Query fan-out: write for the sub-questions an AI answer is built from

What fan-out looks like

Take the prompt 'What is the best accounting software for a small design agency?'. A system could search for several things: small business accounting software comparisons, features agencies need such as time tracking and project billing, pricing of leading tools, reviews from agency owners, and integrations with common payment providers.

Each of those is a search in its own right, with its own results. The final answer draws on whichever pages answered each part best. The pages that rank for the original long question are only a subset of the pages that shape the response.

Why this changes how you plan content

Keyword-era planning tends to produce one page per head term. Under fan-out, the useful unit is the sub-question. A buyer's single prompt touches pricing, fit, comparisons, trust and implementation, and an engine may look for each separately.

If you have a strong page on 'best accounting software' and nothing on 'how much does accounting software cost for a ten-person agency', you may lose the pricing sub-question to a competitor and appear thinner in the final answer.

This also explains why narrow, specific pages can be cited more than broad ones. A page that thoroughly answers one concrete question is easy to use as evidence.

Related reading: How to write a page AI answers will actually cite and How ChatGPT, Gemini, Perplexity, Claude and Google AI build an answer.

How to find the sub-questions

Ask the engines. Run your important prompts and read the answers: the sections of a good answer correspond to sub-questions. Where search is shown, note the queries or the sources cited; the sources reveal which sub-topics were looked up.

Use your own data. Search Console queries around your topic, support tickets, sales call questions and the 'people also ask' style questions all show how buyers break a decision apart.

List the sub-questions per topic and group them: definition, fit, price, comparison, risk, setup, proof. Most buying topics fall into the same few groups.

How to write a page that answers one sub-question

Lead with the answer. State the conclusion in the first sentence or two, in plain words, then support it. An engine scanning for a quotable answer should find it without reading the whole page.

Be specific. Prices, ranges, dates, numbers of users, named integrations and clear conditions are all things an engine can quote. Vague claims give it nothing to cite.

Say who it is for and who it is not for. Honest limits make a page more credible and more useful, and engines tend to repeat the framing they find.

Use headings that match how people ask. A heading phrased as the question makes the structure obvious to a reader and to a model.

Keep one main question per page. If you cover several, give each its own clearly labelled section so each part stands alone.

Support the page so it is trustworthy

Show who wrote it and when it was updated. Freshness and accountability are cheap signals that help both people and systems judge reliability.

Add structured data where it fits, such as Article and FAQPage, so the page's type and its questions are explicit. Structured data helps interpretation but does not guarantee inclusion.

Link related sub-question pages together so a reader, and a crawler, can move from the overview to the detail.

Measure whether it worked

Track a prompt group that includes both the head question and the sub-questions you wrote for. Watch whether your brand is named more often, whether new pages of yours appear among the cited sources, and what the answer now says about the topic.

In OClever, Content Studio can draft pages from the gaps found in your tracked prompts, using your brand facts so claims stay accurate, and the Prompts view shows afterwards whether those pages are being cited.