Question-format H2s: why AI assistants prefer them

Question-formatted H2s — section headings phrased as questions rather than statements — produce roughly 3× higher AI citation rates than statement-formatted headings per our internal audit observations across 20+ client engagements. The mechanic is direct: ChatGPT, Perplexity, Claude, and Gemini all preferentially extract passages that immediately follow a question heading because the structural pattern mirrors how the assistant produces its own response. A passage following "How does X work?" gets extracted as the answer to similar buyer queries about X. The same passage following "X Overview" gets extracted at materially lower rates because the structural signal doesn't match the assistant's generative pattern. Question-format H2s are dimension #2 in the 40-Point AEO Content Geometry Standard, allocated 5 of the 100 total points. Articles where 100% of H2s are question-formatted score full points on this dimension; articles where 60-79% of H2s are question-formatted score 2 points; articles below 40% score zero. This guide unpacks why the mechanic works, the practical rules for phrasing H2s as buyer questions, and the common authoring mistakes that produce statement headings under the false belief they're more "professional."

Why do AI assistants prefer question-format H2s?

The mechanic reflects how language models learn to associate structural patterns with content types. AI assistants are trained on massive corpora that include Stack Overflow Q&A, Quora discussions, FAQ pages, documentation organized around questions, and countless other sources where text is structured as "question, then answer."

The model learns to associate the structural pattern question heading followed by paragraph with citable answer content. When generating a response to a user query, the model preferentially extracts from sources whose structure mirrors the response it's about to produce.

This produces three observable behaviors:

Higher extraction rate on question-headed sections. A passage following "What is AEO?" gets extracted as the answer to user queries about AEO at materially higher rates than the same passage following "AEO Overview" or "About AEO."

Cleaner attribution preservation. Question-headed sections produce verbatim extraction more often than statement-headed sections, which the model tends to paraphrase. Verbatim extraction preserves brand attribution; paraphrasing dilutes it.

Better alignment with buyer query phrasing. Buyers phrase questions to ChatGPT conversationally — "how do I find a reputable painter in San Diego" rather than "best San Diego painters." Question-format H2s that match buyer phrasing route the article into the exact queries it's optimized to answer.

The 3× extraction rate ratio comes from internal A/B observation across audit work. Articles where authors rewrote statement headings to question format saw measurable citation lift on ChatGPT and Perplexity within the next crawl cycle.

What makes a good question-format H2?

The structural pattern is "question word + verb + noun" or close variants. Examples that work:

✅ "How does ChatGPT decide what to cite?" ✅ "What is brand mention frequency?" ✅ "Why does NAP consistency matter?" ✅ "Where should brands start with AEO?" ✅ "When does Knowledge Graph eligibility expire?" ✅ "Which AI bots should you allow?" ✅ "What's the difference between AEO and GEO?"

Each leads with a question word (how, what, why, where, when, which), uses an active verb, and references a specific entity or concept buyers would actually search for.

Variants that work less well:

⚠️ "How Does It Work" (title case feels more like a statement and less like a buyer query) ⚠️ "Why Is This Important" (vague — what's "this"?) ⚠️ "What About X?" (informal — okay for less-important sections but not pillar H2s)

Variants that don't work:

❌ "Working Mechanism Overview" (statement, no question signal) ❌ "Importance Considerations" (abstract noun phrase, no buyer-query alignment) ❌ "X: A Deep Dive" (colon-separated title-style heading)

The empirical rule: phrase every H2 as a question a buyer might literally type into ChatGPT. If the H2 doesn't sound like a buyer's spoken question, rewrite it.

What about H3s and deeper headings?

H3s benefit from question format but not at the same intensity as H2s. H2s are the highest-extraction structural signal because they typically introduce article sections that AI assistants treat as standalone answer candidates. H3s introduce sub-sections that get extracted less frequently as standalone answers.

The practical rule:

• H2s: question format always. Required for the 40-Point Standard scoring.
• H3s: question format when natural, declarative when not. Optimization-positive but not required.
• H4+: declarative formatting fine. Deep headings function as structural organization more than citation signal.

The FAQ section at the end of articles is a special case — every Q&A pair has a question in its "Q" position by definition. The schema is doing the work; the heading formatting matters less.

How do question H2s interact with SEO keyword targeting?

Question-format H2s naturally align with conversational AI queries but can feel constraining if the article also needs to target traditional fragmented keywords for Google SEO.

