Query Fan-Out

TL;DR — Query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics. is how AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. turns one question into several searches. Instead of running a single search, the system breaks your question into a handful of related sub-questions, searches all of them at once, and stitches the results into one answer. It’s why a Google AI OverviewAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. can cover a topic from several angles — and why your content needs to answer more than just the exact words someone typed.

What query fan-out is

Google describes query fan-out in AI Mode as issuing multiple related searches across subtopics and data sources. Evidence for this claim Google describes AI Mode as using a query fan-out technique that issues multiple related searches across subtopics and data sources. Scope: Google AI Mode's documented behavior; exact orchestration, ranking, and generated queries are product-specific and can change. Confidence: high · Verified: 2026-07-14 Google: AI Mode in Search That documented product behavior should not be generalized to every model or search system. Evidence for this claim Query fan-out is a retrieval-orchestration pattern that expands one request into multiple related searches before synthesis. Scope: Editorial abstraction from Google's product description, not a claim that every AI answer uses this pipeline. Confidence: medium · Verified: 2026-07-14 Google: AI Mode in Search

When you type a question into a regular search box, the engine runs that query and shows you a list of pages. AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. works differently. Behind the scenes, it takes your one question and fans it out into multiple related searches — one for each angle or sub-topic your question implies — runs them all, and then writes a single answer from everything it pulled back.

Google confirms this happens in its own documentation: both AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics and data sources — to develop a response.”

A quick example. Ask “how do I get my lawn green again?” A regular search runs that one phrase. A fan-out might instead search for the best fertilizer timing, how to fix bare patches, whether to water in the morning, and how to deal with weeds — then combine the answers. You asked one question; the AI quietly asked several.

Why it matters to you

Here’s the practical part. Because the AI is searching for the implied questions, not just the one you typed, your content has a better chance of being used if it genuinely covers a topic in depth — enough that it answers some of those hidden sub-questions on its own.

That’s a shift from the old game of ranking one page for one keyword. A single page stuffed around a single phrase can still rank in normal search, but fan-out rewards content that actually covers the sub-topics a real person would care about.

The other thing to know: this all runs over Google’s normal search indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.. There’s no separate “AI index” you submit to. The same fundamentals that have always mattered — being crawlable, getting indexed, writing clearly — are what decide whether your page can be retrieved for any of those sub-queries.

Want the real mechanics — how fan-out fits inside the retrieve-then-generate pipeline, what “8–12 sub-queries” really means, and why citations are getting less rank-dependent over time? Switch to the Advanced tab.

TL;DR — Query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics. is the retrieval step inside RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights./groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it.: the model decomposes a single query into multiple related sub-queries, runs them concurrently across the core Search indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed., and synthesizes the retrieved passages into one answer. Google confirms both AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics and data sources.” There’s no separate AI index and no separate AI ranking system — the same crawl → index → snippet-eligible gates apply. The SEO consequence is that content has to be retrievable for the implied sub-questions, not just the head query. Google’s own account is that fan-out maturing is why direct top-10 sourcing fell from roughly 76% (mid-2025) to about 38% (early 2026). The “8–12 sub-queries” figure is industry estimation, not an official Google number.

Where fan-out sits in the pipeline

Fan-out is a retrieval strategy within documented products, not a publisher-controlled ranking factor. Evidence for this claim Google describes AI Mode as using a query fan-out technique that issues multiple related searches across subtopics and data sources. Scope: Google AI Mode's documented behavior; exact orchestration, ranking, and generated queries are product-specific and can change. Confidence: high · Verified: 2026-07-14 Google: AI Mode in Search Exact generated queries, weights, and orchestration remain system-specific. Evidence for this claim Query fan-out is a retrieval-orchestration pattern that expands one request into multiple related searches before synthesis. Scope: Editorial abstraction from Google's product description, not a claim that every AI answer uses this pipeline. Confidence: medium · Verified: 2026-07-14 Google: AI Mode in Search

Query fan-out isn’t the whole system — it’s one step inside it. AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode are powered by retrieval-augmented generation (RAG), which Google calls groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it.: retrieve relevant passages from an index, inject them into the model’s context, and generate an answer grounded in them. Fan-out is what happens at the start of the retrieval phase — the decomposition of one user query into many.

