By
Austin Coker
Query fan-out (also written “query fan out”) is the technique AI search systems use to break a single prompt into multiple related searches, run them at the same time, and build one answer from the results. Google confirmed the name and the mechanism in its May 2025 AI Mode announcement, which describes AI Mode splitting a question into subtopics and issuing many queries simultaneously. Most advice written since then treats every generated sub-query as a new keyword to build a page for. That is where it goes wrong.
This post covers:
Every figure below links to its source, with the sample size and date attached.
Google’s patent application “Search with stateful chat” (US20240289407A1) describes a system that generates variants of a user’s query and runs them against a search backend. It is a patent application, not a confirmed description of how the live product behaves. Treat it as a blueprint for the category, not proof of what AI Mode did this morning. That said, the variant types it describes match what practitioners see in fan-out logs.
Here is each type applied to one example prompt: “best CRM for a 20-person B2B SaaS sales team.”
The implicit and comparative types decide most B2B outcomes. A buyer asking for the “best CRM” never types “annual contract minimum” or “does it support multi-currency invoicing.” Yet those are exactly the sub-queries that determine which vendors survive the shortlist the model builds.
Three things happen in order:
The scale depends on the feature. Google’s I/O 2025 announcement describes Deep Search as the same fan-out technique taken further: it can issue hundreds of searches to build a fully cited report. A standard AI Overview sits at the other end of the range.
The practical consequence is that you are no longer competing for one ranking position. You are competing on how well you cover a cloud of related searches, several of which you will never see in any keyword tool.
More than most people assume, and it is all in official documentation. Google’s Search Central guide, AI features and your website, says three things:
On eligibility, the documentation is unusually plain. To appear in these features, your pages need to be indexed and eligible to show in Search with a snippet, and there are no additional technical requirements. There is no fan-out schema, no special markup, and no separate opt-in.
Google’s product team added detail in July 2025. As reported by Search Engine Journal, Google VP of Product Robby Stein said:
Mike King of iPullRank told Digiday in June 2025 that fan-out targets the “subintents” behind a query rather than the query itself.
Anything beyond that, including the exact number of sub-queries a given prompt triggers, is measurement and inference, not confirmation.
ChatGPT does something functionally similar when it searches the web, even though “query fan-out” is Google’s term. The clearest public measurement is MJ Cachón’s study of brand prompts in ChatGPT. It analyzed 1,797 sub-queries and found that a single run averages fewer than three sub-queries. Running the same prompt four times raises the number of distinct sub-queries to about 10.3.
That gap is the finding that matters: one run tells you very little about which sub-queries a prompt can generate.
Perplexity and Claude also break prompts down and search when connected to the web. Be careful, though, about describing how they do it with the confidence Google’s documentation allows for AI Mode. Neither has published a comparable description of its process, and stating one as fact is how bad GEO advice gets made. For how B2B SaaS buyers move through these assistants, see our B2B SaaS ChatGPT guide.
We’ll run your buyers’ prompts across ChatGPT, Perplexity, and Google AI Mode and show you where you’re cited, where competitors show up instead, and which sub-queries you’re missing.
Two shifts do the damage, and both are measurable.
Citations no longer follow classic rankings. A December 2025 Surfer analysis covered 173,902 URLs across 10,000 keywords, reported by MD Marketing Digital. It found that 67.82% of pages cited in AI Overviews were not in the top 10 for the main query or for any of its subqueries. That fits Google’s own statement that fan-out surfaces a broader set of supporting links. A page can be cited because it answered a sub-query well, without ever ranking for what the user typed.
Visibility becomes probabilistic instead of fixed. Surfer’s fan-out research across 1,600 runs found that only about 27% of fan-out queries were consistent across runs. It also found that 66% appeared only once across 10 runs of the same prompt. The sub-query set is regenerated every time, so there is no stable list to rank for. The right target is how often you appear across many runs, not a position.
For high-value B2B services, this is where deals are quietly won and lost. A recruitment agency can be missing from the main answer and still get cited on the sub-query “average time to fill senior engineering roles,” which is the one the hiring VP actually cares about.
This is the most common piece of bad advice in circulation. Lily Ray has argued directly against playing whack-a-mole with individual fan-out queries. She warns that publishing large volumes of thin long-tail pages risks running into Google’s scaled content abuse policy.
The data makes the strategic case too. If 66% of fan-out queries appear only once across 10 runs, a page built for one of them is a page built for something that may never be generated again. You would be spending production budget on the random output of a sampling process.
The alternative is aggregation:
That is the same content-architecture logic behind our SEO services, applied to AI retrieval instead of a page of ten blue links. It is also why consolidated pages tend to collect citations across several sub-queries at once.
