Fan-Out Query

Query fan-out is what happens when an AI search engine like Google AI Mode expands one user query into dozens of synthetic sub-queries, fetches sources for each in parallel, and synthesizes them into a single answer. Knowing which sub-queries fire for a seed term is what AEO turns on.

What Query Fan-Out Is

Traditional search treats a query as a single string to match against an index. AI search engines treat the same query as a starting point.

Google AI Mode, ChatGPT Search, and Perplexity all use the same basic architecture: take the seed query, generate synthetic sub-queries that surround it, run those against the index in parallel, and synthesize a grounded answer from the retrieved sources.

Google describes this directly in its AI Mode announcement, calling it a “query fan-out technique” that issues multiple related searches at once across subtopics and data sources. The engine reads the user’s intent, decomposes it, and runs the expanded set against its corpus before generating the response.

For a treatment center marketer, the practical consequence is large. The single keyword a facility tries to rank for is not the only string the engine searches.

A query like “alcohol rehab in San Diego” can fan out into ten or fifteen distinct sub-queries covering insurance, levels of care, modalities, accreditation, family involvement, and aftercare. Citation depends on whether the facility’s content satisfies those sub-queries, not just the seed.

Why Fan-Out Matters for Treatment Centers

Behavioral health marketing has spent two decades optimizing for individual keywords. The whole industry has been organized around it: rank-tracker reports, keyword volume buckets, location-page templates targeting “rehab in [city].”

That model assumed the user typed one query and saw a list of links. Both halves of that assumption are breaking.

The user side broke first. Treatment seekers and their families no longer type three-word queries. They ask AI tools full questions: “what’s the difference between PHP and IOP for someone who works full-time,” or “is there a rehab near me that takes Anthem and treats co-occurring depression.”

The engines treat those questions as fan-out targets, not as keyword strings.

The engine side broke next. Google AI Mode does not match the query to a page; it matches the synthetic sub-query set to a set of pages, then synthesizes.

A facility that ranks #1 for “alcohol rehab San Diego” but has no content on insurance verification, family programming, or dual diagnosis may not appear in the AI answer at all. The page wins the keyword and loses the citation.

For operators, this means the unit of optimization has moved up a level. The work is no longer to rank a page; it is to cover the sub-query set well enough that an AI engine pulling sources for any of them lands on the facility’s domain.

How AI Engines Generate Sub-Queries

Fan-out is not random. The engines decompose the seed query using a small number of recurring expansion patterns. Understanding those patterns is what makes content planning predictable instead of guesswork.

Intent Expansion

The engine asks what the user actually wants behind the query. “Best rehab near me” is rarely a request for a ranked list.

The intent fans out into: “what does best mean for rehab,” “what should I look for,” “how do I evaluate a treatment center,” “what red flags should I watch for.” Each becomes its own retrieval call.

Entity Expansion

The engine pulls in related entities. For “alcohol rehab,” that includes detox, medical stabilization, MAT, naltrexone, residential treatment, PHP, IOP, aftercare, and 12-step versus non-12-step models. For “treatment for teens,” it adds adolescent levels of care, family therapy, school re-entry, and parental insurance coverage.

Comparison Expansion

The engine generates head-to-head sub-queries. Rehab A versus rehab B. Inpatient versus outpatient. PHP versus IOP. Insurance-covered versus private-pay. Comparison expansion is where directories and review aggregators tend to win citation because their pages are explicitly structured around it.

Location Expansion

For any local-intent query, the engine fans out across city, neighborhood, regional, and insurance-market layers. “Rehab in San Diego” expands into La Jolla, Pacific Beach, North County, Mission Valley, and into the insurance plans active in that market.

A facility whose location content is thin on any of those layers loses ground in the local fan-out.

Example: Fan-Out for “Alcohol Rehab in San Diego”

Running the fan-out on a representative behavioral health seed makes the pattern concrete. Below is a realistic sub-query set an AI engine would generate for “alcohol rehab in San Diego,” grouped by expansion type.

• Intent expansion: “what should I look for in an alcohol rehab,” “how do I know if I need rehab or outpatient,” “what does a good first call to a rehab look like.”
• Entity expansion: “alcohol detox San Diego,” “medical detox versus social detox,” “naltrexone and acamprosate for alcohol use disorder,” “PHP for alcohol,” “IOP for alcohol,” “dual diagnosis treatment San Diego.”
• Comparison expansion: “inpatient versus outpatient alcohol rehab,” “30-day versus 60-day programs,” “luxury rehab versus standard residential,” “insurance-covered versus private-pay rehab.”
• Location expansion: “alcohol rehab in La Jolla,” “alcohol rehab in Pacific Beach,” “alcohol rehab in North County San Diego,” “rehabs that take Anthem in San Diego,” “Tricare-covered rehab San Diego.”

Fifteen sub-queries from one seed, and that is a conservative count. The facility’s content cluster either covers them or it does not. There is no middle ground in citation eligibility.

How to Optimize Content for Fan-Out

Fan-out optimization is a cluster problem first and a page problem second. The work runs in a specific order.

