Information-gain assets to add before publish:
- A worked fan-out map for one of your own live pages: the main query, the sub-questions you extracted, and which you already cover.
- Before and after heading structure for that page, as screenshots.
- Citation status 30 and 60 days after the restructure.
An AI Overview is not an answer to your query. It is an answer assembled from answers to several related questions the system generated on your behalf. Those generated questions are the fan-out, and pages that rank for them are markedly more likely to be cited than pages that rank only for the query you typed.
The short answer
Google decomposes a query into related sub-questions, retrieves passages for each, and synthesises them. Surfer SEO’s analysis of 10,000 keywords and 173,902 URLs found pages ranking for fan-out queries were 161% more likely to be cited than pages ranking only for the main query. Structure your page around the sub-questions and you enter more of those retrieval pools.
What the fan-out data shows
The Surfer study extracted roughly 33,000 fan-out queries across its sample. The distribution of citations is the interesting part:
| Ranking pattern | Share of AI Overview citations |
|---|---|
| Ranks for the main query only | just under 20% |
| Ranks for fan-out queries only | about 30% |
| Ranks for both | about 51% |
Two things follow. First, fan-out-only ranking produces more citations than main-query-only ranking. Second, the combination produces more than either alone, which is what you would expect if retrieval runs separately for each sub-question.
The study also found a strong correlation, Spearman 0.77, between ranking across multiple fan-outs and citation probability, and reported that roughly 68% of cited pages did not rank in the top ten for either the main query or a fan-out. Ranking helps. It is not the gate.
How to build a fan-out map
You cannot see Google’s generated sub-questions directly. You can approximate them well enough from public signals.
- Start with the main query and search it in an incognito window with the correct location.
- Record every People Also Ask question, then expand each one and record the new questions that appear beneath. Two levels is usually enough.
- Record the related searches at the foot of the page.
- Record the autocomplete suggestions for the query plus each of who, what, when, why, how, cost, best, vs.
- Read the AI Overview itself if one appears. Its internal structure often mirrors the sub-questions it answered.
- Group the resulting questions by intent. Several will be phrasings of the same thing; collapse them.
What you end up with is a list of eight to fifteen genuine sub-questions. That is your heading structure.
Turning the map into a page
The temptation is to make each sub-question an H2 and write a paragraph under it. That produces a page that reads like a quiz. Do this instead:
- Group sub-questions into three to six themes. Each theme becomes an H2 with a descriptive, human heading.
- Put individual sub-questions as H3s inside the relevant theme, phrased the way a person would ask them.
- Answer immediately under each H3, in 40 to 60 words, as a self-contained statement. Someone reading only that paragraph should get a complete answer.
- Then expand. Nuance, exceptions and examples go after the direct answer, not before it.
- Use a table when the sub-question is comparative, and a numbered list when it is procedural. Both are easier to extract and easier to read.
What a self-contained answer looks like
Weak: “This depends on a number of factors, which we will explore below.” Nothing is answerable from that.
Strong: state the answer, give the main condition, note the exception. Forty to sixty words. Then keep going for the reader who wants the detail.
Coverage, not repetition
A common failure is to answer the same sub-question five times in slightly different words, on the theory that more mentions means more relevance. It does not. What increases citation likelihood is covering more distinct sub-questions well, not covering one sub-question repeatedly.
If two sections would say the same thing, they are one section.
One page or several?
Fan-out coverage argues for depth on one page. Site architecture sometimes argues for separate pages. The test:
- Same intent, different angle: one page, multiple sections.
- Different intent, even if the words overlap: separate pages, linked to each other.
“How much does X cost” and “is X worth it” look adjacent but are different decisions at different stages. Splitting them lets each be a complete answer; merging them makes both partial.
What this does not require
- No special markup. Google’s documentation on AI features states you do not need new machine-readable files, AI text files or markup to appear in these features.
- No artificial chunking. Clear structure helps. Chopping prose into fragments to look machine-friendly does not.
- No thirty-question FAQ block. Coverage of real sub-questions is the point; volume of questions is not.
How to check whether it worked
- Record which fan-out queries your page ranks for before the restructure.
- Restructure.
- Wait for recrawl and reindexing, then recheck the same list after 30 and 60 days.
- Track AI Overview citation separately from ranking, because the two move independently.
- Judge by the number of sub-questions you now rank for, not by the position on the head term.
Frequently asked questions
What is a query fan-out?
It is the set of related sub-questions a generative search system creates from your original query, retrieves passages for, and synthesises into a single answer. The answer you read is assembled from several retrievals, not from one ranked page.
How do I find the fan-out queries for my topic?
Approximate them from People Also Ask, expanded two levels deep, plus related searches, autocomplete variations and the internal structure of any AI Overview already showing. Group the results by intent and collapse duplicates into a single sub-question each.
Does ranking for fan-out queries guarantee a citation?
No. Surfer SEO’s data shows a 161% higher likelihood of citation for pages ranking on fan-outs, and a strong correlation between breadth of fan-out coverage and citation probability. It raises the odds substantially; it does not guarantee anything.
Should every sub-question be its own heading?
Group them. Three to six themed H2s with individual sub-questions as H3s reads like a document rather than a quiz, and still gives each sub-question its own extractable answer. Coverage matters; the number of headings does not.
How long should the answer under each heading be?
Forty to sixty words, self-contained, before any expansion. If someone read only that paragraph they should have a complete and correct answer. Everything after it exists for the reader who wants the detail.
Sources
- Ranking for AI Overview fan-out queries boosts citation odds – Search Engine Land, reporting Surfer SEO research on 10,000 keywords and 173,902 URLs
- AI features and your website – Google Search Central