The full loop, roughly: user query → fan-out into sub-queries → retrieve across the index → assemble the best passages → generate a synthesized, cited answer. If you want the retrieval mechanics in depth — chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking., embeddingsEmbeddings are dense numerical vectors — lists of floating-point numbers — that represent the meaning of text in a high-dimensional space. Semantically similar content lands close together, so search and AI systems can match by meaning, not just keywords., vector searchVector search finds content by comparing the meaning of a query against stored content as numerical vectors, retrieving the closest ones in a high-dimensional embedding space. At scale it uses approximate nearest neighbor (ANN) algorithms — not exact comparison — to search billions of vectors in milliseconds., re-rankingReranking is the second stage of a retrieval pipeline: after a cheap, broad first pass pulls a candidate set of documents or passages, a slower, more precise model re-scores and reorders that shortlist by true relevance before the results are served or handed to an LLM. — that’s the province of the retrieval and grounding topics in this cluster. This page is specifically about the fan-out step.

What Google has actually confirmed (and what it hasn’t)

Be careful to separate the confirmed mechanism from the industry theory built on top of it. What Google states plainly:

• Both AI Overviews and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics and data sources — to develop a response.”
• The model issues “concurrent, related queries… to request more information and fetch additional relevant search results,” which Gemini then “analyzes… extracts the most relevant facts, and stitches them together into a single, comprehensive answer.”
• It all runs over the core index: “our generative AI features on Google Search are rooted in our core Search ranking and quality systems.” There is no separate AI index or AI ranking algorithm.

What Google has not confirmed: any specific count of sub-queries. The widely repeated “8–12 sub-queries per prompt” figure is SEO-industry estimation, not an official Google number — I’d treat it as illustrative of “several, not one,” not as a hard fact. Google says multiple related searches; it doesn’t publish how many.

Grounding queries — the closest thing to seeing fan-out

You don’t get to see Google’s fan-out sub-queries. Google exposes nothing about them — its June 2026 Generative AI performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. in Search ConsoleGoogle's free tool for monitoring crawling, indexing, and search performance. shows impressions only, broken out by Pages, Countries, Dates, and Devices, with no query dimension.

The one place the industry gets a peek at fan-out-style behavior is Bing/Microsoft. Bing’s AI Performance report in Bing Webmaster ToolsMicrosoft's free portal for monitoring and improving how a site appears in Bing search — the peer to Google Search Console, plus IndexNow instant indexing, richer backlink data, and keyword volumes. Because Bing's index also feeds Microsoft Copilot, it doubles as a window into AI-search visibility. (public preview, February 2026) surfaces grounding queries — the actual retrieval-optimized query phrases that pulled your content into a Copilot or Bing AI answer. These are distinct from what a user typed: they’re simplified, machine-generated search terms. Microsoft Clarity added the same visibility. It’s the nearest thing anyone has to watching the fan-out fire — and notably, Google exposes no equivalent.

Why this is reshaping citations

Here’s the most concrete, measurable consequence of fan-out — and it’s from my own data at Ahrefs.

When we first studied how AI Overview citations line up with organic rankings, about 76% of cited pages also ranked in the traditional top 10 (mid-2025). By the March 2026 update — 863K keyword SERPs and 4M AI Overview URLs, twice the original sample — that number had fallen to about 38%. The rest split almost evenly between positions 11–100 and pages that didn’t rank in the top 100 at all.

The explanation, in our own words: “Google is selecting far fewer pages straight from the original SERP… AI Overviews are relying less on the direct search results and more on the sources showing up in fan out query SERPs.” As fan-out matured, sourcing moved off the head query’s top 10 and onto the SERPs of the sub-queries. Ranking #1 for the exact phrase someone typed is no longer the gate it used to be.

A related pattern from the same work: pages that rank across multiple fan-out queries are meaningfully more likely to be cited — a direct signal that breadth across the sub-topic space beats a single narrowly-optimized page. And YouTube is a disproportionate, ranking-independent citation source: 18.2% of AI Overview citations that don’t rank in Google’s top 100 for the query are YouTube URLs.

