What is query fan-out? How AI search turns one question into many

If you’ve ever wondered why an AI answer cites a page that doesn’t rank for the question you asked, query fan-out is usually the reason. The assistant didn’t search your question. It searched five smaller ones, and that page was the best answer to one of them.

How query fan-out works

Take a question like “What’s the best CRM for a 10-person real estate team that already uses Google Workspace?” No single page answers all of that. So the assistant breaks it up:

Each part of the final answer may come from a different page. A page about Google Workspace integrations can be cited for that one point without ever mentioning real estate.

Who uses it

The names differ, but the idea is the same everywhere: the search the AI runs isn’t the question the user typed.

This is also why AI answers often feel more complete than a single search result. The tool is doing the work a careful researcher would do: searching the parts of the question separately, reading several sources, and putting the pieces together. For anyone trying to be cited, the implication is simple. You’re not competing for one search. You’re competing for a place in several, and each one can be won by a different page, on your site or someone else’s. The rest of this guide shows how to find those searches and make sure you have a strong answer for the ones that matter to buyers.

Why it changes which pages get cited

• Rankings for the main keyword matter less. Across ChatGPT, Gemini and Copilot, only 12% of cited links ranked in Google’s top 10 for the same query in Ahrefs’ study. The sub-searches pull from different results.
• Narrow pages can win. A focused page that answers one sub-question precisely can be cited alongside much bigger sites.
• Sections matter more than pages. The AI takes the passage that answers its sub-search. A page with ten clear sections has ten chances.
• Coverage across sources matters. If roundups and reviews answer the “best for” sub-search and you aren’t in them, you miss that part of the answer.

How to find the sub-queries for your topics

You can’t see Google’s or ChatGPT’s fan-out directly, but you can reconstruct it closely enough to plan content. Use four sources:

1. Read the AI answer. Ask the question in ChatGPT, Perplexity and Google AI Mode. Each distinct point or paragraph usually maps to one sub-search. Write them down.
2. Check the cited sources. Note which passage each source supplied. The page title often reveals the sub-search it matched.
3. Use “People also ask” and related searches in Google. They’re close cousins of the fan-out queries.
4. Read your Bing grounding queries. If you’ve verified your site in Bing Webmaster Tools, the AI Performance report shows a sample of the phrases Copilot used to find your cited pages. That’s real fan-out data for your own site.

Bing notes that grounding queries aren’t users’ prompts. They’re the AI’s own retrieval phrases, which is exactly what you want to see.

A fan-out map template

For each important buyer question, build a map like this and keep it in a spreadsheet:

How fan-out differs by question type

Different kinds of questions split in predictable ways. Use this as a starting point when you map your own topics:

Notice how often reviews, alternatives and pricing appear. Those sub-searches show up in almost every buying question, which is why pricing pages, comparison pages and review profiles carry so much weight in AI answers.

How to prioritise sub-queries

A single topic can produce dozens of sub-queries. You can’t cover them all at once, so score each one from 1 to 3 on three things and start with the highest totals:

• Buying intent: does answering it help someone choose a product? Pricing and comparison sub-queries score high; definitions score low.
• Gap: is the current cited source weak, outdated or missing? A thin or old source is easier to replace.
• Fit: can you answer it better than anyone, from your own data or experience?

A sub-query scoring 8 or 9 deserves a dedicated section or page this month. One scoring 3 or 4 can wait, or be handled by getting listed on the source that already answers it.

What fan-out means for measurement

Fan-out makes keyword rank tracking a weak guide to AI visibility. You might rank fifth for your main keyword and still be cited for three of its sub-queries, or rank first and be cited for none. Track three things instead: whether you appear in the AI answer for the main question, which of your pages are cited and for which part of the answer, and, through Bing’s grounding queries, the phrases that led AI to your pages. Our guide to tracking brand mentions in ChatGPT shows a simple monthly method.

How to write for fan-out

You don’t need a new kind of content. You need ordinary pages organised so each part can stand alone:

• One sub-question per section, with the question as the heading.
• Answer in the first sentence of the section, then explain.
• Repeat the subject by name instead of “it” or “this”, so the passage makes sense out of context.
• Use HTML tables for comparisons and pricing, which are common sub-searches.
• Cover the qualifiers people add: team size, industry, budget, integrations, location. Those words show up in fan-out queries.
• Link related sections and pages so crawlers find the whole cluster.

A 30-minute fan-out exercise

Try this on your single most important buyer question before you plan anything bigger:

1. Minutes 0–5: write the question the way a real buyer would ask it, with their situation included: team size, industry, budget, tools they already use.
2. Minutes 5–15: ask it in ChatGPT (with search), Perplexity and Google AI Mode. Copy each answer into a document.
3. Minutes 15–20: split each answer into its separate points. Label each point with the sub-question it answers.
4. Minutes 20–25: list the sources cited for each point, and mark the ones that are yours.
5. Minutes 25–30: circle the sub-questions where you have no page and no presence on the cited sources. Those are your first three tasks. Repeat the exercise for your next four questions over the following week and you’ll have a content plan grounded in real answers.

A common result the first time you do this: your own site answers the product question, and almost nothing else. The pricing, comparison, review and “for my situation” parts of the answer come from other people.

Fan-out and content clusters

If you’ve built topic clusters for SEO, a hub page with supporting articles around it, you’re already halfway there. Fan-out rewards the same structure, with two changes.

First, the supporting pages should map to sub-queries you’ve actually seen in AI answers, not just keyword variations from a tool. Second, the hub page itself should answer each sub-question briefly in its own section and link to the deeper page, so both the hub and the supporting page have a chance to be cited. A cluster that looks tidy in a keyword tool but doesn’t match what AI tools search for won’t help much.

A worked example

Say you sell accounting software for freelancers. Your main page targets “accounting software for freelancers”. When someone asks ChatGPT “What’s the easiest way for a freelance designer in the UK to handle taxes and invoicing?”, the fan-out likely includes searches about UK self-assessment for freelancers, invoicing tools for designers, expense tracking, and accounting software comparisons.

Your product page answers maybe one of those. A plan built around fan-out would add: a guide to self-assessment for UK freelancers with a clear deadline table, an invoicing page for creative freelancers, an expenses guide, and a fair comparison page. Each answers one sub-search directly. Together, they give you a chance at every part of the answer.

Common mistakes

• Tracking only the head keyword. Track the sub-queries too, because that’s where most citations are won.
• Assuming fan-out queries are the same across tools. Google, ChatGPT and Copilot split questions differently. Check each.
• Writing headings for style. “The bottom line” matches nothing; “How much does X cost for a 10-person team?” matches a sub-search.
• Skipping Bing Webmaster Tools. Its grounding queries are the most direct fan-out data you can get for free.

For the bigger picture of planning content around AI questions, see our guide to prompt research, and for applying this to Google specifically, how to get cited in AI Overviews.