Ask Google's AI Mode something like "things to do in Nashville with a group" and the system doesn't run that search once. According to Robby Stein, Google's VP of Product for Search, the model breaks the question into a batch of related searches: restaurants, bars, and kid-friendly options, then runs them in the background before drafting a reply. That mechanism, known as query fan-out, became the GEO industry's go-to explanation this year for why some pages get cited by AI engines and others don't. An April 2026 study of more than 800,000 scored pages tested the advice built on top of that explanation directly against real citation data. The advice didn't hold up.
What query fan-out actually is
Query fan-out is the process by which an AI search system splits one question into several related sub-queries, runs them in parallel against its index, and merges the results into a single answer. Google uses it inside AI Mode, Deep Search, and some AI Overviews; other engines run comparable multi-query retrieval under different names.
Stein's Nashville example, reported by Search Engine Journal, shows the mechanism at its simplest: one query becomes several. Deep Search takes it further, issuing dozens to hundreds of background searches on a single complex question, a process that can take several minutes to finish. The legal description behind it is a Google patent called Thematic Search, US12158907, filed in December 2024, which lays out a system that identifies themes inside a query, generates narrower sub-queries from those themes, and summarizes the grouped results with a language model.
None of that mechanism is in dispute. What a content team should actually do about it is where the disagreement starts.
The 800,000-page study that tested the "cover everything" advice
An AirOps report with Growth Memo analyst Kevin Indig, published in April 2026, ran 16,851 real queries through ChatGPT three times each, scraped 353,799 cited and uncited pages, and scored how much of each query's fan-out sub-topics every page actually covered. The finding: how much of a query's fan-out a single page covers barely predicts whether it gets cited.
Raw citation rates did creep upward with coverage on their own: pages answering none of a query's sub-topics were cited 30.6% of the time, pages answering 51 to 100% of them were cited 35.2% of the time. But once the researchers controlled for how closely a page matched the primary query, using a relevance score of 0.8 or higher, the pattern reversed. Pages covering 26 to 50% of the sub-queries were cited 38.2% of the time; pages covering all of them dropped to 34.0%. Partial coverage paired with strong relevance beat exhaustive coverage.
Why retrieval rank still beats coverage
Position in the underlying search results predicted citations far more reliably than fan-out coverage did, in the same report. Pages that ranked first behind the scenes were cited 58.4% of the time. Pages ranked tenth were cited just 14.2% of the time. That four-fold gap dwarfed anything coverage scoring produced.
That gap lines up with how these systems retrieve text once a page is already in the running. Passage retrieval, the step where a model matches sub-query text against a specific chunk of a page rather than the page as a whole, rewards a tightly focused section over one trying to cover every angle a query could fan out into. A closer look at how AI engines chunk content before they cite it found the same pattern: narrow, self-contained sections consistently outperform sprawling ones built to catch every possible follow-up question.
What this means for a content strategy built around fan-out
The practical answer isn't to chase every sub-query fan-out can generate. It's to identify the handful that overlap most with what real buyers ask, answer those directly in a well-structured section, and let the strength of a whole topic cluster do the rest of the work a single page can't.
That's also a different kind of keyword planning than classic SEO ever required, one reason GEO and SEO diverge on fundamentals that used to be settled. A separate 2026 comparison of AI citation data found that topical authority, how thoroughly a domain covers a subject across multiple pages, predicted citations better than raw domain rank. Fan-out coverage crammed into one page is a weak substitute for the same signal a genuinely authoritative cluster already sends. Pair that with an answer-first structure, where each section opens with a direct response to one specific question, and a page ends up matching two or three of the sub-queries an engine is likely to generate without having to guess at all of them.
Why fan-out tracking tools don't agree with each other
Tools promising to show "your brand's fan-out queries" are measuring something inherently unstable. Independent testing of the underlying process found that only 27% of the sub-queries generated for a given search stayed consistent when the same search ran again, and 66% of fan-out queries showed up in just one run out of several.
That instability echoes a separate 126-million-prompt study of the gap between AI mentions and AI citations: these systems answer the same question differently from one run to the next far more often than a classic search engine does, which is exactly why a single fan-out snapshot makes a fragile foundation for a content plan. A report listing forty queries a page should supposedly cover is describing one sample of a probabilistic process, not a fixed target.
Fan-out is real. Fan-out optimization, as currently packaged and sold, mostly is not.
Frequently Asked Questions
What is query fan-out in AI search?
Query fan-out is the process where an AI system turns one question into several related sub-queries, runs them in the background, and merges the results into a single answer. Google uses it inside AI Mode and Deep Search. Other AI engines use comparable multi-query retrieval to cover a topic more broadly than a single search would.
Does covering every fan-out sub-query improve citation chances?
No. An April 2026 AirOps study of over 800,000 scored pages found that once relevance to the main query was controlled for, pages covering 26 to 50% of sub-queries were cited more often (38.2%) than pages covering all of them (34.0%). Exhaustive coverage did not outperform focused, highly relevant coverage.
What predicts AI citations better than fan-out coverage?
Retrieval rank. Pages that ranked first in the underlying search results were cited 58.4% of the time versus 14.2% for pages ranked tenth, a four-fold gap. Topical authority across a whole domain and passage-level relevance to the specific question both outweighed how many sub-queries a single page tried to address.
Are fan-out tracking tools reliable?
Only partially. Independent testing found that just 27% of the sub-queries generated for a repeated search stayed consistent across runs, and 66% appeared in only one run. A fan-out report is a snapshot of a probabilistic process, not a fixed list of queries a page needs to target.
What's the Google patent behind query fan-out?
Thematic Search, US12158907, filed in December 2024. It describes a system that identifies themes in a query, generates narrower sub-queries from those themes, and summarizes the grouped results using a large language model, the same mechanism Google's AI Mode now runs at scale.