Question-formatted headings
What it checks
AI answer engines directly match user questions to heading text. Question-formatted headings (ending with “?”) are the primary signal AI systems use to identify which section answers a specific query.
Why it matters
H2 and H3 headings can be phrased as the natural-language question a user would ask, rather than as topic labels. The claim is that this increases the probability that the passage beneath the heading is matched to a fan-out sub-query and cited — over and above having descriptive headings at all.
Evidence
Question-form headings
Headings as such are vendor-documented. Google states that pages should be ‘organized by paragraphs and sections, along with headings that provide a clear structure’. And Google’s own definition of query fan-out — ‘a set of concurrent, related queries generated by the model’ — supplies a plausible mechanism for why question-shaped subheads would match sub-queries. GEO-SFE measured macro-structure (heading architecture, target depth 3–5 levels) as 44.9% of a 17.3% citation gain (p<0.001) with a −3.5pp ablation penalty. Indig’s dataset found ‘78.4% of citations tied to questions came from headings’, consistent with H2s acting as prompts and following paragraphs as answers. Semrush’s 200k-keyword study confirms AI Overview queries skew to ‘how’/‘what’ questions (35% desktop, 32% mobile), so question-shaped demand is real.
Limits
No vendor documents question-form headings, and Google explicitly says ‘You don’t need to write in a specific way just for generative AI search.’ The 78.4% figure is a conditional proportion: of the citations that involved questions, most came from headings. It does not establish that converting a label heading to a question heading raises the citation rate, and the base rate is unstated. GEO-SFE tested heading depth and architecture, never question phrasing, so the controlled evidence does not cover this signal.
C-SEO Bench found most authored content edits ineffective or actively harmful to ranking, and the GEO critical survey warns that ‘generic heuristics transfer poorly’ and ‘citation-oriented rewrites can impair retrieval.’ The vendor claims circulating in SEO blogs (‘question headings improve citation within 4–8 weeks’) trace to no measurable methodology. Distinguish two audits: ‘has a sane heading hierarchy’ is scorable; ‘headings are phrased as questions’ is not yet.
Sources
- AI features and your website — AI optimization guide (mythbusting section) — Google Search Central, vendor-doc (verified 2026-08-21)
- Structural Feature Engineering for Generative Engine Optimization: How Content Structure Shapes Citation Behavior — Yu, Yang, Ding, Sato (arXiv, March 2026), study (verified 2026-08-21)
- 44% of ChatGPT citations come from the first third of content: Study — Search Engine Land (reporting Kevin Indig / Growth Memo research), study (verified 2026-08-21)
- We Studied 200,000 AI Overviews: Here’s What We Learned — Semrush, dataset (verified 2026-08-21)
- AI Citation Ranking Factors Analysis — Cyrus Shepard, Zyppy, article (verified 2026-08-20)
- C-SEO Bench: Does Conversational SEO Work? — Puerto, Gubri, Green, Oh, Yun (arXiv; NeurIPS 2025 Datasets & Benchmarks), study (verified 2026-08-20)
- Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023–2026) — arXiv, study (verified 2026-08-21)