Glossary

Query fan-out

Updated

Query fan-out is a technique in which an AI search engine splits one question into several related search queries, runs them, and writes one answer from the combined results. Google uses the term for AI Mode and AI Overviews, and other AI search tools, such as ChatGPT search, also turn one prompt into one or more searches. A page therefore competes for the queries the engine writes, not only for the words the person typed.

Query fan-out diagram The question "best CRM for a small agency?" splits into five searches: best CRM for small agencies, agency CRM pricing per seat, CRM reviews for agencies, CRM with client reporting, agency CRM alternatives compared. The results of all five merge into one answer that cites a review site, a pricing page and a comparison. ONE QUESTION best CRM for a small agency? SEARCHES THE ENGINE WRITES best CRM for small agencies agency CRM pricing per seat CRM reviews for agencies CRM with client reporting agency CRM alternatives compared ONE ANSWER, WITH CITATIONS review site pricing page comparison
An illustrative fan-out. The engine writes several searches for one question, reads the results of each, and cites the pages that answered a part.

How query fan-out works, step by step

The engine does not search for the question as typed. It plans the searches it needs first. The steps below describe the general pattern. Each engine has its own details, and most do not publish them.

  • Read the question. The engine works out what the person wants and which parts the question has.
  • Write the sub-queries. The engine writes several search queries, one for each part or angle. These queries often add words the person never typed, such as "pricing", "reviews", "vs" or a use case.
  • Run the searches. The engine sends the queries to a search index, often at the same time. Some engines then run follow-up searches based on what the first results show.
  • Read and select. The engine reads the top results for each query and picks the passages that answer each part.
  • Write one answer. The engine merges those passages into one response and links the pages it used as sources.

A worked example

Take the question "what is the best CRM for a small agency". The answer names a few products and cites a few pages. A review site can be cited for the reviews part. A vendor pricing page can be cited for the cost part. A comparison article can be cited for the comparison part. None of these pages needs to rank for the original question.

The example is illustrative. Engines do not publish their exact queries for a given question, so the list below shows the kind of searches an engine can write.

  • "best CRM for small agencies": the question itself, rewritten as a search.
  • "agency CRM pricing per seat": the cost.
  • "CRM reviews for agencies": what users say.
  • "CRM with client reporting": a feature that agencies need.
  • "agency CRM alternatives compared": comparisons between named products.

Which engines use query fan-out

Google uses the name in its own documentation. The other engines describe a similar pattern in their own words. None of these companies publishes how many queries it writes for a given question or how it chooses them.

  • Google AI Mode. At Google I/O in May 2025, Google said that AI Mode uses "our query fan-out technique". It breaks a question into subtopics and issues a multitude of queries at the same time. Google said its Deep Search feature takes the same technique further and can issue hundreds of searches.
  • Google AI Overviews. Google's AI features and your website page says that both AI Overviews and AI Mode may use query fan-out. The page says the technique runs related searches across subtopics and data sources, so the response can show a wider and more diverse set of links than a classic search.
  • ChatGPT search. OpenAI's help article on ChatGPT search says ChatGPT rewrites a prompt into one or more targeted queries for its search partners. It can send more specific follow-up queries after it reads the first results.
  • Gemini. Google's Gemini API documentation says that with Google Search grounding, the model decides whether a search helps, generates one or more search queries, runs them, and returns the queries it ran.
  • Perplexity. Perplexity's getting started guide says Pro Search breaks a question into smaller steps and pulls from a broader range of sources.

Why query fan-out matters for SEO and GEO

In classic search, a page competes for the words a person types. With query fan-out, a page competes for the queries the engine writes. Those queries are more specific than the question, and there are more of them.

Google says there are no extra requirements and no special optimizations needed to appear in AI Overviews or AI Mode. A page must be indexed and eligible to show in Google Search with a snippet. Query fan-out does not change that. It changes which questions your pages need to answer, in three ways.