What fan-out means for your content

Strip away the mechanism and the playbook is straightforward:

• Cover the sub-topics, not just the keyword. Fan-out retrieves for the questions around the question. Genuine topical depth — the kind that answers the obvious follow-ups on the same page or across a well-linked cluster — is what makes you retrievable across more of those sub-queries. Depth beats a page stuffed around one head term. Retrieval is a prerequisite, not a promise: being pulled into a sub-query’s candidate set doesn’t guarantee the model selects, cites, or accurately attributes your page in the final synthesized answer — treat broader coverage as improving your odds, not as a disclosed citation guarantee.
• Being indexed is still the prerequisite. No separate AI index means the crawl → index → snippet-eligible chain has to be intact. A page that can’t be crawled and indexed can’t be retrieved for any sub-query. Google is explicit: to be shown as a supporting link “a page must be indexed and eligible to be shown in Google Search with a snippet… There are no additional technical requirements.”
• Don’t over-optimize for the mechanism. Google directly rebuts the tactics the fan-out era has spawned: you don’t need llms.txtllms.txt is a proposed (not adopted) Markdown file at /llms.txt that gives AI systems a curated map of a site's most important pages. Proposed by Jeremy Howard in 2024, it's read mostly by coding agents like Claude Code — not search crawlers — and Google ignores it. or special AI markup (“Google Search itself doesn’t use them”), and “there’s no requirement to break your content into tiny pieces for AI to better understand it.” ChunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking. your content into fragments to “match sub-queries” is exactly what Google says not to do — its systems already understand multiple topics on a page.
• Structure for humans. Google’s actual guidance is deliberately minimal: organize content “by paragraphs and sections, along with headings that provide a clear structure.” Clear headings and self-contained sections happen to make individual passages easier to retrieve for a sub-query — but you write them for readers, not for a fan-out.
• Off-site signals still dominate. In the broader research, branded web mentions correlate with AI Overview visibility more strongly than Domain Rating (0.664 vs. 0.326). Fan-out changes which SERP you need to show up on; it doesn’t change that authority and mentions drive citation.

The one-sentence version: fan-out didn’t invent a new optimization — it widened the target. You’re no longer optimizing one page for one query; you’re trying to be a credible answer across the whole neighborhood of sub-queries a topic generates.

TIP

Turn a research hypothesis into a reviewable coverage map with my free Query Fan-Out Simulator Free

1. Start with one seed question and inspect the generated related branches.
2. Compare a page or supplied copy against each branch; treat lexical and cosine coverage as directional, not as Google’s hidden queries.
3. Strengthen genuinely missing answers or evidence, merge overlapping branches, and rerun the check.

Tap to enlarge

The result lists five modeled subqueries: technical SEO audit at 100 percent coverage; What is technical SEO audit at 100 percent; technical SEO audit benefits and limitations at 60 percent; technical SEO audit vs alternatives at 60 percent; and How to choose technical SEO audit at 75 percent. A note says the run used lexical BM25 and cosine coverage only and did not evaluate embeddings.

AI summary

A condensed take on the Advanced version:

• Query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics. = one question, many searches. The model decomposes a single query into multiple related sub-queries, runs them concurrently, and synthesizes the results into one answer. It’s the retrieval step inside RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights./groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it., not the whole system.
• Google confirms it: AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics and data sources.” Gemini then “stitches them together into a single, comprehensive answer.”
• No separate AI indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. or ranking system — it runs over the core Search index; the usual crawl → index → snippet-eligible gates apply.
• The “8–12 sub-queries” figure is industry estimation, not official. Google confirms multiple related searches; it doesn’t publish a count.
• You can’t see Google’s fan-out. GSCA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results.’s Generative AI report is impressions-only, no query dimension. Bing is the exception — its AI Performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. and Microsoft Clarity surface groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries (the machine-generated retrieval phrases).
• Citations are getting less rank-dependent. Direct top-10 sourcing of AI Overview citations fell from ~76% (mid-2025) to ~38% (early 2026) as fan-out matured — sourcing shifted onto sub-query SERPs. Pages ranking across multiple fan-out queries are more likely to be cited.
• SEO upshot: cover the sub-topics (breadth beats one stuffed page); stay indexed; don’t chunk or add special AI markup (Google says not to); structure for humans; off-site authority/brand mentions still dominate.

