The mechanics

From “best CRM for small businesses”, a fan-out system internally derives several sub-questions: comparison criteria, pricing models, integration requirements, user experiences. Each sub-query runs its own retrieval process (RAG); the answer is composed from the results of all runs. Google uses the technique in AI Overviews and even more aggressively in AI Mode — where fan-out also feeds the session's follow-up questions.

The visibility consequence: a domain can rank #1 for the main query and still be absent from the AI answer, because the answer is assembled from sub-query results in which it does not appear.

The evidence in real data

Fan-out is not a theory: complete fan-out prompts now appear as “queries” in Search Console data — multi-line, conversational requests with instruction character (“… name specific providers”) that no human types into a search box. They are the visible trace of the internal decomposition machinery and a new analysis window at the same time: they show for which sub-questions your domain is already being retrieved.

Identifying sub-queries

  1. SERP signals: People-Also-Ask, related searches and the source clusters visible in AI Overviews for the main query.
  2. Asking LLMs directly: query models for the decomposition (“which sub-questions would you research for X?”) — triangulated across several models.
  3. GSC query data: mine conversational long-form queries as real fan-out traces.

Content consequence: clusters instead of single pages

Fan-out forces cluster thinking: the topical map plans the sub-query spaces, individual passages answer one sub-question each standalone (passage level), and internal linking makes the cluster machine-readable. A single page, however comprehensive, loses against a clean cluster, because retrieval runs per sub-query prefer the most precise source.

Related terms

Fan-out is the query-processing stage of RAG and shapes AI Overviews and AI Mode. The content response is topical maps and passage optimisation. The full analysis: Fan-out queries.

Key point

The AI does not answer your query — it answers its own

Generative systems decompose queries into sub-queries with separate retrieval each. Visibility goes to whoever covers the sub-question spaces — not to whoever ranks best for the main query.

FAQ on query fan-out

What is query fan-out? ▾

The technique generative search systems use to internally decompose a query into several sub-queries, run separate retrieval for each, and synthesise the answer from all results. Google uses it in AI Overviews and — more aggressively — in AI Mode.

Why is a top ranking for the main query not enough? ▾

Because the answer is composed from sub-query results. If you do not cover the sub-questions (criteria, prices, comparisons, experiences) with precise passages of your own, you are absent from the synthesis — the main-query ranking does not change that.

How do you find a topic's sub-queries? ▾

From three sources: SERP signals (People-Also-Ask, related searches, AI Overview sources), asking LLMs directly for the decomposition across several models, and conversational long-form queries in your own Search Console data as real fan-out traces.

What does fan-out mean for content structure? ▾

Clusters instead of single pages: a topical map plans the sub-question spaces, each passage answers one sub-question standalone, internal linking connects the cluster. Retrieval prefers the most precise source per sub-query — not the longest page.