What is query fan-out in Google AI Mode, and why one big practice page is not enough

What is query fan-out in Google AI Mode?

Query fan-out is the retrieval method AI Mode uses to answer a question: it decomposes the question into a set of related sub-queries, searches for each one at the same time, and synthesizes the results into a single response. A traditional search returns one ranked list for one query. AI Mode runs many queries behind one answer, so a page can be retrieved for a sub-topic of the question without ever being the page that ranks for the question as typed.

Google announced the technique by name in the AI Mode rollout post published on blog.google on May 20, 2025. Elizabeth Reid, Google's VP and Head of Search, wrote that AI Mode uses the query fan-out technique, breaking a question into subtopics and issuing a multitude of queries simultaneously on the user's behalf. The company has repeated and extended the explanation twice since, in a July 2025 interview and a March 2026 blog post, both cited below.

The mechanism matters more for a law firm than the label does. A person typing a full sentence into AI Mode, rather than a two-word keyword phrase, is handing Google's system several distinct questions bundled into one. The firm that answers more of those bundled questions, on pages the system can retrieve, has a better chance of being cited in the synthesized answer than the firm that wrote one long page and hoped it would rank for the sentence as a whole.

Who explained query fan-out at Google, and when?

Three named Google employees have described query fan-out in public between May 2025 and March 2026, each adding a detail the one before did not give. The timeline is short enough to state directly: Elizabeth Reid introduced the term in May 2025, Robby Stein gave a concrete example in July 2025, and Dounia Berrada extended the same mechanism to image search in March 2026.

Robby Stein, Google's VP of Product for Search, walked through an example in an interview reported by Search Engine Journal on July 30, 2025. Asked how AI Mode handles a broad question, Stein said that a prompt like things to do in Nashville with a group leads the system to think of a bunch of related questions, such as great restaurants, great bars and things to do with kids, and search for each of them at once.

Dounia Berrada, Google's Search Senior Engineering Director, described the same fan-out mechanism applied to a photo rather than text in a company blog post dated March 5, 2026. Berrada said AI Mode is basically doing a dozen searches in the time it takes to do one, using the example of a photo of a garden producing simultaneous searches for the care requirements of every plant in the image. The number is not a formal count of sub-queries; it is Berrada's own shorthand for the scale of the parallel search.

How many sub-queries does a single search actually generate?

Google has not published an exact count, and any specific number in circulation is an outside estimate rather than a company figure. Semrush's own analysis puts the range at roughly 8 to 20 or more background searches for a complex question, with a simple prompt sometimes generating only one. Search Engine Journal separately reported that Google's related Deep Search feature can issue dozens or even hundreds of background queries for a single request.

Treat every specific count, including the ones above, as an illustrative range rather than a target to hit. What is consistent across Google's own statements is the order of magnitude: Stein's Nashville example produced three or four visible sub-topics, and Berrada's dozen-searches description matches that scale. A law firm planning content should read this as evidence that a meaningful question produces somewhere between a handful and a few dozen underlying searches, not as a number to engineer a page count around.

What does fan-out look like for a legal search term?

A single question typed into AI Mode about a legal problem carries several distinct legal sub-questions inside it, each of which the system can search for separately. The table below shows how that plays out for five practice areas, phrased the way a person actually types rather than as a keyword. The sub-queries are illustrative examples of the kind of decomposition Google has described, not a transcript of an actual AI Mode session, since Google does not publish the sub-queries it runs for a given search.

Why does one comprehensive practice page lose to several targeted pages?

A single broad practice page can rank for the practice area's main keyword and still be invisible in an AI Mode answer, because the answer is assembled from whichever pages best address each individual sub-query, not from whichever page best addresses the topic as a whole. A firm with five narrow pages, each written to a specific sub-question, has five separate chances to be the source the system retrieves; a firm with one long page has one.

This is a shift from how SEO worked when a single query returned a single ranked list. Under that model, consolidating a practice area's content onto one authoritative page and building links to it was the standard advice, because the page only had to win one competition. Under fan-out, the same page is competing in several competitions inside one search, and depth on the sub-topic beats breadth across the topic in each one.

Ranking position for the practice area's head term becomes a weaker signal of AI Mode visibility as a result. Ahrefs' March 2026 citation study found that only about 38% of the pages Google's AI Overviews cite rank in the traditional top 10 for the query, down from roughly 76% in mid-2025. A firm that ranks first for "personal injury lawyer [city]" is not guaranteed a citation in an AI Mode answer to a longer, more specific question built from that same search.

Should a law firm split a practice page into several sub-practice pages?

In most cases, yes, once the practice area is large enough to carry more than two or three genuinely distinct fact patterns. The table below compares the two structures on the dimensions that matter for fan-out retrieval. Neither structure is wrong on its own; the choice depends on whether the firm has real expertise and content to support each sub-page, not just a template to fill in.

The caution runs the other direction from the instinct to write more pages. Google's scaled content abuse policy treats content published primarily to manipulate search rankings, including content generated at scale with little added value, as a violation regardless of whether a human or an automated process produced it. A sub-practice page earns its place by answering a fact pattern the firm's attorneys actually handle, in language a client in that situation would use, not by existing to claim another slot in the sitemap.

Does fan-out mean a firm needs new schema markup or an llms.txt file?

No. Google's own developer guidance states plainly that there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode, and that no AI-specific text files or special schema.org markup are required. The precondition is the same as it has always been: the page has to be indexed and eligible to be shown with a standard search snippet before it can be retrieved into a fan-out sub-query at all.

That guidance reframes fan-out as a content-coverage problem rather than a technical-markup problem. The fix is writing the sub-pages that answer the sub-queries, not adding a file or a markup type Google has not asked for. Firm AEO covers the separate, narrower questions of which schema types are worth adding anyway and whether an llms.txt file does anything measurable in their own posts; this pillar is about page structure, not markup.

Why does this matter now, at AI Mode's current scale?

Because AI Mode is no longer a small preview feature. Alphabet's Q2 2026 earnings report, published July 22, 2026, stated that AI Mode had surpassed one billion monthly active users. Google's Gemini app separately reached one billion monthly active users, reported August 11, 2026. Between the two products, fan-out retrieval is now running behind more than a billion monthly searches, not a beta test a firm can reasonably wait out.

A firm does not need every one of those billion users searching for a lawyer to feel the effect. It needs the share of legal searches that already happen inside AI Mode to keep growing, which is the direction every usage figure in this post points. A practice page structure built for one-query, one-ranking search is increasingly answering a question that AI Mode has already split into several, whether the firm has written for that split or not.

How does a firm know whether its sub-practice pages are being retrieved?

Mainly by watching which specific, longer queries start surfacing in Google Search Console for the new pages, and by running the firm's own target questions in AI Mode periodically to see which pages, if any, the answer draws from or links to. Neither method isolates fan-out cleanly, since Google does not report which sub-query a citation came from, but a jump in impressions for a narrow, fact-pattern-specific query after a new sub-page goes live is a reasonable signal that the page is being pulled into the retrieval set.

Firm AEO's separate posts on tracking AI mentions and on Bing's AI Performance report cover the fuller measurement protocol, including run counts and sample sizes. The short version for fan-out specifically: measure the sub-page's own search performance before treating a citation count as proof the page is working, because a page can be retrieved and read without ever being the one named in the answer.