Official documentation

Primary-source documentation from Google and Microsoft on fan-out, groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it., and how generative features source their answers.

Google

• AI features and your website — the source that confirms AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode “may use a ‘query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics.’ technique,” plus the snippet-eligibility requirement and content controls (nosnippetnosnippet is a robots directive (meta tag or X-Robots-Tag header) that suppresses all snippet text and the video preview for a whole page; data-nosnippet is an HTML attribute on span, div, or section that excludes just that section from the snippet., noindexNoindex is a directive that tells search engines to keep a page out of their index, so it won't appear in search results. It works only on pages a crawler can actually fetch — a page blocked in robots.txt can never be noindexed.).
• Google’s Guide to Optimizing for Generative AI Features — defines RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights./groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. over the core indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed., and directly rebuts chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking., llms.txtllms.txt is a proposed (not adopted) Markdown file at /llms.txt that gives AI systems a curated map of a site's most important pages. Proposed by Jeremy Howard in 2024, it's read mostly by coding agents like Claude Code — not search crawlers — and Google ignores it., and “special AI markup” tactics.
• AI Overviews and AI Mode in Search — confirms no additional requirements beyond standard indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. and snippet eligibility.
• Introducing Search Generative AI performance reports in Search Console — the June 2026 report; impressions-only, no query dimension (so no way to see fan-out sub-queries).
• Generative AI performance report — Search Console Help — the dimensions it does and doesn’t expose (Pages / Countries / Dates / Devices; no clicks, CTR, or queries).

Bing / Microsoft

• Introducing AI Performance in Bing Webmaster Tools (public preview) — surfaces “grounding queries,” the closest public window into fan-out-style retrieval phrases.

Quotes from the source

On-the-record statements from Google. Deep links jump to the quoted passage where the source page allows it.

Google — the confirmed fan-out mechanism

• “Both AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode may use a ‘query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics.’ technique — issuing multiple related searches across subtopics and data sources — to develop a response.” — Google Search Central, “AI features and your website.” Jump to quote

Google — no separate AI indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed.

• “In short, yes! The best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems.” — Google Search Central, AI optimization guide. Jump to quote

Google — eligibility gate (fan-out can’t retrieve what isn’t indexed)

• “There are no additional requirements to appear in AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. or AI Mode, nor other special optimizations necessary.” — Google Search Central, “AI features and your website.” Jump to quote

Google — don’t chunk for the mechanism

• “There’s no requirement to break your content into tiny pieces for AI to better understand it. Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users.” — Google Search Central, AI optimization guide. Jump to quote

Google — structure for readers

• “People generally appreciate it when web pages are organized by paragraphs and sections, along with headings that provide a clear structure to navigate content.” — Google Search Central, AI optimization guide. Jump to quote

The mental models

1. One query in, many queries out. The whole idea in a sentence: a single user question becomes several concurrent sub-searches, and the answer is synthesized from all of them. When an AI answer covers angles the person never explicitly asked about, that’s fan-out at work.

2. Fan-out is the retrieval step of RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights.. Don’t treat fan-out as a separate system. The pipeline is user query → fan-out → retrieve across the indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. → assemble passages → generate a grounded, cited answer. Fan-out is what makes the retrieve step plural.

3. There is no separate AI index. Fan-out runs over the same core Search index and the same ranking/quality systems as blue links. So the crawl → index → snippet-eligible chain is the prerequisite: if a page can’t be indexed, no sub-query can retrieve it.

4. Confirmed mechanism vs. industry theory. Google confirms the technique (multiple related searches) but not a count. Treat “8–12 sub-queries” as a useful mental image of “several, not one” — not as fact.

5. The target widened. Old game: rank one page for one query. New game: be a credible answer across the whole neighborhood of sub-queries a topic generates. Optimize the topic, not the single keyword. Pages that rank across multiple fan-out queries win more citations.

