Question-Format Headings: How to Write H2s and H3s That AI Engines Extract
AI engines extract answers from the first clean paragraph after a heading. When that heading is phrased as a question, the extraction model can match it precisely to a user query. Topic-statement headings lose that signal. This guide covers why question-format headings matter, how to write them, and how to audit your existing pages to maximise AI citation coverage.
This is not theory. It is how retrieval-augmented generation works in practice. When an AI engine receives a user query, it searches for content that matches. A question-format heading gives the engine a direct semantic match. A topic-statement heading gives it a loose topic match. Direct matches win.
How AI engines read headings
When an AI engine processes a page, it scans the heading hierarchy to understand what the page covers. It models each heading as a potential question, then looks for the paragraph that answers it. This is not猜测. It is how transformer-based retrieval works.
If the heading reads "llms.txt guide", the AI engine interprets this as: the section covers llms.txt. Good, but imprecise.
If the heading reads "How does llms.txt work?", the AI engine interprets this as: this section answers "how does llms.txt work?". Better. It knows exactly what question the content is designed to answer. The paragraph below becomes a candidate answer for that specific query.
This matching happens across ChatGPT, Gemini, Perplexity and Google AI Overviews. Each platform uses slightly different extraction logic, but all of them weight question-format headings more heavily than topic statements. The pattern is consistent because it is fundamental to how retrieval-augmented generation works: match query to content, extract the closest answer.
<h2>llms.txt guide</h2>
Interpretation: "The section covers llms.txt as a topic."
<h2>How does llms.txt work?</h2>
Interpretation: "This section answers 'how does llms.txt work?'" - exact match.
Question-phrase patterns AI engines recognize
These are the heading formats that trigger the strongest extraction signal:
How to check your current headings
For an automated check across all your pages: run a free SearchScore audit. The E-E-A-T category includes question-phrase coverage scoring, so you can see exactly which pages need attention without manual inspection.
The question hierarchy technique
The most effective pages use a nested question structure where H2s ask broad questions and H3s ask specific follow-ups. This mirrors how users actually search: they start with a broad question, then drill into specifics.
This structure gives the AI engine multiple entry points. If a user asks a broad question, the H2 matches. If they ask a specific follow-up, the H3 matches. Both lead to your content being extracted.
Rewriting headings without changing content
Headings to avoid entirely
Some heading patterns actively hurt AI extraction, even if they look natural in a traditional editorial context:
- Clever or witty headings. "The Secret Weapon You Never Knew You Needed" tells an AI engine nothing about what the section covers. Rewrite as a question.
- Vague nouns. "Overview", "Introduction", "Background" carry no semantic signal. Replace with "What is [topic]?" or "How did [topic] develop?".
- Single-word headings. "Pricing", "Features", "FAQ". These work for human navigation but not for AI extraction. Use "How much does [product] cost?", "What does [product] do?", "Common questions about [topic]".
- Clickbait patterns. "5 Things You Need to Know" is not a question. Rewrite as "What are the 5 most important things about [topic]?".
Common topic-statement headings to rewrite
Frequently Asked Questions
Will rewriting headings hurt my Google rankings?
No. Short, descriptive headings are what Google expects. Question-phrase headings are more specific - not a downgrade.
Google has never penalized for headings that read as questions. Featured snippets actively reward short, direct headings that answer the searcher's question. There is no known risk to rewriting headings for clarity.
Do H3s also need to be question-phrase?
Yes, where practical. Sub-sections should follow the same pattern as H2s. H3s phrased as questions compound the extraction signal.
H3 headings work as sub-questions within a section. If your H2 is "How does llms.txt work?", the H3s underneath should be "How do I create an llms.txt file?" and "What should I include in llms.txt?". The question hierarchy helps the extraction model navigate the page structure.
What if my topic does not fit a question format?
Most topics can be phrased as a question. "Case studies" becomes "Who uses [product] and what results do they get?" Test different question framings.
If a section genuinely has no question framing ("Case studies", "About us", "Contact"), it is fine to leave it as a topic statement. The target is 80%+ question-phrase on your main content pages - not 100% across every page.
Does this work for product pages and landing pages too?
Yes. Product pages benefit even more because AI engines frequently answer "which [product] should I use?" queries. Your product page headings should match those queries.
A product page with headings like "Features", "Pricing", "Testimonials" is a missed opportunity. Rewrite as "What does [product] do?", "How much does [product] cost?", "Who uses [product] and what results do they get?". These match the queries AI engines receive from users comparing options.
How many question headings should a page have?
Aim for 5-10 question-format headings per comprehensive guide page. Fewer than 5 suggests shallow coverage. More than 10 risks diluting the page's focus.
The ideal number depends on the topic depth. A pillar page covering a broad topic might have 8-12 question headings. A focused sub-article might have 4-6. The key metric is coverage: are you answering the questions your audience is actually asking AI engines? If your heading hierarchy maps to common queries, the number will naturally fall into the right range.
See your full score across all Q&A structure signals.
Run a free audit at searchscore.io - question-phrase headings, answer-first paragraphs, and FAQ schema all scored together.
Check your AI visibility
Free audit. Instant results. No sign-up required.