Three approaches reconcile the tension:

Approach 1: Question H2s with keyword-rich answers. Author the H2 as a natural buyer question. Cover the target keyword in the section's prose. Both AI assistants and Google index the body content; the question H2 doesn't prevent keyword ranking.

Approach 2: Hybrid — question H2s for major sections, occasional keyword-anchored H3s. Use question H2s for the structural skeleton. Within sections, allow keyword-anchored H3s when they match natural sub-question structure.

Approach 3: Two-track articles for queries with strong divergence. Rare but real: some target queries have strong AEO/GEO interpretation ("how do I get cited by ChatGPT") and strong traditional SEO interpretation ("ChatGPT citation API"). For these, two separate articles addressing the two interpretations may outperform a single hybrid article.

In practice, Approach 1 covers 90% of cases. Question H2s with substantive body content rank fine on Google while producing the AI citation lift the standard targets.

What are common authoring mistakes that produce statement H2s?

Five authoring patterns produce statement H2s under the false belief they're more "professional":

Mistake 1: Title-case formatting feels more authoritative. "The Importance of Brand Mention Frequency" feels more authoritative than "Why does brand mention frequency matter?" — but the statement format scores worse on AEO and converts no better on conversion-rate measurement. Authorial preference for title case usually reflects style habits from print editorial rather than empirical optimization.

Mistake 2: Avoiding question marks because they "feel informal." Question marks are not informal. Major editorial publications (Wall Street Journal, New York Times, Harvard Business Review) use question-format headings frequently. The question mark signals the structural pattern AI assistants prefer.

Mistake 3: Compressing question structure for "tighter" headings. "AEO Definition" feels tighter than "What is AEO?" but the compressed version loses the structural signal. The 3-word question is the optimization target, not editorial concision.

Mistake 4: Statement openings forced by content management systems. Some CMSs auto-format headings as title case, producing "What Is AEO" instead of "What is AEO?" The capitalization mostly doesn't matter; the missing question mark does. Author the heading with the punctuation that signals the question pattern.

Mistake 5: Confusion about pillar vs supporting article heading conventions. Some authors use statement H2s for pillar articles ("they're more authoritative") and question H2s for supporting articles ("they're more conversational"). The standard treats both article types the same: question H2s throughout.

How do you retroactively fix statement-headed articles?

The retroactive refactor procedure:

Step 1: List every H2 in the article. Document each.

Step 2: For each statement H2, draft 2-3 question-format alternatives. Phrase as a buyer query. Don't simply add a question mark to existing statement headings — restructure to genuine question form.

Step 3: Pick the question alternative that best matches buyer query phrasing. Test mentally: would a buyer literally type this question into ChatGPT?

Step 4: Update the article. Edit the MDX file with the new H2.

Step 5: Re-run Phase 4 audit. Question-format H2 percentage should now be 100%. Score this dimension at full 5/5 points.

The refactor typically takes 15-30 minutes for an article with 6-12 H2s. The lift on overall score is 3-5 points if the original article had statement H2s throughout.

Frequently asked questions

Do question-format H2s help SEO too?

Yes secondarily. Google's "People Also Ask" feature surfaces question-format content directly. Featured snippet eligibility increases for content organized around question-and-answer structures. The AEO/GEO benefits are larger, but the SEO benefits are real.

Does the rule apply to every article, including landing pages?

For editorial content (blog posts, guides, articles), yes. For sales landing pages and conversion-focused pages, the rule is more flexible — landing pages have different optimization goals (conversion rate, not citation) and statement H2s can work fine. The standard is calibrated for editorial AEO content; landing pages are a different surface.

Should every H2 end with a literal question mark?

Strongly preferred. The question mark is part of the structural signal AI parsers and Google's structured-data systems use. Heading "How does X work" without the question mark scores less reliably than "How does X work?" with the question mark.

Can H2s be too question-formatted?

Rare but real. Articles where every H2 follows the same question pattern ("How does X work? / How does Y work? / How does Z work?") can feel monotonous. Vary the question structure across the article — mix "How", "What", "Why", "Where", "When", "Which" to keep editorial rhythm without sacrificing the structural signal.

What about H1 — should the article title be a question?

Often yes. Pillar articles especially benefit from question-format titles. Statement-format titles work fine when they incorporate the buyer query phrasing naturally. The 40-Point Standard scores H2 percentage explicitly; H1 is not separately scored but follows the same logic.

---

Companion guides: How to write content that ChatGPT will cite · The 40-Point AEO Content Geometry Standard · Named entity density · The 134-167 word extractable passage rule · Source links and date stamps.