Query Fan-Out Explained for SEO and AI Search

What is query fan out? Query fan-out is the technique AI search systems use to split one question into several related sub-searches, run them in parallel, and combine the results into a single answer. Google uses query fan-out in AI Mode and AI Overviews, and ChatGPT, Perplexity and Grok do something similar when they search the web. For SEO, it means a page can be cited for a sub-search it ranks for, even if it does not rank for the original question.

What is query fan out according to Google

Google introduced the term publicly when it launched AI Mode as a Labs experiment in March 2025, and described it in more detail at Google I/O on 20 May 2025. In the I/O announcement, Elizabeth Reid, Google’s VP and Head of Search, wrote that AI Mode uses a “query fan-out technique” that breaks a question into subtopics and issues many searches at the same time, so it can explore the web more deeply than a single search would. Deep Search applies the same idea at a larger scale, running hundreds of searches to build a fully cited report. The Google I/O 2025 post remains the clearest official description.

Google’s Search Central documentation says both AI Overviews and AI Mode may use query fan-out, which is why they can show a wider and different set of links than the classic results. Because the two features can also run on different models, their sources differ: Ahrefs found that AI Mode and AI Overviews shared only 13.7% of cited URLs for the same queries.

How google query fan-out works step by step

Google has not published the full pipeline, but its own descriptions, its patents and practitioners’ analysis point to roughly these stages:

1. Understand the question. The system classifies the intent and decides whether a search is needed at all. In Google’s Gemini API, for instance, the model first decides whether searching would improve the answer.
2. Generate sub-queries. The model writes several related searches: rephrasings, narrower subtopics, comparisons, and follow-up questions the user did not ask yet.
3. Retrieve in parallel. Each sub-query runs against the search index and returns its own set of pages.
4. Select passages. From all retrieved pages, the system picks the passages most relevant to each part of the answer.
5. Synthesize and cite. The model writes one answer and links to the sources it relied on.

Take a prompt such as “best running shoes for flat feet on a budget”. A fan-out could plausibly include searches like “running shoes for overpronation”, “stability running shoes under $100”, “flat feet running shoe reviews 2026” and “motion control vs stability shoes”. These examples are illustrative; Google does not show the sub-queries it runs in AI Mode.

What iPullRank’s research adds

Mike King of iPullRank has done much of the public analysis of fan-out. His breakdown of Google’s “Search with stateful chat” patent describes synthetic query reformulations, query classification that routes to specialized models, and retrieval at the level of passages rather than whole pages. He also built Qforia, a query fan-out simulator that suggests the sub-queries a prompt might expand into.

In iPullRank’s AI Mode webinar recap (July 2025), King explains that fan-out is why pages cited in AI Mode often do not rank in the top 100 for the original query. He compares visibility to a raffle: every relevant passage that matches a sub-query is another ticket. The practical shift is from optimizing whole pages for one keyword to writing clear, specific passages that each answer a distinct sub-question.

Ahrefs’ data points the same way. In a study published in March 2026, only 37.9% of AI Overview citations came from pages in the top 10 for the query, down from about 76% in July 2025, and 31% came from beyond the top 100. Ahrefs’ hypothesis, which it states is not proven, is that fan-out sub-queries now drive citation more than the original query does.

Query fan-out beyond Google

Other engines fan out too, at very different rates. Peec AI, an AI search analytics company, measured about 5 million fan-outs in April 2026 and found an average of 1.4 sub-searches per prompt for Perplexity, 2.1 for ChatGPT and 6.8 for Grok.

Peec also tracked how ChatGPT’s fan-outs change. Across more than 20 million fan-outs from October 2025 to January 2026, the average sub-query length doubled from about 6 to about 12 words, while the number per prompt stayed flat at about 2.3 to 2.8 across five countries. Longer sub-queries are more specific, which favors pages that answer specific questions.