Query fan-out — cheat sheet

What it is in one line One user query is decomposed into multiple related sub-queries, run concurrently over the core Search indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed., and synthesized into a single answer. Confirmed by Google for AI Overviews and AI Mode.

Where it sits

user query → [FAN-OUT into sub-queries] → retrieve across index → assemble passages → generate cited answer

Confirmed vs. estimated

Where you can (and can’t) see fan-out

Do / don’t

• Do: cover sub-topics with genuine depth; stay crawlable + indexed + snippet-eligible; structure with clear headings for readers.
• Don’t: chunk content into fragments to “match” sub-queries (Google says not to); add llms.txtllms.txt is a proposed (not adopted) Markdown file at /llms.txt that gives AI systems a curated map of a site's most important pages. Proposed by Jeremy Howard in 2024, it's read mostly by coding agents like Claude Code — not search crawlers — and Google ignores it. or special AI markup (Google ignores them); assume ranking #1 for the head query guarantees a citation (it doesn’t anymore).

Fan-out readiness checklist

A quick pass to confirm your content can be retrieved across the sub-queries a topic generates:

• Every important page is crawlable and indexedStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. (fan-out can’t retrieve what isn’t in the index).
• Pages are eligible to show with a snippet (not blocked by nosnippet / noindex where you want visibility).
• The topic is covered in genuine depth — the obvious follow-up questions are answered on the page or across a well-linked cluster, not left to a single stuffed keyword page.
• Content is organized with clear headings and self-contained sections (for readers — which also makes passages easier to retrieve).
• You have not fragmented content into tiny AI-only chunks (Google explicitly advises against this).
• No llms.txtllms.txt is a proposed (not adopted) Markdown file at /llms.txt that gives AI systems a curated map of a site's most important pages. Proposed by Jeremy Howard in 2024, it's read mostly by coding agents like Claude Code — not search crawlers — and Google ignores it. / “special AI markup” being relied on as a fan-out lever (Google ignores them).
• Off-site signals — brand mentions, authority — are being built, since they correlate with AI OverviewAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. visibility more than Domain Rating.
• For Bing/Copilot, you’ve checked groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries in Bing Webmaster ToolsMicrosoft's free portal for monitoring and improving how a site appears in Bing search — the peer to Google Search Console, plus IndexNow instant indexing, richer backlink data, and keyword volumes. Because Bing's index also feeds Microsoft Copilot, it doubles as a window into AI-search visibility. / Microsoft Clarity to see which retrieval phrases actually pull your content.
• You’re not assuming a #1 organic ranking guarantees a citation (only ~38% of AI Overview citations now come from a top-10 organic result).

Fan-out, worked through

A single question becomes several. Say the user asks: “Is intermittent fasting good for weight loss?” A regular search runs that one phrase. A fan-out spreads it across the implied sub-questions — roughly:

• how intermittent fasting affects weight
• intermittent fasting vs. calorie restriction
• intermittent fasting side effects
• who should not do intermittent fasting
• best intermittent fasting schedule for beginners

Each sub-query retrieves its own set of pages; the model then synthesizes one answer with citations pulled from across all of them. Note what this means for a publisher: a page that only answers the literal head question, and none of the follow-ups, competes for just one slice of the retrieval — while a genuinely thorough page (or a tight internal cluster) can be retrievable for several.

Why the citation moved off your #1 ranking. Imagine you rank #1 for “best running shoes for flat feet.” Pre-fan-out, that ranking was close to a guarantee of AI OverviewAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. inclusion. Post-fan-out, the overview might be built from sub-query SERPs like “overpronation running shoe features,” “flat feet arch support explained,” and “stability vs. motion-control shoes” — and the pages it cites are the ones that rank for those, which may not be your #1 head-term page at all. That’s the mechanism behind top-10 sourcing dropping from ~76% to ~38%.

