What Is Query Fan-Out, and How Do LLMs Use It to Retrieve Search Results?

How does query fan-out work?

A typical sequence:

1. The user asks a question, often long and with conditions: "What's the best CRM for a 10-person estate agency that uses Gmail, under $50 per user?"
2. The system breaks it into sub-queries covering the parts of the need, such as the category, the use case, the integration and the price.
3. It searches for each sub-query, in a web index, and sometimes in other sources such as maps, shopping data or a knowledge graph.
4. It reads and selects passages from the results for each sub-query.
5. It writes one answer, often citing several of the pages it used.

Google has described AI Mode as grouping a question into subtopics and searching for each one simultaneously. Other assistants that search the web behave similarly, deciding for themselves which searches to run.

What might the sub-queries look like?

For the CRM question above, plausible sub-queries include:

Each sub-query retrieves different pages: a comparison article for the first, a vendor's integrations page for the third, a pricing page for the fourth. The final answer blends them.

What are grounding queries, and how are they different from search volume?

Grounding queries are the searches an AI system runs to find sources that ground its answer, in other words the sub-queries. Microsoft's Bing Webmaster Tools shows a sample of the grounding queries that led to your pages being cited in Copilot and Bing's AI answers.

They are not the same as search volume:

• Search volume estimates how often people type a keyword into a search engine.
• Grounding queries are written by the AI system, not the person, and reflect how it decomposes questions.

A grounding query can be something almost nobody types, yet it decides which pages are retrieved for many different user questions. That makes grounding query data, where you can get it, a useful guide to the sub-questions your content should answer.

How does query fan-out change keyword research and search intent?

• From single keywords to question trees. Map a buyer question into its parts: category, use case, constraints, evaluation, and next steps.
• From head terms to specific sub-topics. Pages that answer narrow parts well, such as integrations, pricing or suitability for a segment, get retrieved for sub-queries even when they would never rank for the head term.
• From one page per keyword to coverage of a need. A cluster of pages, or clear sections on one page, should cover the sub-questions a buyer's prompt implies.
• Intent becomes compound. One prompt can contain informational, comparative and transactional intent at once, so content needs to support several.

Classic keyword data still helps: many sub-queries look like ordinary searches, and ranking for them matters, especially in Google's AI features. How to organise tracking around needs rather than keywords is covered in whether GEO strategy should track intent clusters instead of keywords.

How do you optimise content for fan-out sub-queries?

1. List the prompts your buyers ask, with their usual conditions.
2. Break each into likely sub-queries: what the category is, who it suits, how it integrates, what it costs, how it compares.
3. Check what ranks for those sub-queries in Google and Bing, and what AI answers cite.
4. Make sure you have a clear answer for each sub-query, as a section with a matching heading or a dedicated page.
5. State the specifics each sub-query needs: prices, integrations, limits, segments, in plain text.
6. Link related pages so engines and readers can move from one sub-answer to the next.

An illustrative example: a CRM vendor's site had one long features page. After mapping sub-queries, it added a "CRM for estate agents" page, a Gmail integration page with setup details, a pricing page with per-user prices in text, and a fair comparison page against two rivals. Each page answers one part of the prompt, and together they give the vendor four ways to be retrieved.

How do you target sub-queries so Claude includes your content?

When Claude uses web search, it runs its own searches and reads the results, so the same principles apply:

• Allow Anthropic's search and user-request crawlers, Claude-SearchBot and Claude-User, so your pages can be indexed and fetched.
• Answer specific sub-questions clearly, with the answer at the start of the relevant section.
• Be findable in Brave Search for those sub-queries. Anthropic lists Brave Search as its web search provider, and in Profound's analysis of about 35,000 Claude citations, 79% came from Brave's top 10 results against 34% from Google's top 10. Search your sub-queries on search.brave.com and see whether your pages appear.
• Be present on the third-party pages that come up for comparison sub-queries.

Where an assistant shows the searches it ran, note them; they are direct evidence of the sub-queries that matter for your topic. For ChatGPT, see how to use browser tools to reveal ChatGPT's search queries and fan-outs.

How do you find out which sub-queries matter for your topic?

• Bing Webmaster Tools: the AI Performance report's sample of grounding queries for your cited pages.
• Visible search steps: some assistants show the searches they run; record them for your key questions.
• Search Console: long, specific queries that bring impressions to your pages often resemble sub-queries.
• Reverse engineering: look at the sources cited for a prompt and ask what search each one was probably retrieved for.

What mistakes should you avoid?

• Writing one generic page for a complex buying question and expecting it to win every sub-query.
• Creating dozens of near-duplicate pages for every possible variation. Cover distinct sub-questions, not every phrasing.
• Ignoring evaluation sub-queries such as reviews and comparisons, which are often answered by third-party pages.
• Assuming fan-out makes classic SEO irrelevant. Many sub-queries are retrieved from ordinary search results.

What do people ask about query fan-out?

Is query fan-out only a Google feature?

Google describes it for AI Overviews and AI Mode, but many AI assistants that search the web break questions into several searches in a similar way.

Can I see the exact fan-out queries for my prompts?

Only partly. Bing shows a sample of grounding queries for your cited pages, and some assistants show their search steps. Otherwise, you have to infer them.

Does fan-out favour long-form content?

It favours content that answers specific sub-questions clearly. That can be a long page with clear sections or several focused pages.

Should I write content for sub-queries nobody searches for?

If the sub-query reflects a real part of a buyer's need, yes. Grounding queries do not need human search volume to matter.