Run the fan-out before drafting. For each priority seed query, generate the actual sub-query set the engine would produce. Webserv uses an internal /fanout skill for this; the same exercise can be run manually by prompting AI Mode and ChatGPT Search and recording the questions they ask back.

Map sub-queries to pages. Some sub-queries belong on the hub (the broad service or location page). Most belong on sibling pages in the cluster.

A page that tries to answer everything in the fan-out answers none of it well enough to be cited. Topical authority at the cluster level is what carries the citations.

Structure each page for extraction. The page that answers a given sub-query should answer it directly in the first 40-60 words, with clear H2s, FAQ blocks where appropriate, and schema. Semantic triples embedded in the prose give the engine clean subject-predicate-object facts to lift.

Bind the cluster with internal links. Hub-to-sibling, sibling-to-sibling, and sibling-back-to-hub links signal a coherent body of work on the topic.

This is where the AEO citation lift compounds, because the engine treats the cluster as a single source of topical relevance rather than a set of isolated pages.

Fan-Out vs Traditional Keyword Targeting

Traditional keyword targeting starts from a list of high-volume terms and builds one page per term. Fan-out optimization starts from a seed query, generates the sub-query set the engine would actually fire, and builds a cluster of pages designed to satisfy that set. The two approaches produce different content libraries.

A keyword-targeted library is wide and shallow: hundreds of pages, each chasing a specific phrase, each thinly connected to the others.

A fan-out-optimized library is narrower and deeper: fewer pages, each tightly bound to a cluster, each answering a defined slice of a sub-query set. The second pattern is what wins AI Overview citation.

Mike King at iPullRank documents this shift in his analysis of query fan-out, latent intent, and source aggregation, where he shows how AI Mode’s retrieval architecture rewards source aggregation rather than keyword targeting.

The implication for treatment center content strategy is direct: the page is no longer the unit of optimization. The cluster is.

Tools and Techniques for Fan-Out Analysis

Three approaches work in practice. Each has a place depending on how much depth a facility needs and how fast the analysis has to ship.

Manual Prompting Against AI Mode and ChatGPT Search

Ask the seed query directly in AI Mode and in ChatGPT Search. Capture the follow-up questions the engine surfaces, the sources it cites, and the subtopics it groups. This is the slowest method but the most accurate, because the output is what the live engine is actually doing.

Programmatic Fan-Out Generation

Webserv’s internal fan-out tooling generates the synthetic sub-query set for a seed without a live engine call. The output is a structured list grouped by expansion type (intent, entity, comparison, location), ready to map against an existing content cluster to find coverage gaps.

SERP Inspection for Sub-Query Surface Area

Inspect the actual AI Overview and AI Mode SERP for the seed query. The “People also ask” block, the AI Overview citation list, and the related queries panel all leak sub-queries the engine considered.

This is the cheapest method and a good ongoing monitoring loop after the initial fan-out is mapped.

Building Content That Survives the Fan-Out

Fan-out is the architecture of AI search, and it has already changed what good content looks like for treatment centers.

Webserv’s SEO practice runs fan-out analysis as the first step of every cluster build, so the content library is shaped against the sub-query set the engines actually fire rather than the keyword list a facility used to track.

Frequently Asked Questions

What is a query fan-out?

A query fan-out is the expansion of a single user query into many synthetic sub-queries that an AI search engine runs in parallel before generating its answer. Google AI Mode, ChatGPT Search, and Perplexity all use the pattern. The engine decomposes the seed across intent, entity, comparison, and location dimensions, retrieves sources for each, and synthesizes a single response.

How does query fan-out affect SEO for treatment centers?

Traditional SEO targets a single keyword per page. Fan-out moves the unit of optimization from page to cluster. A facility ranking for “alcohol rehab in [city]” but thin on insurance, levels of care, modalities, or family involvement may not appear in the AI answer at all, because the engine pulls sources for the full sub-query set.

How many sub-queries does an AI engine generate?

The count varies by seed complexity and engine. For a typical behavioral health local query like “alcohol rehab in San Diego,” the fan-out commonly produces ten to twenty sub-queries spanning intent, entity, comparison, and location expansions. Higher-complexity queries (dual diagnosis, adolescent care, insurance-specific searches) can produce thirty or more sub-queries in a single fan-out cycle.

How do I find the sub-queries for my seed term?

Three methods work in combination. Run the seed in Google AI Mode and ChatGPT Search and record the follow-up questions and cited sources. Use a programmatic fan-out tool to generate the synthetic sub-query set offline. Inspect the live SERP for “People also ask,” AI Overview citations, and related queries. The three together produce the sub-query map you optimize against.

Is query fan-out the same as semantic search?

No. Semantic search is the underlying retrieval method that matches meaning rather than exact keywords. Query fan-out is the orchestration layer on top: the engine generates multiple semantic searches from one seed query, runs them in parallel, and aggregates the results. Fan-out is what makes the retrieval set wide; semantic search is what makes each individual retrieval call accurate.