Where you actually get to watch it. On Bing, the AI Performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. will show you the groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries that surfaced your page — e.g. your article on robots.txtA plain-text file at the root of a host that tells crawlers which URLs they may and may not request. It controls crawling, not indexing — a blocked URL can still be indexed if it's linked from elsewhere. might have been retrieved by a groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. query like “robots.txtA plain-text file at the root of a host that tells crawlers which URLs they may and may not request. It controls crawling, not indexing — a blocked URL can still be indexed if it's linked from elsewhere. block botsA crawler — also called a spider or bot — is an automated program that fetches web pages, extracts their links, and queues new URLs to visit. Search engines use crawlers to discover and download content for their index. site access,” not the user’s literal “how do I stop Google crawlingCrawling is how search engines use automated bots (like Googlebot and Bingbot) to discover URLs and download pages. A page has to be crawlable to be indexed, but crawling on its own isn't a ranking factor. a page.” That gap between what the user typed and what the machine searched is fan-out made visible — and Google gives you no equivalent view.

Simulating fan-out for research

You can’t read Google’s actual sub-queries, but you can approximate the sub-topic space a query fans out into — useful for planning depth. Below are practical ways to enumerate the “questions around the question.”

Shell — pull People Also Ask–style expansions from a keyword list

# Given a seed query, list adjacent sub-questions to make sure your content covers them.
# (Feed real PAA / related-search data you've exported; this just structures the review.)
seed="intermittent fasting weight loss"
printf '%s\n' \
  "how does $seed work" \
  "$seed vs calorie restriction" \
  "$seed side effects" \
  "who should avoid ${seed% *}" \
  "best $seed schedule for beginners" \
| sort -u

Chrome DevTools Console — extract the “People also ask” questions on a live SERP Run this in the Console on a Google results page to grab the visible PAA questions — a rough, human-visible cousin of the sub-queries a fan-out would generate:

// On a Google SERP with a "People also ask" box open:
[...document.querySelectorAll('div[jsname] [role="button"] span, .related-question-pair span')]
  .map(el => el.textContent.trim())
  .filter(t => t.endsWith('?'))
  .filter((t, i, a) => a.indexOf(t) === i);

Bookmarklet — one-click PAA + related-searches grab Save this as a bookmark and click it on a Google SERP to dump the related-search and PAA phrases to the console for your sub-topic mapping:

javascript:(()=>{const q=[...document.querySelectorAll('.related-question-pair span, a[data-hveid] div')].map(e=>e.textContent.trim()).filter(t=>t&&t.length<90);console.log([...new Set(q)].join('\n'));})();

Python — regex-extract sub-question candidates from your own content Check whether a draft actually answers the sub-questions in a topic before you publish:

import re

subquestions = [
    "how does it work", "vs", "side effects",
    "who should avoid", "schedule for beginners",
]
draft = open("draft.md", encoding="utf-8").read().lower()

for q in subquestions:
    hit = re.search(re.escape(q), draft)
    print(f"{'✓' if hit else '✗'}  {q}")

None of this reveals Google’s real fan-out — it’s a planning aid to make sure your content covers the sub-topic neighborhood a fan-out would probe. For the machine-visible version, use Bing’s groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries (below).

Tools for working with fan-out

• Bing Webmaster ToolsMicrosoft's free portal for monitoring and improving how a site appears in Bing search — the peer to Google Search Console, plus IndexNow instant indexing, richer backlink data, and keyword volumes. Because Bing's index also feeds Microsoft Copilot, it doubles as a window into AI-search visibility. — AI Performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. — the one place you can see real fan-out-style retrieval phrases: it surfaces groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries (the machine-generated searches that pulled your content into a Copilot / Bing AI answer), plus total citations and cited pages.
• Microsoft Clarity — also exposes the groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries behind AI citationsAn AI citation is the visible source link an AI answer engine shows next to its generated text — the clickable reference that credits the web page it used. A citation's presence is a separate thing from whether the cited page actually supports the statement, and from being retrieved (read behind the scenes) or merely mentioned (named without a link); citation is driven more by brand mentions and being retrievable than by traditional ranking., distinct from what users typed.
• Google Search ConsoleA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results. — Generative AI performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. — impressions of your pages in AI features (Pages / Countries / Dates / Devices). Useful for visibility, but it shows no query dimension, so it won’t reveal fan-out sub-queries.
• GSC — regular Performance report + URL InspectionA Google Search Console feature that reports how Google sees one specific URL on a property you own. By default it shows the last-indexed snapshot; a separate \"Test live URL\" mode fetches the current version. — AI feature impressions are folded into the “Web” search type; use this plus URL Inspection to confirm a page is indexedStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. and snippet-eligible (the prerequisite for being retrieved at all).
• Ahrefs (Brand Radar / Site Explorer) — track which pages get cited in AIThree distinct states of AI visibility: retrieved (an AI fetched your page as source material), mentioned (your brand appears in the answer text), and cited (your URL is linked as a source). They don't always happen together, and each is measured with a different tool. answers and filter organic keywords by the AI Overview SERP feature to find pages pulling double-duty, since GSC can’t disambiguate AI clicks.
• People Also Ask / related-searches data — any keyword tool’s PAA and related-terms export is a rough proxy for the sub-topic space a query fans out into — handy for planning content depth.