Language matters as well. In a separate Peec dataset, reported by Search Engine Journal in February 2026, 43% of ChatGPT’s fan-out steps for non-English prompts ran in English, and 78% of non-English runs included at least one English sub-search. For brands outside English-speaking markets, that means English-language sources can shape answers to local-language prompts. Peec is a vendor and its data comes from customer-defined prompts, so treat the exact figures as directional.

What query fan-out seo means for your content

Fan-out changes what “ranking” means in AI search. Five implications follow:

• Topic coverage beats a single keyword. A page that covers the main question and its natural follow-ups (cost, alternatives, comparisons, risks) can match several sub-queries.
• Passages matter more than pages. Each section should answer one sub-question clearly and early, so it can be selected on its own. Our guide to how AI search engines choose sources covers the selection step.
• Classic SEO still feeds the system. Sub-queries run against a search index, so you still need to rank, just for more queries than before.
• Do not build a page per sub-query. Google’s 2026 AI optimization guide warns that creating separate pages for every fan-out variation can violate its scaled content abuse policy. Cover sub-questions within strong pages instead.
• Fan-out is unstable. The model writes new sub-queries on each run, so chasing one specific sub-query is a losing game. Recurring themes across many runs are what to optimize for.

For Google specifically, our guide on appearing in Google AI Overviews puts these points into a checklist.

How to map fan-out queries for a topic

You cannot see every sub-search an engine runs, but you can build a solid approximation. Here is a practical method:

1. List 20 to 30 prompts for the topic. Use real buyer questions, not keywords.
2. Collect observed sub-queries where engines expose them. The Gemini API with Google Search grounding returns the search queries the model ran alongside its answer. Some developer tools and browser extensions can surface ChatGPT’s search queries; these rely on undocumented behavior that can change at any time.
3. Simulate the rest. Tools such as Qforia generate likely sub-queries for Google’s AI features. Treat the output as hypotheses, not data.
4. Run each prompt several times. Sub-queries differ between runs, so three to five runs per prompt show which themes recur.
5. Cluster the sub-queries into themes. Group by entity, attribute, comparison and use case. Count how often each theme appears.
6. Check your coverage. For each recurring theme, note the page and passage that answers it, whether that page ranks for the theme, and which competitors are cited instead.

A simple mapping sheet looks like this (the rows are a hypothetical example, not measured data):

How to track the effect of fan-out on your visibility

Because fan-out varies from run to run, measure outcomes rather than sub-queries: how often your brand is mentioned and your URLs are cited across repeated runs of the parent prompts. Our guide to AI visibility metrics defines the numbers to use. Asheard, for example, tracks mentions and citations for the same prompts across ten AI engines, including Google AI Mode and AI Overviews; see the AI visibility tracking page. Pair that with classic rank tracking for your theme queries, since sub-queries still run against search indexes.

Frequently asked questions

What is query fan out in Google AI Mode?

Query fan-out is the technique Google AI Mode uses to break a question into subtopics and run many related searches at the same time. It then combines what it finds into one answer with links. Google described it at I/O in May 2025, and says AI Overviews may use it too.

Can I see the fan-out queries Google runs?

Not in AI Mode or AI Overviews; Google does not display them. The Gemini API with Google Search grounding does return the queries the model ran, which gives a useful approximation. Simulators such as Qforia can suggest likely sub-queries, but their output is a hypothesis rather than observed data.

Should I create a page for every fan-out query?

No. Google’s 2026 AI optimization guide warns that creating separate pages for every fan-out variation can violate its scaled content abuse policy. Fan-out queries also change from run to run. Cover the recurring sub-questions as clear sections within strong pages instead of producing thin pages for each variation.

Does query fan-out make rankings irrelevant?

No. Sub-queries run against search indexes, so ranking still decides which pages are retrieved. What changes is that you need to rank for the sub-questions around a topic, not just the head term. That is why pages outside the top 10 for the original query can still be cited.