Query Fan-Out: How AI Search Decomposes One Question Into Many

Query fan-out is the technique modern AI search systems use to answer questions. Instead of matching one query to one ranked list of pages, an AI engine takes the user's question, decomposes it into a multitude of sub-queries that cover different facets of the intent, retrieves relevant passages for each sub-query in parallel, then synthesizes one answer that cites the strongest sources.

Google named the technique publicly at I/O 2025: "AI Mode uses our query fan-out technique, breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf." The same pattern is used by ChatGPT search, Perplexity, Microsoft Copilot, and Gemini. Deep-research modes fan out further - hundreds of sub-queries across roughly 20 iterations - to assemble multi-page reports.

The mechanism is described in Google's Thematic Search patent (US12158907B1), where sub-queries are called "themes." Independent research from Surfer SEO's December 2025 study of 173,902 URLs and iPullRank's technical breakdown documented eight distinct sub-query variant types: reformulations, related queries, implicit queries, comparative queries, entity-specific queries, personalized queries, recency queries, and disambiguations.

Classic search optimized one page against one query. Whichever page had the strongest backlink profile, on-page signals, and freshness for that exact query won the click. Query fan-out breaks that model. The user never sees the sub-queries; they only see the synthesized answer. The competition has moved underneath, into a layer where each sub-query is decided independently and the best-matching passages get cited.

Surfer SEO's December 2025 study quantified the shift: 68% of pages cited by AI search systems do not appear in the top 10 organic results. Pages that match fan-out sub-queries get a 161% citation lift. The same study found that fan-out query stability is roughly 27%, meaning the same user can ask the same question twice and 73% of the sub-queries change between runs. There is no fixed list of sub-queries to optimize against - what matters is whether your content covers the underlying intent broadly enough that some passage matches whichever sub-queries fire.

For commerce specifically, fan-out is the mechanism behind the most common product visibility complaint we hear from retailers: "my pages rank fine on Google, but ChatGPT and Perplexity recommend the competitor." The pages that win in classic SERPs were optimized for one query. The pages that win in AI search are the ones whose product attributes, schema, and category content satisfy the most fan-out sub-queries. That is rarely the same page.

A buyer asks ChatGPT or Perplexity, "What is the best sunscreen for sensitive skin?" The shopper sees one answer with two or three product recommendations. Underneath, the AI engine has fanned out roughly twelve sub-queries: reef-safe formulation, mineral versus chemical UV filters, SPF rating and broad-spectrum coverage, fragrance-free options, pediatric-safe alternatives, dermatologist recommendations, ingredient lists for irritants, sensitive-skin specific testing, water resistance, application methods, comparable price points, and trusted brand citations.

Each sub-query is run in parallel against the AI engine's index. The engine retrieves the top-matching passages for each, weights them by source authority and freshness, and assembles a final answer that cites the products best supported across the most sub-queries. A retailer whose product page lists only "SPF 50, mineral, broad spectrum" satisfies maybe three of those twelve sub-queries. A retailer whose page also includes ingredient breakdowns, allergen flags, dermatologist-tested certifications, reef-safety certifications, and comparable-product context satisfies eight or nine. That second retailer gets the recommendation.

Critically, none of this depends on the literal phrase "best sunscreen for sensitive skin" appearing on the page. Fan-out optimizes for the underlying intent, not the surface keyword. This is also why classic SERP rank checks miss what AI search is actually rewarding.

The shift from classic SEO to fan-out-aware optimization changes the entire stack of decisions content and catalog teams make:

• Query length. Classic search averaged 3-4 word queries. AI search queries average 70-80 words because users explain context to a chatbot. Optimization needs to anticipate longer, more conversational intent.
• Unit of optimization. Classic SEO optimized at the page level. Fan-out retrieves at the passage level. Each H2 section needs to independently answer a distinct sub-query, with no sequential dependency on prior paragraphs.
• Authority signal. Backlinks and domain authority still matter, but mentions, citations, and entity authority across the open web matter more for whether the AI cites your passage.
• Output format. Classic search returned a list of links. AI search returns one synthesized answer. There is no second-place click. You either get cited or you do not.
• Measurement. Classic SEO measured ranking position. Fan-out optimization measures share of voice across the universe of sub-queries that fire for your category. Tools like Paz.ai's catalog visibility tracking are built for the second model.

Retailers who only measure traditional SERP rank will systematically underestimate how visible (or invisible) they are inside AI search. The Surfer SEO 68% finding - most AI-cited pages are not in the top 10 organic - is the simplest proof that the two systems are now scoring different things.

Three concrete actions move the needle on fan-out visibility:

1. Restructure content for passage-level retrievability. Each H2 section should independently answer one sub-query. Avoid sections that only make sense after reading the whole article. The AI is going to retrieve your passages out of context, so each one needs to stand alone.
2. Enrich product data so every SKU satisfies more sub-queries. A product page with 5 attributes covers a small number of fan-out themes. A product page with 30 covers many more. Add ingredient lists, certifications, use cases, comparable-product context, dimensional data, and any attribute that could be a sub-query in your category. This is the single largest lever for retailers.
3. Measure share of voice across the fan-out, not just rank. Sample the question space your customers actually ask AI engines, simulate the fan-out, and track how often your products and brand appear across the resulting sub-queries. That is what Paz.ai's AI Readiness platform measures.

Retailers who treat fan-out as a content-only problem will plateau. The biggest gains come from the catalog layer: structured data, attribute completeness, schema, and feed alignment with the protocols (ACP, UCP, MCP) that AI retrieval systems actually consume.