Query Fan-Out

Query fan-out is the technique in which an AI search system expands a single user query into several related sub-queries, runs them in parallel, and composes its answer from the combined results. It is the main reason a page can be cited in an AI answer without ranking for the query the user actually typed.

Google documents the behaviour directly: "Both AI Overviews and AI Mode may use a 'query fan-out' technique — issuing multiple related searches across subtopics and data sources." It is a form of retrieval-augmented generation in which the retrieval step is plural rather than singular.

The Mechanism

A conventional search matches one query to one ranked list. A fan-out system first interprets the query, then generates the narrower questions a thorough answer would need to cover. A query about choosing a product might fan out into sub-queries on price, alternatives, reliability and specific use cases. Each sub-query returns its own results, and the model draws on the pool.

Ahrefs describes the selection effect this produces: "the initial query is split into multiple related sub-queries. The pages that appear most often within those sub-query SERPs then get cited." The sub-queries themselves are not shown to the user or reported to site owners, so the set of searches a page must surface in is unobserved.

The Evidence: Top-10 Overlap Halved

The clearest sign of fan-out is the falling overlap between cited pages and the organic top 10 for the original query. In July 2025 Ahrefs measured roughly 76% of AI Overview citations coming from the top 10. In March 2026, across 863,000 keyword results pages and 4 million cited URLs, it measured 37.9%, with 31.2% ranking between positions 11 and 100 and 31.0% beyond position 100.

Two caveats matter. First, fan-out is Ahrefs' explanation, not a measured cause: the same article notes that AI Overviews moved to Gemini 3 in January 2026 and that Ahrefs improved its citation parsing between the two studies, so part of the change may be a model change or a methodology change. Second, the studies disagree on the level. BrightEdge's 17% and Ahrefs' 38% use different methods and cannot be averaged. What they agree on is direction: most cited pages now come from outside page one for the head query.

Fan-out also helps explain why different engines rarely cite the same pages. Grossman et al. (SIGIR 2026), using 11,500 real queries, found cross-platform source overlap below 0.2 on a Jaccard scale. Each engine expands the query differently and searches a different index.

What Surfaces Through Fan-Out

Pages reached through sub-queries are not a random sample of the web. In the March 2026 Ahrefs data, YouTube accounted for 18.2% of cited URLs that ranked nowhere in the top 100 for the original query, or 5.6% of all AI Overview citations. Narrow, specific content on large platforms appears to answer sub-questions that a broad page on the head term does not.

Implications

Topical breadth beats single-keyword targeting. If citation depends on appearing across many sub-query result sets, the unit of competition shifts from one page ranking for one term to a site covering the questions around a topic. Ahrefs draws this conclusion from its own data.

Sub-query coverage is a planning exercise. Because the sub-queries are hidden, the workable approach is to enumerate the comparisons, definitions, prerequisites and follow-ups a careful answer would address, and to check that a clearly titled page or section exists for each.

Ranking still matters, one level down. Fan-out does not replace ranking; it multiplies the rankings that count. The C-SEO Bench study (NeurIPS Datasets and Benchmarks 2025) found that most content-rewriting tactics aimed at AI engines were ineffective, while ranking higher in the retrieved context worked better. SAGEO Arena (KDD 2026) found that such rewrites "often harm retrieval" in full pipelines. Being retrieved for the sub-query is the precondition for everything else.

Measurement gets harder. Rank tracking on head terms captures a shrinking share of the path to citation. Whether a brand appears in the answer has to be sampled directly, as covered under AI visibility measurement.