What is query fan-out? How AI search turns one question into many

Query fan-out is the technique AI search engines use to answer one question with many searches. The model splits the question into subtopics, runs a related search for each at the same time, and writes a single answer from what comes back. Google built AI Mode on it and says AI Overviews may use it too, while ChatGPT, Perplexity and Claude run their own versions. For a site owner, it means a page can be cited for a sub-question it never targeted and passed over for the head query it ranks for.

The fix is smaller than most advice suggests. You don't need a page per sub-query. You need the page you already have to answer the next four or five questions a careful reader would ask.

What is query fan-out?

Query fan-out is Google's name for letting the model write several related searches from one question and run them together. Google's guide to optimizing for generative AI search defines it as "a set of concurrent, related queries generated by the model". Its example turns a question about fixing a lawn full of weeds into searches for the best lawn herbicides, removing weeds without chemicals and preventing weeds in the first place.

The Search Central page on AI features says both AI Overviews and AI Mode may run related searches across subtopics and data sources to build a response, and that the models find more supporting pages while the answer is being written. That is why an AI answer links a wider spread of sites than a classic results page.

How AI Mode query fan-out works

AI Mode breaks the question into subtopics, searches each at the same time and writes one answer, with links, from the combined results. Google has described the technique in three announcements, and each one widened it.

Part of the answer to a product question can come from that shopping data instead of a web page, so keep your Merchant Center feed as accurate as the product page.

AI Overviews may fan out too, but Google says the two features may use different models and techniques, so the same question can link different pages in each. Our post on Google AI Mode covers how the mode picks and displays sources.

Google does not publish the sub-queries. Search Console's Generative AI performance report, open to all sites since August 31, 2026, counts impressions by page, country, device and date. It has no query dimension. You can see that Google used a page in an AI answer, never which fan-out search pulled it in.

Query fan-out example: one question, six searches

Here is how one question could fan out. It is our illustration, not a logged Google run, modeled on the device comparison Google used to launch AI Mode. Each search is labeled with the gap it fills.

Question:        smart ring vs smartwatch vs sleep mat for tracking sleep

related          how accurate are smart rings for sleep stages
implicit         track sleep without wearing anything to bed
comparative      Oura Ring vs Apple Watch sleep tracking
recent           best sleep trackers 2026
entity           Withings Sleep Analyzer price and features
reformulation    which sleep tracker is most accurate

A smart ring review that answers the question as asked has a fair shot at the first and last lines. The other four go to pages that explain how a mat tracks sleep without a wearable, compare two named products, carry a current date or state a price. Each fan out query is its own retrieval. Your page has to come back for that search and then hold a passage that answers it.

How ChatGPT and other AI engines rewrite queries

Every AI engine with web search rewrites the question before it searches. They differ in how much they publish and in whether the searches run at once or in rounds.

ChatGPT's follow-up searches depend on its first results, so a product name it finds on one page can become its next search. How ChatGPT search works walks through OpenAI's own example step by step.

Anthropic's documentation puts a number on the spread. Simple factual questions usually take one to three searches, while comparative research covering several products or companies can take ten or more. Comparisons fan out the widest, so they get the most room below.

Query fan-out SEO: what changes for your content

Fan-out moves the competition from the question to its parts, and the page that answers the parts wins. Ranking first for the head term no longer covers you. In a study of 863,000 keyword results published March 2, 2026, Ahrefs found that about 38% of pages cited in AI Overviews ranked in the top 10 for the query, down from about 76% in its July 2025 study. Ahrefs also improved its citation parsing in between and offers fan-out as a possible explanation, not a proven one.

Related and implicit questions

Write down what someone asks right after the question, and what they need but didn't type. For the sleep tracker question that's accuracy, battery life and whether the person wants to wear anything in bed at all. Give each a heading in the words people search with and a first sentence that answers it.

Comparisons

A comparison sub-query is answered only when both sides are on the page with the criteria that separate them. "Smart rings are more comfortable" is half an answer. A table of price, battery life, subscription cost and what each device measures is a whole one. If you sell one of the products, name the competitor anyway. A page that never mentions the alternative gives the "vs" search nothing to find.

Recent information

Searches with a year, "latest" or "new" want facts tied to a date. Put the date next to the fact, as in "prices checked October 2026" above a price table, and change the facts when they change. A fresh date stamp on old numbers answers nothing.

Named products, brands and places

If the question names or implies a product, a company or a place, it gets its own searches. State model names, prices, specs and locations in plain text, not only in images or a PDF.

For the writing itself, how to write passages AI can quote covers answer-first sections and sentences that stand on their own.

How to find the sub-queries your page misses

You can't see Google's real sub-queries, so rebuild a likely list from what the engines show you, then check the page against it.

1. Pick one query and the one page that should answer it. Fan-out gaps exist per page, not per site.
2. Run the query in AI Mode and read the answer as an outline. Each subheading, comparison and group of bullets is roughly one sub-query, and the links beside it show which page won that part.
3. Note the People Also Ask questions and autocomplete suggestions for the query in regular Google. Our Question Finder collects the autocomplete questions for a topic, groups them by intent and marks which ones your page already answers.
4. Fill the gaps by angle, so the list has at least one related, implicit, comparative, recent and named-entity search.
5. For each search, find the section that answers it in its first two sentences. Mark it answered, partial or missing. Partial means the page mentions the subject without the fact the search wants.
6. Add the missing answers that belong on this page. Send the rest to the page that should own them and link to it.

Treat the list as an estimate. Google's guide says no third-party tool has access to its ranking or AI systems, and that includes ours.

What not to do about query fan-out

Most bad fan-out advice means more pages or smaller pieces, and Google's AI optimization guide warns against both.

One thin page per sub-query. Google says that making separate content for every variation of how people search, fan-out queries included, mainly to manipulate rankings or AI answers violates its scaled content abuse policy. Twenty near-identical pages on sleep tracker accuracy also leave you twenty pages to keep current.

Content chopped into fragments. The same guide says there's no need to break content into tiny pieces for AI to understand it. A section that answers in two sentences and then explains is the right shape. A page of 40 one-line FAQ answers is not.

Search strings pasted in as headings. "best sleep trackers 2026" as an H2 reads like a keyword list. Write the question a person would ask, such as "Which sleep tracker is most accurate in 2026?", and answer it.

Answers you can't stand behind. A comparison with a product you never used, or a 2026 figure you didn't check, hands the engine a wrong passage to quote. Cover the sub-queries you can answer accurately and link out for the rest.

Check which fan-out searches your page answers

The Fan-out Coverage Checker takes a query and the URL of the page meant to answer it. It asks ChatGPT the query with web search on and records up to eight of the searches ChatGPT ran. A model then predicts the related, implicit, comparative, recent, named-entity and reworded searches an engine would likely add, for up to 20 in all. Another model pass reads your page's sections, marks each search answered, partial or missing by meaning, and names the section that answers it. The score is answered searches plus half the partial ones, as a percentage.

The report also says whether ChatGPT cited your site and which domains it cited instead. The searches come from ChatGPT and a model, not from Google, and the checker reads the HTML your server sends, so text that appears only after JavaScript runs isn't checked. Answers vary between runs, so treat each run as a sample. A run costs 95 credits.