Query Fan-Out: How One Search Becomes a Dozen (and Decides Who Gets Cited)

Type one question into Google AI Mode and the system quietly runs a dozen. That is query fan-out. Instead of matching your search to a single ranked list of pages, the model reads your query as an intent, generates a set of narrower sub-queries around that intent, retrieves sources for each one in parallel, and stitches the results into a single answer. You typed one thing. The system searched for many.

This matters for anyone trying to get cited by AI search, because fan-out changes the unit of competition. You are no longer competing for one query. You are competing for a whole cluster of sub-queries you never see, generated on the fly, each one deciding a small slice of whether your brand shows up in the final answer.

What query fan-out actually is

Traditional search works on a simple loop. A user types a query, the engine matches that query against an index, and it returns ranked results for that exact string. One query in, one set of results out.

Query fan-out breaks that loop apart. When an AI system receives a query, especially a complex or open-ended one, it does not treat the query as a single string to match. It treats the query as a question that has several component parts, and it generates separate sub-queries for each part. Each sub-query gets its own retrieval pass. The system pulls sources for sub-query one, sources for sub-query two, sources for sub-query three, and so on, then a synthesis layer reads everything it retrieved and writes one coherent answer, choosing which sources to cite along the way.

Suppose someone asks an AI system, "should I switch my HVAC company to a service agreement model?" A single-query search engine matches that string against pages containing similar language. A fan-out system might instead generate sub-queries like "what is a service agreement in HVAC," "pros and cons of HVAC membership plans," "average cost of an HVAC service agreement," and "how do service agreements affect technician retention." Four sub-queries, four separate retrievals, one final answer that draws from whichever sources answered each piece best.

Why AI search needed fan-out in the first place

Traditional keyword search assumes the user already knows how to ask the question well. Real questions rarely arrive that clean. People type vague, compound, or comparative questions that don't map to one clear intent. "Best CRM for a small agency" is really three or four questions bundled together: what counts as best, what counts as small, what a CRM needs to do for an agency specifically, and how price factors in.

A single retrieval pass against a query like that returns a muddy mix of results, because no single page perfectly matches every implied sub-question. Fan-out solves this by doing the decomposition work the user didn't do themselves. It separates the bundled question into its component parts, retrieves cleanly for each part, and lets the synthesis step recombine everything into an answer that actually addresses the full intent.

This is also why fan-out shows up specifically in AI Mode style products rather than classic search. Generating a full written answer from scratch requires more raw material than returning ten blue links. A generated answer needs to cover multiple angles convincingly, and one retrieval pass rarely surfaces enough varied, high quality source material to do that. Fan-out is the mechanism that gathers enough raw material, from enough different angles, to write a complete answer instead of a shallow one.

Google AI Mode and the fan-out technique specifically

Google has been the most public about building fan-out directly into its AI Mode product. Rather than treating AI Mode as a chat wrapper sitting on top of the regular search index, Google designed it to actively decompose harder questions before running any retrieval at all. The system looks at the query, decides how many angles it has, generates a query for each angle, and searches the index once per generated query.

This explains a pattern many site owners notice without fully understanding why: a page can rank respectably for its target keyword in classic search and still never appear in an AI Mode answer for that same topic. The page was never competing against the keyword in AI Mode. It was competing against a set of sub-queries the system invented from the keyword, and it may simply not have answered any of them directly enough to win a citation.

It also explains why AI Mode answers on complex topics tend to cite more sources than a typical search snippet ever would. Each cited source usually corresponds to a different sub-query, not a different opinion on the same question. A four-source answer often means four sub-queries ran, and four different pages each won one slice.

A worked example: watching fan-out run on one real question

Abstract descriptions of fan-out are harder to act on than a concrete walkthrough. Take a question a home services business might care about: "is it worth switching customers to a maintenance plan instead of one-off repair calls?"

A fan-out system reading that question sees at least four distinct facets buried inside it. It generates a sub-query asking what a maintenance plan actually includes, because the answer depends on the reader understanding the product first. It generates a sub-query comparing maintenance plans against one-off repair billing, because the word "instead" signals a comparison. It generates a sub-query around the cost and revenue impact of the switch, because business decisions usually come down to numbers. And it generates a sub-query around customer retention, because "worth it" points to a business outcome rather than a product description.

Each of those four sub-queries runs its own retrieval. The definitional sub-query might pull from a page that explains maintenance plans clearly in its first paragraph. The comparison sub-query might pull from a page with an actual side-by-side table of billing models. The cost sub-query might pull from a page with real numbers instead of vague ranges. The retention sub-query might pull from a page that specifically ties maintenance plans to repeat business.

Four sources, four different pages, each winning a narrow slice of the final answer. None of them needed to cover all four facets to get cited. But a single page that covered all four facets well would have had four separate chances to win a citation slot, instead of one.

This is the pattern worth internalizing: fan-out does not reward the page that talks about maintenance plans the most. It rewards whichever page answers each specific facet the most directly, and it is entirely possible for four different competitors to split the citations on a single question, each one winning ground the others left uncovered.

