Glossary

What is query fan-out?

The Brimm team · grounded in the docs

Query fan-out is the technique behind Google's AI Mode and AI Overviews where 1 question is decomposed into many related sub-queries, each searched in parallel, and the results synthesized into a single answer. Google describes this in Google's AI optimization guide. The short version: “one question quietly becomes many searches behind the scenes.” So you are not competing for one query, you are competing for the cluster of sub-questions the system generates.

One question becomes a search of searches

When you type a question into a person-facing search box, you get one set of results for that exact phrase. AI Mode does not work that way. It reads your question, infers the smaller questions hidden inside it, and runs a search for each of them at the same time. Then it reads across all of those result sets and writes one synthesized answer that stitches the pieces together.

Take a question like "what is the best CRM for a small law firm." A traditional search returns pages targeting that phrase. Fan-out instead spins off a handful of sub-questions: which CRMs serve legal practices, what features a small firm actually needs, how the pricing compares, what other lawyers say, whether there is a free tier. Each of those is searched on its own, and your page only enters the final answer if it is the best source for one of those sub-questions.

How fan-out works, step by step

The mechanics are consistent across the AI surfaces that use it:

Decompose. The model breaks the original question into related sub-queries, including follow-ups a curious person would ask next.
Search in parallel. Each sub-query runs as its own retrieval against the index, so the system pulls back many candidate pages, not one ranked list.
Read and extract. The system reads the retrieved pages and pulls the specific passages that answer each sub-query.
Synthesize. Those extracted passages are merged into one answer, often citing several different pages for different parts of it.

The consequence is that a single AI answer can be assembled from five or ten different sources, each contributing the one passage it answered best. There is no longer a single "position one" to win.

Why fan-out changes the math for your site

Under classic ranking, you picked a target query and tried to be the top result for it. Under fan-out, the question you can see is only the doorway. Behind it, the system is generating sub-questions you never see, and each is a separate chance to be cited or to be left out.

This is closely tied to two other shifts. One is the great decoupling, where impressions in AI answers rise while clicks to your site fall, because the answer is delivered on the results page. The other is passage-level retrieval: the system cites a passage, not a whole page, so the unit of competition shrinks from "your article" to "this paragraph." Fan-out multiplies the number of those paragraph-sized contests a single page can enter.

If your page answers only the headline question and ignores the obvious follow-ups, it can win the doorway and lose every room behind it. A page that also answers the sub-questions cleanly can be pulled into several of the fanned-out searches at once.

What to do about it

You cannot see the exact sub-queries Google generates, but you can predict most of them, because they are the questions a real reader would ask next. The work is to cover the topic completely and make each answer easy to lift out:

Cover the topic, not just the keyword. Answer the obvious follow-up questions on the same page, with their own clear headings that match how a person would phrase them.
Make each sub-answer a clean, extractable passage. Lead each section with a self-contained answer in plain text, so a retrieval system can pull it without dragging in noise. See how to make your content extractable for AI.
Be specific. Include the number, the named entity, the concrete detail. Vague paragraphs lose to specific ones in every sub-query.
Stay readable without JavaScript. If the retrieval crawler cannot see your passage in the raw HTML, none of the above counts.

This is the same discipline behind generative engine optimization: write the page so the machine can take the exact part it needs. Fan-out just raises the reward for doing it across a whole topic instead of one query.

Check your own page

You can map the likely sub-questions by hand, or you can paste your link into our GEO audit and see how extractable your passages are, whether your answer survives without JavaScript, and where a page covers the headline question but skips the follow-ups. Run it in the app and we print the gaps in fix order.