What is query fan-out in AI search?

Fan-out is an engineering word: one input spread across many outputs. In AI search, the input is your question and the outputs are the searches the system launches on its own to answer it. For a store, the consequence is concrete: your product page can end up cited in the answer to a question it doesn't fully answer, because it answers one part of it well, and it can be left out even if it ranks for the original question.

How it works

Google defines it in its guide to generative AI features: a set of simultaneous, related queries that the model generates to ask for more information and bring back more results to answer the person's question. It gives an example: for "how to fix a lawn full of weeds", the fan-out queries might be "best weed killers for lawns", "remove weeds without chemicals" and "how to prevent weeds in a lawn".⁠^{[1]Google Search Central. Optimizing your website for generative AI features on Google Search}

Both of Search's AI features use it. AI Overviews and AI Mode can run several related searches across subtopics and data sources to build the answer, and while they generate it their models identify more pages that support it, so they can show a wider and more varied range of links than a classic search.⁠^{[2]Google Search Central. AI features and your website} When it introduced AI Mode in May 2025, Google put it this way: under the hood, AI Mode breaks your question into subtopics and issues a multitude of queries at once on your behalf. Its deep research mode, Deep Search, takes the idea further and can run hundreds of searches to write a cited report.⁠^{[3]Google, The Keyword (Elizabeth Reid, May 20, 2025). AI Mode in Google Search: Updates from Google I/O 2025}

Fan-out queries retrieve pages from the index, and the model writes from them. That second part, retrieving and then generating, is what is called RAG or grounding, and Google does it with its usual ranking systems.⁠^{[1]Google Search Central. Optimizing your website for generative AI features on Google Search} Fan-out decides what gets searched; the rest of the system decides what gets read and what gets cited.

Outside Google

ChatGPT doesn't use the name, but it describes the same move. When it searches through outside providers, it usually rewrites your question into one or more focused queries and, after reviewing the first results, can send more specific ones. Its example: for "what's new in CCR8 drugs for cancer?", the first query might be "CCR8 immunotherapy drug development 2025", and the next a more specific one about a particular drug and a conference. It can add your approximate location and, with memory on, what it remembers about you.⁠^{[4]OpenAI Help Center. Searching the web with ChatGPT} That way "good restaurants near me" can become "top restaurants San Francisco".⁠^{[4]OpenAI Help Center. Searching the web with ChatGPT}

Bing shows site owners a reflection of those queries. Its AI Performance report, in Bing Webmaster Tools, shows the grounding queries: the key phrases the AI used when retrieving the content it ended up citing. It is a sample, not the full record.⁠^{[5]Microsoft, Webmaster Tools blog. Introducing AI Performance in Bing Webmaster Tools Public Preview}

What not to do with this

The temptation is to manufacture a page for every possible sub-query. Google rules it out in the same guide: its systems understand synonyms and the general meaning of what is searched, and creating separate content for each variant, fan-out queries included, mainly to manipulate AI answers, breaks its policy against scaled content abuse.⁠^{[1]Google Search Central. Optimizing your website for generative AI features on Google Search} The useful move is the opposite: make the page you already have answer each part of the question well.

In a store

A fictional, teaching example. The sub-queries in this table are a hypothesis written to explain the idea: they are not real queries from Google or any assistant, none of which publish theirs. The question: "I'm looking for a camera backpack I can take as a carry-on that will hold up in the rain."

A page that only says "ideal for travel" answers none of the four. One that gives the dimensions, the weight and the rain protection can be the source for two or three parts of the answer, even if it never shows up for the whole question.

How to check it

1. Write your customer's full question and break it up yourself into the three or four doubts that make it up. Label the list a hypothesis, because it is one.
2. Find the fact on your product page that answers each doubt, in text and with its unit. If it is missing, add it to the page you already have instead of creating another.⁠^{[1]Google Search Central. Optimizing your website for generative AI features on Google Search}
3. Ask the full question in AI Mode and note which pages it links and which part of the answer each one supports, with the date.⁠^{[2]Google Search Central. AI features and your website}
4. If your site is in Bing Webmaster Tools, open AI Performance and read the grounding queries for your cited pages.⁠^{[5]Microsoft, Webmaster Tools blog. Introducing AI Performance in Bing Webmaster Tools Public Preview}
5. Check the Search Console generative AI performance report for which pages earn impressions in AI Overviews and AI Mode. If a product page shows up that you didn't expect, it probably answers part of some question.⁠^{[6]Search Console Help. Generative AI performance report (Search)}

Related concept: RAG

The two go together and are often confused. Query fan-out is the step of deciding what to search: multiplying one question into several queries.⁠^{[1]Google Search Central. Optimizing your website for generative AI features on Google Search} RAG is the whole scheme of searching and then writing from what was found; fan-out is one way to enrich its first half. A system can use RAG with a single query and no fan-out at all, and fan-out is useless without a model that reads what comes back.

To keep going, the Google AI Mode entry explains where this technique is most visible, and the AI search reading path puts the other terms in order.