How the fan-out process actually runs

Strip away the marketing language and the process breaks into four steps.

Step 1: intent analysis

The system reads the original query and identifies what it is actually asking. Is it a definition question, a comparison, a how-to, a recommendation, or some combination? This step decides how aggressively the query needs to be broken apart. A narrow factual query like "what year was the iPhone released" needs no fan-out at all. A broad, comparative, or research-style query gets decomposed heavily.

Step 2: sub-query generation

Based on the intent analysis, the system generates a set of related sub-queries. These cover the different facets of the original question: definitions, comparisons, pricing, process steps, exceptions, related questions a person would naturally ask next. This is the step people mean when they talk about "fan-out," because it is the moment one query becomes several.

Step 3: parallel retrieval

Each sub-query runs its own retrieval against the index, independently of the others. This is why the process is called fan-out rather than a simple chain. The sub-queries execute in parallel, not one after another, and each one returns its own set of candidate sources.

Step 4: synthesis and citation selection

The system reads across everything the parallel retrievals returned, identifies which sources most directly and clearly answered each sub-query, and writes a combined response. Sources that answered their sub-query cleanly get cited. Sources that were retrieved but did not clearly answer the specific facet they were pulled for get dropped from the final synthesis, even if they were technically relevant to the broader topic.

That last point is the one worth sitting with. Being retrieved is not the same as being cited. A page can show up in the candidate pool for a sub-query and still lose the citation to a page that answered that specific sub-question more directly. Our piece on how AI engines choose sources goes deeper on what separates a retrieved source from a cited one.

It also helps to understand what the synthesis step is actually optimizing for. It is not trying to represent every candidate source. It is trying to produce the shortest, most useful answer it can defend with citations. That means a page competing for a sub-query citation is competing against the field and against something else entirely. It is competing against the system's own bias toward brevity. A source that says the same thing as three others, only with more hedging and more words, is the one most likely to get left out of the final synthesis even if it was retrieved.

The kinds of sub-queries fan-out tends to generate

Fan-out is not random. It tends to generate a handful of recognizable sub-query types, and knowing them helps explain why comprehensive content outperforms narrow content under this system.

Definitional sub-queries. What is the core concept the user is asking about? If someone asks about switching to a service agreement model, a definitional sub-query establishes what a service agreement actually is before the system can address whether switching makes sense.

Comparative sub-queries. How does the option in question stack up against alternatives? Comparison language in the original query, or implied by it, triggers sub-queries that pull pages structured around comparisons, tables, and pros-and-cons breakdowns.

Quantitative sub-queries. What does this cost, how long does it take, how common is it? These sub-queries favor pages with specific numbers over pages with vague reassurance.

Process sub-queries. How does someone actually do the thing being asked about? These favor step-by-step content over descriptive prose.

Follow-up sub-queries. What would a person naturally ask next, once the main question is answered? These are the sub-queries that reward content anticipating a reader's next question rather than stopping cold at the first one.

Not every query triggers all five types. A narrow factual question might only generate one or two sub-queries. A broad, decision-oriented question, the kind that starts with "should I" or "is it worth," tends to trigger the full set, because deciding requires understanding the concept, weighing alternatives, checking the cost, and knowing what happens next.

Why this favors a small number of deep pages over many shallow ones

A common instinct in content teams is to publish a separate page for every angle. A page for the definition. A page for the comparison. A page for the pricing. This made sense when a single-query search engine rewarded exact keyword matches, because each page could target its own keyword.

Fan-out changes the math. When a system generates five sub-queries from one broad topic, it is not searching five different keywords in five different content silos. It is searching the same index five times in quick succession, looking for the best answer to each facet. A single comprehensive page that answers all five facets clearly, each in its own section, can win multiple citation slots inside the same answer. Five separate thin pages competing against each other dilute the signal instead of concentrating it, and none of them individually covers enough ground to feel authoritative on its own.

The pages that consistently win multiple fan-out citations tend to share a structure: a direct answer near the top, then dedicated sections for each major facet, each one written so it could stand alone as a complete answer to its own sub-question. That structure is not a stylistic preference. It is what lets one page compete across an entire cluster of generated sub-queries instead of just one.

This table shows how the two approaches tend to perform once fan-out is running the retrieval.

Query fan-out versus a single-query search

The difference isn't cosmetic. It changes what "ranking well" even means.

In single-query search, your job is to rank for the exact phrase or its close variants. Keyword matching, on-page optimization, and backlinks all point at winning one query. In a fan-out system, your job is to be the best answer to any one of several sub-queries a system might generate from a broader question, none of which you can predict in advance or target directly.

This shifts the competitive unit from the page to the topic. A page optimized for one keyword can win a single-query search and still lose in a fan-out system, because it only answers one facet of a multi-part question. A page that thoroughly covers a topic, its definition, its comparisons, its costs, its process, and its exceptions, has more surface area to win across a wider set of generated sub-queries, even without targeting any of them by name.