Model a query fan-out without inventing demand

Paste a primary question, audience, and verified keyword/question research:

Decompose the primary question into the distinct subquestions an answer system may
need to resolve. Use only the supplied research as evidence of demand. Group each
subquestion by intent, entity, comparison, constraint, or follow-up. Mark speculative
branches clearly, merge duplicates, and map each retained branch to an existing page
or a documented content gap. Do not invent search volume.

Test whether one page covers a fan-out coherently

Paste the proposed fan-out and the page’s headings plus section summaries:

For each fan-out branch, identify the best matching section and quote the exact
heading that covers it. Label branches as covered, partially covered, absent, or
better served by another page. Flag sections that try to satisfy unrelated intents.
Recommend heading or internal-link changes without manufacturing facts or forcing
every branch into one URL.

Resources worth your time

My related writing & research

• Update: 38% of AI Overview Citations Pull From The Top 10 — the study behind the “fan-out is doing more of the sourcing” finding (top-10 sourcing fell from ~76% to ~38%); 863K SERPs, 4M AI OverviewAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. URLs.
• 76% of AI Overview Citations Pull From the Top 10 — the original 1.9M-citation study this trend is measured against.
• Insights From 55.8M AI Overviews Across 590M Searches — my largest-scale AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. study: prevalence, triggering conditions, and top-cited domains.
• 8.64% of AI Overviews Appear Outside Position #1 — my study of 10M SERPs on where AI Overviews actually sit.
• What We Actually Know About Optimizing for LLM Search — the Ahrefs write-up using my data on what correlates with AI citationAn AI citation is the visible source link an AI answer engine shows next to its generated text — the clickable reference that credits the web page it used. A citation's presence is a separate thing from whether the cited page actually supports the statement, and from being retrieved (read behind the scenes) or merely mentioned (named without a link); citation is driven more by brand mentions and being retrievable than by traditional ranking. (brand mentions, freshness).
• GEO? AEO? LLMO? What’s With All This AI SEO Stuff? — my Ahrefs Evolve 2025 talk on the AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. landscape and why the indexingStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed. prerequisite hasn’t changed.

My speaking

• GEO/AEO/LLMO — AI SEO webinar — where fan-out, groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it., and retrieval fit into the bigger AI-search picture.

From around the industry

• AI features and your website — Google’s own confirmation of the “query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics.” technique (official, the primary source).
• Google’s Guide to Optimizing for Generative AI Features — Google’s mythbusting on chunkingChunking is splitting a document into smaller passages so AI systems can embed, index, and retrieve the single most relevant piece — not the whole page — in response to a query. It's a foundational step in RAG pipelines and the conceptual cousin of Google's passage ranking., llms.txtllms.txt is a proposed (not adopted) Markdown file at /llms.txt that gives AI systems a curated map of a site's most important pages. Proposed by Jeremy Howard in 2024, it's read mostly by coding agents like Claude Code — not search crawlers — and Google ignores it., and structured dataStructured data is a standardized way of labeling page content (using the schema.org vocabulary in JSON-LD, Microdata, or RDFa) so search engines can understand its meaning. It's not a direct ranking factor — its value is rich results and entity understanding. in the fan-out era.
• Query Fan-Out Guide — Search Engine Land’s practitioner guide to the concept.
• Introducing AI Performance in Bing Webmaster Tools — Microsoft’s report that surfaces groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries (the closest public view of fan-out-style retrieval).
• Grounding Queries and Query Fan-outs — Hive Digital on reading Bing’s grounding queries as fan-out signals.
• Microsoft Clarity Now Shows Grounding Queries Behind AI Citations — Search Engine Journal (Dan Taylor) on the Clarity disclosure.

