Query fan-out explained: how AI search breaks one question into many

Query fan-out is what happens when an AI search system takes one question and runs several related searches behind the scenes before it writes an answer. Google's definition is "a set of concurrent, related queries generated by the model to request more information and fetch additional relevant search results to address the user's query."

It matters to your business because the pages an AI answer links to are not only the pages that match the words the user typed. They can come from any of the related searches. A page that clearly answers one sub-question may be shown for a broad question it might not rank for by itself in classic search.

The practical lesson is simple, and it comes with a warning from Google. Cover the real sub-questions your customers have, on pages that deserve to exist. Do not build a separate page for every query variation you can think of. Google says doing that to manipulate rankings or AI responses violates its spam policy.

What Google says fan-out is

Google has described the technique in several places, in slightly different words. Reading them side by side gives you the full picture without anyone's guesswork added.

Three things come through in these descriptions. The model generates the extra queries itself. The queries run at the same time. And the results are combined into one response. Google's AI features documentation adds a consequence that matters to site owners: while responses are being generated, its models "identify more supporting web pages, allowing us to display a wider and more diverse set of helpful links associated with the response than with a classic web search."

Note the word "may." Google does not say every AI answer uses fan-out, how many queries a typical answer uses, or how it picks among the pages those queries return.

A worked example

Google's own example is a good one. A person searches "how to fix a lawn that's full of weeds." According to Google's guide for site owners, the fan-out queries "might include 'best herbicides for lawns,' 'remove weeds without chemicals,' and 'how to prevent weeds in lawn.'"

Look at what happened there. The user asked one broad how-to question. The system turned it into a product question, an alternative-method question and a prevention question. A lawn care company with a strong page on preventing weeds could be a supporting link for a question about fixing them.

Here is a hypothetical example for a service business, to show the pattern. The sub-questions are our illustration, not Google's output.

You cannot know the exact queries. You can know the questions a thoughtful person would need answered before deciding, and those are the same questions your customers ask you on the phone.

Other AI systems do something similar

"Query fan-out" is Google's term, but the general behavior is documented elsewhere too.

• ChatGPT search. OpenAI's help center says "ChatGPT search typically rewrites your query into one or more targeted queries that it sends" to its search providers, and that after reviewing initial results it "may send additional, more specific queries." Its example: a question about the latest drug development targeting CCR8 for cancer might first become "CCR8 immunotherapy drug development 2025." OpenAI adds that saved memories may be used when rewriting a query, if memory is enabled.
• Perplexity. Perplexity says its Deep Research mode "performs dozens of searches automatically, reads hundreds of sources."
• Microsoft Copilot and Bing. Microsoft's AI Performance report shows "grounding queries," defined as "the key phrases the AI used when retrieving content that was referenced in AI-generated answers." Microsoft notes that "they are not full user questions or prompts."

The shared idea is that the search the machine runs is often not the question the person asked. That is the part worth planning around. The details of how each system selects and weighs results are not published, and we won't pretend otherwise. For background on the general process, see how AI answers work.

Why this changes content planning

Classic keyword planning starts from a phrase and asks which single page should rank for it. Fan-out shifts the unit of planning from the phrase to the topic.

Two cautions keep this honest. First, Google says its AI responses rely on its "core Search ranking systems" to retrieve pages from the Search index, so a page still has to be indexed, and the same ranking systems decide whether it is retrieved for the sub-question it answers. Fan-out is not a back door around ordinary search quality. Second, more sub-questions is not the same as more pages. Many sub-questions belong as sections on a page you already have.

If you already organize content into clusters, this will feel familiar. Our guide to creating topical clusters for GEO covers the structure, and GEO keyword research covers how to gather the questions.

Google's warning about fan-out pages

Once people learned the term, a tactic followed: guess the fan-out queries for a topic and publish a page for each. Google addresses this directly. Its guide says that creating "separate content for every possible variation of how people might search (for example, by focusing on other queries that people have asked, or fan-out queries), doing so primarily to manipulate rankings or generative AI responses in Google Search violates Google's scaled content abuse spam policy."

This has teeth. Google clarified on May 15, 2026 that its spam policies also apply to generative AI responses in Search. Among its listed examples of scaled content abuse is "using generative AI tools or other similar tools to generate many pages without adding value for users."

The test in Google's wording is purpose and value. Ask these questions about any page you plan to add:

• Would a customer be glad this page exists if search engines did not?
• Does it contain something the neighboring pages do not: a different answer, different data, a different audience?
• Could it be a section on an existing page without losing anything?
• Is someone who knows the subject going to write or review it?

If the honest answers are no, no, yes and no, it is a fan-out page in the sense Google warns about. Microsoft takes a similar line: Bing's webmaster guidelines say thin, ad-heavy or affiliate-only URLs may lose ranking eligibility and grounding visibility. We look at the content side in why generic content never gets cited by AI.

How to cover a topic's real sub-questions

1. Start with a decision. Pick one thing customers come to you to decide, such as repair or replace, or which plan to buy. A decision is a better starting point than a keyword.
2. List the questions behind it. Use what you already have: sales calls, support tickets, emails, proposal objections, the questions your team answers every week.
3. Check your own search data. In Search Console's Performance report, look at the queries that already bring impressions to pages on the topic. In Bing Webmaster Tools, look at the grounding queries in the AI Performance report.
4. Map each question to a home. For each one, decide whether it is already answered, belongs as a new section on an existing page, or needs its own page.
5. Give a page only to questions that need one. A sub-question earns its own page when the full answer is long, the audience is different, or people search for it as a topic in its own right.
6. Answer each one specifically. Put the direct answer first, then the conditions, numbers and exceptions from your own experience. Google's top recommendation is "non-commodity content," meaning material a general summary could not supply.
7. Link the pieces together. Internal links are on Google's list of best practices for AI features. Connect the main page to each supporting page and back.
8. Prune overlaps. If two pages answer the same question, merge them. Google's guide lists reducing duplicate content among its recommendations.

What you can and can't measure

You can't see Google's fan-out queries. Google's Generative AI performance report in Search Console is described on its help page with impressions and four tabs: Pages, Countries, Dates and Devices. No queries tab is listed. What you can do is watch which pages gain impressions in that report after you fill a gap. The report and the related settings are covered in our guide to Google AI Mode.

Microsoft gives you more on its side. The AI Performance report in Bing Webmaster Tools lists grounding queries and maps them to the pages that were cited. Microsoft says the data is a sample of citation activity, and it covers only Copilot, Bing's AI summaries and unnamed partner integrations, but it is the closest first-party view of machine-written queries that site owners have. Setup is covered in Bing Webmaster Tools and IndexNow for AI search.

Be careful with tools that claim to reveal the exact fan-out queries Google runs for a keyword. Google's guide states: "No third-party tool has access to our internal ranking or AI systems." A tool can suggest plausible sub-questions, which is an estimate, not a readout. And none of this matters for a page that crawlers can't reach, so check your AI crawler and robots.txt settings first.

Conclusion

Query fan-out helps explain why AI answers can link to pages that never matched the user's exact words: the system may search for the parts of the question, not just the whole. The useful response is to know your customers' real sub-questions and answer each one well, on a page or section that earns its place. The response Google warns against is mass-producing a page per variation. If you would like help finding the gaps in a topic you care about, request a free GEO audit. More guides are in our GEO and AEO Guide.

Frequently asked questions