On this page · 11 sections
- What query fan-out actually is
- How fan-out works, step by step
- The evidence: fan-out is live and large
- Why single-keyword pages lose ground
- What wins: topic clusters, passages, and entities
- How to build for query fan-out
- Measuring in a fan-out world
- India-specific considerations
- FAQ
- How eCorpIT can help
- References
Summary. Query fan-out is how Google's AI Mode answers a question. Instead of matching one query to one page, it uses a large language model to generate a set of related sub-queries, runs them across Google's indexes at once, and has Gemini synthesise a single answer with links. Google's vice president of product for Search, Robby Stein, described AI Mode issuing background searches a user never typed, and said these AI search experiences now serve about 1.5 billion people a month, drawing on real-time systems like a shopping catalogue of 50 billion products updated 2 billion times an hour. The technique runs in AI Mode, Deep Search, and some AI Overviews, and it closely matches a Google "thematic search" patent from December 2024 that groups results by inferred topic rather than keyword. The consequence for strategy is direct: a page optimised for one keyword can satisfy one sub-query, while a connected cluster of pages can satisfy many. With AI answers already cutting the top result's click-through rate by 58% (Ahrefs, December 2025) on a search-ad business the UK regulator valued at £10 to £20 billion in 2024, content architects have to build for topics, not terms. This guide explains the mechanism and the fix.
What query fan-out actually is
Query fan-out is the step where an AI search engine turns one prompt into many searches. Rather than looking for a single best page, the system reads the intent behind the question, writes several related sub-queries, retrieves passages from multiple sources for each, and assembles one synthesised answer. Google AI Mode does it, and so do ChatGPT Search and Perplexity.
Robby Stein gave a plain example in a July 2025 interview reported by Search Engine Journal: "If you're asking a question like things to do in Nashville with a group, it may think of a bunch of questions like great restaurants, great bars, things to do if you have kids, and it'll start Googling basically." The user typed one question. The system asked four or five. Stein described AI Mode using Google Search as a backend tool, issuing multiple queries and combining the results into a single response with links.
The important shift is what the engine is matching against. A traditional ranking looks for the page that best fits the exact query. Fan-out looks for the passages that best answer each inferred sub-question, then stitches them together. Your page is no longer competing to be the one blue link; it is competing to supply one or more of the parts that build the answer.
How fan-out works, step by step
The pipeline is consistent across AI Mode's answers. First, a language model interprets the prompt and its intent. Second, it generates synthetic sub-queries, including angles the user never spelled out. Third, those sub-queries fan out simultaneously across Google's systems, from the web index to structured sources. Fourth, Gemini aggregates the retrieved passages. Fifth, it writes one answer and attaches links to the sources it leaned on.
Stein was specific about the structured sources feeding step three: "We've integrated most of the real-time information systems that are within Google. So it can make Google Finance calls, for instance, flight data, movie information. There's 50 billion products in the shopping catalog, updated I think 2 billion times every hour or so." For a shopping or finance query, fan-out is not just web pages; it is live data pulled through internal tools and summarised.
When Google decides a question needs more reasoning, a mode called Deep Search can issue dozens or even hundreds of background queries and take minutes to finish. Same idea, larger fan.
The evidence: fan-out is live and large
This is not a thought experiment. Stein said AI-powered search experiences, including query fan-out, now serve roughly 1.5 billion users each month across text and multimodal input, and he called Google Search "the largest AI product in the world." The technique is active in AI Mode, Deep Search, and some AI Overview results.
It also has a paper trail. Search Engine Journal matched Stein's description to a Google patent from December 2024 on "thematic search," which creates sub-queries based on inferred themes, groups results by topic, and generates summaries with a language model, with each theme able to link to source pages. The patent organises content around inferred topics rather than specific keywords. That single design choice is why the old one-page-per-keyword model is losing force.
The stakes ride on the same numbers that make AI answers matter at all. Ahrefs found the top-ranking page loses about 58% of its click-through rate when an AI Overview appears, and the Pew Research Center found users click a normal result in only 8% of searches that show an AI summary, against 15% without one. Fan-out is the engine behind those answers.
Why single-keyword pages lose ground
Picture a page built to rank for one phrase, "best running shoes." Under classic search it could win the click. Under fan-out, the engine breaks that intent into sub-queries: cushioning for long runs, shoes for flat feet, best value under a budget, durability over months. A single page pitched at the head term answers maybe one of those well and gestures at the rest. It supplies one ingredient to an answer that needs six.