  • Cited pages need not rank for the question. A page that answers one sub-query well can be cited in an answer to a broader question that the page never mentions.
  • Narrow pages gain value. A clear pricing page, a fair comparison or a page for one use case matches the specific queries that engines write.
  • Rank tracking misses part of the picture. A keyword report shows where a page ranks for the question. It does not show the sub-queries that decided which pages the answer cites. See AI citation for how citations differ from mentions.

How to see the fan-out queries

Some engines expose the searches they run, and some do not. There are four ways to see them or to infer them.

  • API responses. The Gemini API returns the queries it ran when Google Search grounding is on. OpenAI's web search tool reports each search action, and its documentation says the action usually includes the queries.
  • The chat interface. Some interfaces show the searches while the answer is written, or list them next to the answer. Others show only the final sources.
  • Cited sources. Where the queries stay hidden, the cited pages are the next best evidence. A cited pricing page suggests a pricing query. A cited review site suggests a reviews query.
  • Manual tests. Ask the engine your question and read the sources. Then search Google for the likely sub-queries and compare the pages that rank with the pages the answer cites.

Fan-out queries in CitePrism

CitePrism samples seven engines on a schedule: ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Google AI Mode and Microsoft Copilot. Where an engine exposes its searches, CitePrism records them for each prompt on its Queries page. Where it does not, the cited sources stand in for them.

How to write for query fan-out

Write for the questions behind the question. None of these steps needs special markup.

  • List the sub-questions. For each buyer question, write down its parts: cost, reviews, comparisons, use cases, setup and alternatives. Use the fan-out queries you can see as the starting list.
  • Answer each part in its own passage. Put the answer in the first sentence under a heading that matches the sub-query. An engine that reads one passage needs that passage to make sense on its own.
  • Publish the narrow pages. A public pricing page, one page for each main use case and a fair comparison page each match a common type of sub-query.
  • Say who the product is for. State your category, audience and main use case in one plain sentence. Engines often add an audience or a use case to the queries they write.
  • Keep facts current. Prices, plan limits and feature lists go stale. Show the date a page was last checked, and update the facts when they change.
  • Earn presence on third-party pages. Many sub-queries return review sites, communities and comparison articles. A brand that is listed there can be named even when its own site is not cited. Citation share shows which of those sites the engines rely on.

Query fan-out vs query rewriting and query expansion

Search engines have changed queries for a long time. Query rewriting fixes or rephrases one query, for example with a spelling correction. Query expansion adds related terms, such as synonyms, to one query so that it matches more documents. Both still produce one search and one ranked list of links.

Query fan-out produces many searches with different intents. Each search returns its own results. The engine then reads across all of them and writes one answer. The output is not a ranked list of links. It is a generated response with a few citations.

FAQ

What does query fan-out mean?

Query fan-out means that an AI search engine splits one question into several related searches, runs them, and writes one answer from the combined results. Google uses the term to describe how AI Mode and AI Overviews find the information for an answer.

Does ChatGPT use query fan-out?

ChatGPT search works in a similar way. OpenAI says ChatGPT rewrites a prompt into one or more targeted search queries for its search partners, and it can send more specific follow-up queries after it reads the first results.

Do Google AI Overviews use query fan-out?

They can. Google Search Central says both AI Overviews and AI Mode may use a query fan-out technique. Google describes AI Mode in more detail: it breaks a question into subtopics and issues many queries at the same time.

How do I find the fan-out queries for my topic?

Use an engine that reports its searches. The Gemini API returns the queries it ran when Google Search grounding is on, and the OpenAI web search tool usually reports its queries too. For engines that hide them, read the cited sources and infer the queries from them. CitePrism records the reported queries for each prompt over time.

How many queries does query fan-out create?

It depends on the engine and the question, and no engine publishes a fixed number. Google says AI Mode issues a multitude of queries for one question and that Deep Search can issue hundreds of searches. OpenAI and the Gemini API both say one or more, so a simple question can need only one search.

Read more

Measure it for your brand

CitePrism computes every metric in this glossary from scheduled samples across seven engines, with the raw answers one click away. Free to start with your own keys.