See how we consolidate content around the questions AI search keeps asking, instead of chasing every synthetic query.
You observe them repeatedly rather than looking them up in a search volume database. A workable set of tools:
Method matters more than which tool you pick:
The sub-queries that keep coming back across runs are your real content briefs. The long tail of one-time appearances is noise: record it, then ignore it.
Retrieval rewards pages that still make sense when read in pieces. In priority order:
Answer the question in the first two sentences under each heading. A retrieved passage has to stand on its own, without the paragraph above it.
Use question-form H2s that match how buyers phrase sub-queries. Keep them short and descriptive. “How much does CRM implementation cost for a 20-person team” gets retrieved. Abstract headings do not.
Make every section self-contained. Repeat the subject instead of relying on “it” and “this,” so a passage lifted out of context still makes sense to a model deciding whether to cite it.
Cover comparative and implicit sub-intents explicitly. B2B fan-out constantly generates implicit queries about:
Most vendor content leaves these for the sales call instead of putting them on a page.
Get mentioned on third-party sites. Fan-out pulls from review sites, listicles, forums, and industry publications, not just your domain. Off-site presence is part of the same optimization job. Our AI search (GEO) engagements treat those placements as a core deliverable alongside on-site work.
Keep the technical basics clean. Google’s documentation says eligibility means indexed and snippet-eligible. A noindex tag, a nosnippet directive, or a template that doesn’t render for crawlers quietly removes you from the whole game.
Paid placement inside AI assistants is also arriving. We’ve written about what ChatGPT Sponsored Agents mean for organic citations.
Book a free strategy call. We’ll map your topic’s recurring fan-out queries and show you which pages to build or merge first. No pitch deck required.
Stop measuring positions and start measuring frequency. Build a fixed set of commercial prompts and run each one repeatedly, on a schedule, in the assistants your buyers use. Record three things:
Sample size is the whole method. Per MJ Cachón’s analysis of 1,797 sub-queries, a single ChatGPT run averages fewer than three sub-queries, while four runs of the same prompt yield about 10.3 distinct ones. That makes a one-off check closer to a coin flip than a measurement. Ten runs per prompt is a reasonable minimum before you report anything to leadership.
Pair those prompt runs with Bing Webmaster Tools’ AI search query report and your referral data from assistant traffic, then compare the results against pipeline. For a high-value service business, a small number of visits from decision-stage prompts can matter more than a large volume of awareness traffic, and your measurement should reflect that. Our B2B SaaS SEO guide covers the broader reporting approach.
Austin Coker is the CEO and founder of 95 Projects. He has worked in SEO since 2020, when he built and sold an ecommerce brand that grew mainly through search. He founded 95 Projects in 2022 to help SaaS, ecommerce, and B2B service companies show up wherever their buyers search, from Google to ChatGPT and AI Overviews.
Have questions about working with us? Book a 30-minute strategy call to discuss your goals and see if we’re a good fit.
No-pressure conversation
Clear next steps (if we’re a fit)
Talk directly to an expert
Query fan-out is when an AI search system takes one prompt, generates several related sub-queries from it, runs them at the same time, and writes a single answer from the combined results. Google confirmed in its May 2025 AI Mode announcement that AI Mode uses this technique to break a question into subtopics and issue many queries simultaneously.
It is confirmed. Google’s Search Central documentation on AI features says both AI Overviews and AI Mode may use a query fan-out technique across subtopics and data sources. A Google VP described the mechanism publicly in July 2025. What remains unconfirmed is how many sub-queries a given prompt triggers.
No. Google’s documentation says pages need to be indexed and eligible to appear in Search with a snippet, with no additional technical requirements for AI Overviews or AI Mode. The practical failure points are noindex tags, nosnippet directives, content that doesn’t render for crawlers, and pages that never got crawled.
Generally no. Surfer’s research across 1,600 runs found that 66% of fan-out queries appeared only once across 10 runs of the same prompt, so most sub-queries are unstable targets. Cluster the recurring ones into fewer, deeper pages with self-contained sections. This also avoids thin-page risk under Google’s scaled content abuse policy.
Partly, and only with repeated sampling. FanoutFox, Peec AI, Profound, and Surfer’s Keyword Surfer extension show sub-queries and citations, and Bing Webmaster Tools has an AI search query report. 95 Projects runs this as a monitored set of prompts across ChatGPT, Perplexity, and AI Mode, so citation share is tracked over time rather than spot-checked.
Rankings still drive retrieval, but they are no longer the whole story. A December 2025 Surfer analysis of 173,902 URLs across 10,000 keywords found that 67.82% of pages cited in AI Overviews were not in the top 10 for the main query or any subquery. A page can earn citations by answering a sub-query well, even without a top position.