Stats worth citing

• ~76% → ~38% — share of AI OverviewAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. citations that also rank in the traditional top 10, mid-2025 vs. early 2026, as query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics. matured. The update is Xibeijia Guan’s Ahrefs research (863K SERPs, 4M AI Overview URLs). Source
• 18.2% — of AI Overview citations that don’t rank in Google’s top 100 for the query are YouTube URLs — a disproportionate, ranking-independent citation source in a fan-out world. Source
• 0.664 vs. 0.326 — branded web mentions correlate with AI Overview visibility more strongly than Domain Rating does; fan-out changes which SERP you need to be on, not that authority drives citation. Source
• “8–12 sub-queries” — the commonly cited fan-out count is industry estimation, not an official Google figure. Google confirms multiple related searches; it publishes no number. Present it as an estimate, not fact. Context
• Impressions only, no query dimension — Google’s June 2026 Generative AI performance reportThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. gives you visibility data but no way to see fan-out sub-queries; Bing’s groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it. queries are the exception. Source

Test yourself: Query Fan-Out

Five quick questions on how query fan-outQuery fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics. works and what it means for your content. Pick an answer for each, then check.

Query Fan-Out

Query fan-out is the technique where an AI search system breaks a single user question into multiple related sub-queries, runs those searches concurrently, and synthesizes the retrieved results into one answer. Google confirms AI Overviews and AI Mode 'may use a query fan-out technique' issuing multiple related searches across subtopics.

Related: Retrieval-Augmented Generation (RAG), Grounding, AI Overviews Optimization

Query Fan-Out

Query fan-out is how an AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. system turns one question into many searches. Instead of running a single query against the indexStoring a crawled page in the search index so it can appear in results. Crawled is not the same as indexed — Google selects what to keep, and indexing isn't guaranteed., the model generates a set of related sub-queries across the subtopics and angles implied by your question, runs them concurrently against the search index, and then stitches the retrieved passages into a single synthesized answer. It’s the retrieval-side behavior that makes AI answers feel comprehensive — they’re assembled from several sub-searches, not one.

Google confirms the mechanism in its own words: both AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode “may use a ‘query fan-out’ technique — issuing multiple related searches across subtopics and data sources — to develop a response.” It’s the retrieval step inside the broader RAGRAG is the retrieve-then-generate pattern behind AI search: the system retrieves relevant passages from an external index at query time, injects them into the model's context, and generates an answer grounded in those sources — without changing the model's weights. (retrieval-augmented generation, which Google calls groundingGrounding is anchoring an AI model's answer to source documents it retrieves at the moment you ask — not to the patterns frozen into its weights during training. Retrieval-Augmented Generation (RAG) is the most common way to do it.) pipeline that powers AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode. Crucially, it runs over Google’s core Search index — there is no separate AI index, and no separate AI ranking system.

For SEO, the practical shift is that your content no longer has to satisfy just the head query a person typed — it has to be retrievable for the implied sub-questions a fan-out spawns. That rewards genuine topical depth over a single page stuffed around one keyword. Google’s own account of why direct top-10 sourcing has fallen (from roughly 76% of AI Overview citations in mid-2025 to about 38% in early 2026) is that fan-out is doing more of the work — pulling from the SERPs of sub-queries rather than only the original query’s top results.

Note that the widely cited “8–12 sub-queries per prompt” figure is SEO-industry estimation, not an official Google number. What Google has confirmed is the technique itself and that multiple related searches are issued — not a specific count.

Related: Retrieval-Augmented Generation (RAG), Grounding, AI Overviews Optimization

Changelog