Two mechanics compound the problem. AI retrieval works at the passage level, not the page level, so the engine lifts the specific paragraph that answers a sub-query, wherever it sits. And entity coverage feeds Google's knowledge graph: sites that cover a subject thoroughly, with accurate references to the people, products, and concepts involved, get mapped as authorities on the topic. A lone page is thin on both counts. It has few passages to offer and a weak entity footprint.
| Dimension | Single-keyword page | Topic cluster |
|---|---|---|
| Maps to fan-out | Answers one sub-query | Answers many sub-queries at once |
| Passage selection | Few extractable passages | Many discrete, answerable passages |
| Entity signals | Shallow, one angle | Deep, many related entities |
| Resilience | Rises and falls with one term | Survives when sub-queries shift |
| Best for | A narrow, stable query | A researched, nuanced topic |
What wins: topic clusters, passages, and entities
The structure that matches fan-out is the topic cluster: a pillar page on the broad subject, linked to a set of specific pages that each handle one sub-topic in depth. A site with deep, interconnected coverage of a subject contributes more to a synthesised answer than a site with one strong standalone page, because it can supply passages for several of the sub-queries at once. Ten thoroughly developed pages on distinct sub-topics build stronger entity associations than fifty thin posts circling the same idea; the quality of the semantic signal matters more than the count of documents.
The unit of optimisation drops from the page to the passage. AI systems select content in discrete blocks, so each section should answer one question in one to three paragraphs, under a clear H2 or H3, and stand on its own if lifted. Write the sub-question as the heading, answer it plainly in the first two sentences, then add the supporting detail. That is the same shape that wins featured snippets, which is why answer-engine optimisation and fan-era SEO point the same way. Our guide to how AEO, GEO and SEO differ covers the overlap.
| Old keyword SEO | Fan-out-era SEO | Why it changed |
|---|---|---|
| One page per head keyword | A cluster per topic | Fan-out asks many sub-queries, not one |
| Optimise the whole page | Optimise each passage | Retrieval is passage-level, not page-level |
| Keyword density | Entity coverage | The knowledge graph maps topics, not terms |
| Rank position is the goal | Inclusion in the answer is the goal | The answer, not the link, gets the attention |
| Standalone hero article | Pillar plus linked sub-pages | Interconnected coverage feeds synthesis |
How to build for query fan-out
Start from the question, not the keyword. Take a core topic and map the sub-queries a real user would need answered, the way AI Mode would fan them out; tools that simulate fan-out can seed the list, but your own knowledge of the topic is the better source. Group those sub-queries into a pillar and its cluster pages, and give each cluster page one clear job.
Then write for extraction. Lead each section with the sub-question as a heading and answer it in the first lines. Name entities precisely and consistently, the products, standards, companies, and people involved, so Google can map your coverage into its knowledge graph. Link the cluster together and back to the pillar, with anchor text that describes the target, so both readers and crawlers can see the topic as one connected body. Keep the facts current and sourced, because a synthesised answer favours specifics it can trust. The tactics that earn citations here are the same ones in our note on why a #1 ranking no longer wins clicks and our 2026 SEO guide.
| Step | Do this | Why it works |
|---|---|---|
| Map sub-queries | List the questions a topic really raises | These mirror the sub-queries fan-out issues |
| Design the cluster | One pillar, several focused sub-pages | Lets the site answer many sub-queries at once |
| Structure passages | One question per section, answered up top | Matches passage-level retrieval |
| Name entities | Reference products, people, standards precisely | Builds the knowledge-graph associations |
| Interlink | Connect sub-pages and pillar with clear anchors | Signals one connected body of coverage |
Measuring in a fan-out world
Fan-out breaks measurement before it breaks ranking. When the engine invents its own sub-queries, the boundary of what counts as "a query" blurs, and attribution gets harder: a user may see your passage inside an answer built from a question you cannot see in any keyword report. The practical response is to widen what you track. Watch Google Search Console's AI-impression reporting to see which pages appear in AI answers, treat topic-level visibility rather than single-term rank as the headline metric, and accept that being included in the synthesised context is the win, even when the click does not follow. Rank for the term if you can, but architect for the topic.
India-specific considerations
For Indian SEO teams, fan-out lands the same way it does anywhere, because AI Mode and AI Overviews reason over intent, not language borders. Content built as a tight topic cluster with clean entity references travels across English and Indian-language queries and across engines, from Google AI Mode to ChatGPT and Perplexity. The gap is often structural rather than linguistic: many Indian sites still publish long single pages targeting head terms, which supply one passage to a fan-out answer instead of many. Rebuilding those into pillars with focused sub-pages is low-cost, since it is editorial and architectural work rather than new licence spend, and it compounds as the knowledge graph maps the coverage. Where a topic touches personal data, design the content and any tools around it to India's Digital Personal Data Protection Act 2023 (DPDP) from the start.
FAQ
How eCorpIT can help
eCorpIT helps SEO strategists and content architects rebuild keyword-era sites for AI search. Our senior engineering and content teams map the sub-queries a topic fans out into, design pillar-and-cluster architectures, restructure pages for passage-level retrieval, and strengthen the entity and internal-link signals that feed the knowledge graph. To turn a set of single-keyword pages into topics that AI search cites, talk to eCorpIT.
References
Last updated: July 7, 2026.