Who actually gets cited when a query fans out

Citation in a fan-out system comes down to whichever source answers a given sub-query most directly, not whichever source ranks highest overall for the broad topic. That means a smaller, less authoritative site can still win a citation for one sub-query slice if its content answers that specific facet more clearly than a bigger competitor's page does.

This is good news for smaller brands and bad news for anyone assuming broad domain authority alone will carry them through. Fan-out evaluates content facet by facet. A comprehensive competitor with a weak pricing section can still lose the pricing sub-query citation to a smaller site that answers pricing precisely and with real numbers.

What this means for how you build content

If a single page can win or lose citations facet by facet, the practical response is to make sure every likely facet of your topic has a clear, direct answer somewhere on the page, not scattered across five thinner posts that each cover one angle shallowly.

Start by mapping the facets a topic actually contains. Take your core topic and list the definitional question, the comparison question, the cost question, the process question, and the follow-up question a reader would ask next. Then check whether your existing content actually answers each one directly, in its own clearly marked section, rather than implying the answer somewhere in the middle of a paragraph about something else.

This is exactly the audit our AEO content checklist walks through, and it is worth running before you write anything new. Reformatting and filling gaps in a page you already have often produces more fan-out citations than publishing another new post on the same topic.

Common mistakes that get pages cut from the fan-out synthesis

The most common mistake is writing content that answers only the headline question and treats every sub-facet as an afterthought. A guide titled "should you switch to a service agreement model" that spends four paragraphs on definitions and one vague sentence on cost is going to lose the cost sub-query to a competitor with a dedicated pricing breakdown, even if the rest of the guide is stronger.

The second common mistake is burying a direct answer inside a long introduction. Fan-out retrieval favors content that states a clear answer near the top of a section, not content that builds up to the answer through several paragraphs of throat-clearing context.

The third mistake is treating comparison questions as an SEO afterthought instead of a real content type. If your topic naturally invites comparison, whether that's product versus product, method versus method, or before versus after, and you have no page structured around that comparison, you are handing that sub-query citation to whoever does.

The fourth mistake is writing every section at the same shallow depth so nothing wins outright. A page that gives four paragraphs of equal, general treatment to definition, comparison, cost, and process ends up mediocre across the board instead of decisive on any one facet. Fan-out retrieval favors the page that answers a specific sub-question best, not the page that mentions every sub-question adequately. Pick the facets your topic actually needs and answer each one like it is the only question being asked.

The fifth mistake is ignoring the follow-up facet entirely. Many pages answer the question in the title and stop. But a reader who just learned what a maintenance plan is, or what a service agreement costs, almost always has a next question already forming. Content that answers that unasked follow-up in its own section is positioned to win a sub-query citation that a competitor's otherwise stronger page never even attempts to address.

Tracking whether fan-out is actually citing you

Because fan-out sub-queries are generated on the fly and never shown to you directly, you cannot look them up the way you'd check keyword rankings. What you can do is run your own target queries against AI Mode and similar systems regularly, read the full generated answer, and note which sources it cited and for which apparent facet of the question. Over a handful of queries in your topic area, a pattern usually emerges: which of your pages win citations, which facets you consistently lose, and which competitor keeps showing up in the gap.

This kind of manual query tracking is slow at scale but reveals exactly what abstract advice cannot: whether your comparison section is actually the one getting cited, or whether a competitor's table is winning that slot every time. Treat the pattern as a to-do list. If you consistently lose the pricing facet, that's the section to rebuild first, not the whole page.

How query fan-out fits into the bigger AEO picture

Fan-out is one mechanism inside a much larger shift in how people find information, and it doesn't replace the fundamentals of getting your content recognized as authoritative and citable in the first place. Schema markup, entity signals, and structured formatting all still matter, because they help retrieval systems understand and trust your content before it ever reaches the synthesis step. If you haven't yet built that foundation, our complete guide to AEO covers the full picture, from technical readiness through content strategy.

What fan-out adds on top of that foundation is a reason to think in topics rather than keywords. A well-built AEO foundation gets your content into the retrieval pool. Comprehensive, facet-complete content is what wins the individual sub-query citations once you're there.

A practical way to think about your next piece of content

Before publishing, ask what the five or six sub-questions are that a fan-out system would likely generate from your target topic, and check whether your page answers each one directly enough to win that specific slice. If a competitor's page would answer any one of those sub-questions more clearly than yours does, that is the gap to close first, not a reason to write an entirely new page.

A simple exercise makes this concrete. Write your target topic at the top of a blank page. Under it, list every question a genuinely curious, uninformed reader would ask before making a decision related to that topic. Most topics produce somewhere between four and seven real questions once you push past the obvious first one. Check your existing page against that list, section by section. Wherever a question has no dedicated section with a direct answer near the top, you have found a sub-query you are currently losing by default, not because a competitor beat you, but because you never entered the competition for it.

Fan-out did not make keyword research irrelevant. It made topic completeness the thing keyword research was always supposed to produce in the first place. The brands that treat their content as a complete answer to a topic, rather than a page built around a single phrase, are the ones positioned to win as more of search shifts toward systems that ask several questions before writing one answer.