Short answer. A question heading followed by a two-sentence answer works because it makes the boundary of an extractable passage explicit. An engine assembling an answer needs a span it can lift without dragging in context; a heading that states the question and a paragraph that resolves it immediately supplies exactly that. It is a structural aid, not a ranking mechanism — the same page written answer-first with no evidence in it is still not worth quoting. Used badly, the pattern degrades into thin repeated Q&A blocks, which is worse than the prose it replaced.
This is a craft guide rather than a strategy one: how to phrase the heading, how long the answer runs, how many sections a page should carry, when a question heading is the wrong choice, and the specific failure modes that make answer-first writing read like a form.
Why do question headings work?
Three mechanisms, none of them mysterious.
They mark the passage boundary. Extraction is easier when a section announces what it resolves and then resolves it. A heading reading "Our approach" tells a retrieval system nothing about what the following paragraphs answer.
They match how questions are actually asked. Conversational queries are phrased as questions. A page whose headings are questions in the buyer's own words matches at the level the question was asked, rather than at the level of a product category.
They align with query decomposition. Generative search systems split a complex question into related sub-questions and retrieve separately for each. A page that covers the natural sub-questions of a topic has multiple independent ways into the candidate pool instead of one.
That third mechanism is the one worth designing around. It means the value of a page is partly a function of how many distinct sub-questions it genuinely answers — not how many times it mentions the topic.
What is an answer-first paragraph?
An answer-first paragraph resolves the heading's question in its first one or two sentences, then earns the claim in the sentences after it.
The test is mechanical: delete everything except the heading and the first two sentences. If what remains is a complete, correct, quotable answer, the paragraph is answer-first. If what remains is a preamble — context, history, a definition of a term nobody asked about — it is not.
| Pattern | Opening move | Extractable? |
|---|---|---|
| Answer-first | States the conclusion, then justifies it | Yes |
| Narrative | Builds context, arrives at the conclusion late | Rarely |
| Definitional preamble | Defines adjacent terms before answering | Only the definition, which was not the question |
| Hedged | Opens with "it depends" and never lands | No |
"It depends" is the most common failure, and it is usually recoverable. The fix is to answer the majority case first and state the exception second: "Usually X. The exception is Y, where Z applies." That is both more honest and more liftable than opening with the caveat.
How specific should the heading be?
Specific enough that the answer beneath it is bounded, general enough that a buyer would actually phrase it that way.
- Use the buyer's words, not internal names. A heading built around a product name answers a question only existing customers ask.
- One question per heading. "How does it work and what does it cost?" produces a section that answers neither cleanly.
- Avoid the exact-match reflex. Repeating the primary keyword in every heading makes the page awkward to read and measured as minimal in effect — and keyword stuffing measured actively negative in the largest controlled study of generative visibility, Aggarwal et al. at ACM SIGKDD 2024, which found statistics improved visibility by up to 40% and quotations by roughly 30% while stuffing scored −10%.
- Prefer a question a competitor would also have to answer honestly. Those are the ones where a specific answer wins.
Not every heading needs to be a question. Comparison tables, procedures, definitions and checklists are often clearer under a descriptive heading, and a page of nothing but questions reads like a form. A reasonable working ratio is most major sections as questions, structural sections as labels.
How many sections should a page carry?
As many as the topic genuinely contains, which for most substantial pages is five to nine major sections and, at 1,400–2,200 words, roughly 150–300 words per section.
The two failure directions:
Too few, too long. A 700-word section answers several questions at once and is not liftable as any of them. Split it at the point where the subject changes.
Too many, too thin. Twenty two-sentence Q&A blocks is not an answer-first page; it is a FAQ pretending to be an article. Thin repeated blocks are the specific pattern that scaled-content guidance targets, and they are also worthless to a reader.
The discipline that resolves both: each section must be readable alone. Take any section out of the page, hand it to someone who has not read the rest, and ask whether it still makes sense and still answers something. Unresolved "as described above" is the tell.
What actually goes underneath the answer?
The heading and the opening sentences make a passage extractable. What makes it worth extracting is what follows.
- A number with a source. Roughly eight per 1,000 words is a workable density; dates, counts, thresholds, versions and durations all qualify if attributable.
- A named authority, quoted. Naming the source in the sentence, not only in a link, so the attribution survives being lifted.
- A formula or defined method, where the subject allows one. These are rare in body copy and unusually citable, because they state how rather than asserting that.
- The limit. Where the claim stops holding. A passage that states its own boundary is more useful to a system assembling a balanced answer than one that does not.
Structure without evidence produces a page that is easy to extract from and contains nothing worth extracting. That is the most common way a well-intentioned AEO rewrite fails.
Where question headings earn their keep beyond the page
Each strong question is a natural anchor for a deeper page, which is how a blog becomes a connected structure rather than a set of unrelated posts. A section answering "How do you measure this?" in 200 words should link to the page that answers it in 2,000, and that page should link back.
Two rules keep this from degenerating:
- Link on the question, not on "click here". The anchor text is a signal about what the destination resolves.
- Do not split a question across two pages. If a section needs its own page, the section becomes a summary with a link — not half an answer that requires the click to be complete.
How Lifewood approaches this
Lifewood applies this as a pre-ship gate rather than an audit afterwards. A draft that fails the deletion test — heading plus first two sentences do not stand alone — or that carries structure without evidence is rewritten before publication, because retrofitting specifics into finished copy costs considerably more than writing them in.
For multi-market work the headings themselves are written in-market, not translated. Across 50+ languages and 40+ delivery centres across 30+ countries, the recurring finding is that the question a buyer asks in one market is frequently not the source-market question restated — so a translated heading set answers questions nobody in that market is asking, however fluently.
See AEO services for how this sits inside a measured programme.
Sources and further reading
- Aggarwal et al., "GEO: Generative Engine Optimization", ACM SIGKDD 2024 — the measured effects of statistics, quotations and keyword stuffing on generative visibility.
- Google Search Central, "Optimizing your website for generative AI features" and the structured data policy guidelines — on people-first writing and markup that matches the visible page.
- Companion guide: What Actually Gets You Cited by AI Answer Engines — the page-level rubric this structural work feeds.