Heading hierarchy matters more for AI Overview citation than for traditional SEO. AI engines use H2/H3 boundaries to delimit extractable passages — get the structure right and citation eligibility goes up across the page.
AI engines extract passages bounded by heading boundaries. A page with a single H1 followed by 3000 words of body text in one continuous block has exactly one extractable passage candidate — the whole thing — which means the AI engine has to summarize-or-skip. A page with an H1 + 8 well-titled H2s + nested H3s has 8 distinct extractable passage candidates, each with its own micro-relevance score against the query.
The practical effect: well-structured pages get cited at higher rates because they offer the AI engine more discrete passage candidates to choose from. Poorly structured pages either get skipped or get cited generically (low relevance, low click-through).
**H1:** the page's primary query, phrased as a natural-language statement (not a question). Length 5-12 words.
**Lead paragraph:** 60-90 word self-contained answer to the primary query. Factually dense. Citation-extractable as-is.
**H2 sections:** one per follow-up query the primary expands into. Title each H2 as the follow-up query phrased as a natural-language statement. 200-500 words per section. Each section should be independently citable.
**H3 subsections (where useful):** sub-aspects within an H2. Use sparingly — too many H3s fragment passage candidates.
**Closing FAQ section:** structured as FAQPage schema (see /ai-overview-optimization/faq-schema-implementation/).
1. **H1 phrased as a marketing slogan.** AI engines key off H1 for relevance scoring; a slogan H1 hurts scoring.
2. **Multiple H1s on a page.** Confuses passage scoring — use one H1, then H2/H3.
3. **H2s phrased generically ('Our process,' 'Why choose us,' 'About us').** AI engines need topic-specific H2s to score passage relevance.
4. **Skipping levels (H1 → H4).** Breaks passage boundary detection.
5. **HTML headings styled to look like body text (or vice versa).** Visually ambiguous boundaries reduce AI engine confidence in passage extraction.
Depends on content depth. A 1500-word page typically has 5-8 H2s; a 3000-word pillar page typically has 8-15. The right number is whatever lets you cover the query fan-out without each H2 section being trivially short (under 100 words) or unmanageably long (over 700 words).
Mirror the intent, not necessarily the exact wording. Natural phrasing of the underlying question works better than awkward keyword-stuffed H2s. AI engines are fluent at recognizing semantic equivalence.
No — title and H1 are separate. But for AEO, the H1 is more important than the title tag (AI engines extract from page body, not from SERP title).