentation says traffic from AI Overviews and AI Mode is included in the Performance report under the “Web” search type. You infer rather than filter:

• Query shape. Growth in long, question-shaped queries against a page is consistent with sub-query retrieval.
• Impressions without proportional clicks on specific pages. A page being shown inside an answer is still an impression. Watch the page-level trend, not the site total.
• Engagement after the click. Google says that when people click through from results with AI Overviews, “these clicks are higher quality, where users are more likely to spend more time on the site,” and it recommends looking at conversions, sign-ups and engagement rather than click counts alone.

Google also notes AI Overviews “show a wider range of sources on the results page,” which is the optimistic reading of fan-out for a smaller site: more slots, each won by a specific answer rather than by domain size. We lay out the full measurement approach in how to measure AI search visibility.

One caution on timing

Structural changes take effect when Google recrawls, and Google says that “can take anywhere from several days to several months.” Restructure the pages that already have indexation and impressions first, since those are recrawled soonest, and judge the result over a quarter rather than a week. The sequencing, from access through structure to measurement, is in our AI Search Playbook; the glossary entries on Google AI Mode and AI Overviews cover the surfaces themselves.

Frequently asked questions

About the work

Webvello is an AI search optimization and index engineering company founded in 2024, based in Roseville, California. We’ve delivered 50+ projects across 37+ cities, with average client ROI in the 3–5x range on properties where the technical foundation was the binding constraint. Founder George Shvaya came to search from cybersecurity and systems engineering, which is why this article reads Google’s documentation as a retrieval spec rather than as marketing. Our generative engine optimization work starts from exactly this question: which sub-queries can your pages win?

Sources: Google’s AI Mode announcement (May 20, 2025), the Search Central post “Top ways to ensure your content performs well in Google’s AI experiences on Search” (May 21, 2025) and the AI features documentation, all linked above and re-read on October 16, 2026. The fan-out example is an illustration of the mechanism, not published Google data.