AEO Heading Hierarchy Checklist: Structuring H1-H4 for AI Answer Extraction
Use one descriptive H1 (≤ 70 characters), question-style H2s that mirror real queries, focused H3 sub-answers, and H4s only for supporting detail. Never skip levels, never duplicate an H1, and place a 40-60 word answer block directly under each heading. This checklist gives 25 verifiable rules to enforce before publishing.
TL;DR
A clean H1-H4 hierarchy is the single biggest structural lever for AEO. Answer engines parse headings as machine-readable signposts, then extract the first 40-60 words underneath. Run through these 25 checks on every article before it ships.
How to use this checklist
Run each item in order. A failed check marked (blocker) stops the publish; (warning) items are recommended fixes. Each rule maps to a known AI-extraction behaviour observed across Google AI Overviews, Perplexity, ChatGPT Search, and Bing Copilot.
- Scope: one article at a time, pre-publish.
- Owner: content editor (final pass) and writer (first pass).
- Time: ≈ 8 minutes per article.
- Output: a green/red status per item, logged in your CMS or a tracking sheet.
H1 rules (1-6)
1. Exactly one H1 per page (blocker)
The H1 is the page's primary topic signal. Multiple H1s confuse extraction and split topical authority across the document.
2. H1 length is 40-70 characters (warning)
Concise H1s are easier for LLMs to extract verbatim. Aim for 40-70 characters; never exceed 90.
3. H1 contains the focus keyword (blocker)
The focus keyword should appear in the first 60 characters of the H1, in natural language, never stuffed.
4. H1 matches the frontmatter title (blocker)
Your H1, the HTML
5. H1 is text, not an image (blocker)
Wrapping an image in an H1 hides the topic from text-based crawlers and most AI parsers.
6. No filler superlatives in H1 (warning)
Drop "ultimate", "complete", "definitive", and "the best" unless they add real meaning. Answer engines weight clarity over hype.
H2 rules (7-13)
7. H2s phrase the major user questions (blocker)
At least 60% of H2s should be question-form or query-mirroring ("What is X?", "How does X work?", "When should you use X?"). Question H2s align with how users prompt AI engines.
8. Each H2 introduces a self-contained section (blocker)
A reader (or LLM) should be able to lift one H2 plus its body and have it still make sense without the rest of the article.
9. The first sentence under every H2 directly answers the heading (blocker)
No throat-clearing. If the H2 is "What is heading hierarchy?", the very next sentence must define it.
10. Answer block is 40-60 words (warning)
Empirically, AI Overviews, Perplexity, and Bing Copilot tend to extract 40-60 word passages. Under 30 words feels thin; over 80 words gets truncated mid-sentence.
11. H2s use sentence case, not Title Case (warning)
Sentence case mirrors natural query phrasing and is more extractable than Title Case shouting.
12. No duplicate H2 text on the same page (blocker)
Duplicate headings collapse in tables of contents, anchor links, and AI summaries.
13. H2s appear before any H3 (blocker)
Never start a section with an H3. The first sub-heading after H1 must always be an H2.
H3 rules (14-19)
14. H3s only ever live under an H2 (blocker)
H3s are sub-points of an H2. If an H3 floats directly under H1, restructure.
15. H3s answer narrower follow-up questions (warning)
Treat H3s as the second-level questions ("How do I implement it?", "What are the trade-offs?") that branch from the H2's main question.
16. Each H3 has at least one paragraph or list under it (blocker)
Empty H3s are dead extraction points. Either add content or delete the H3.
17. H3s do not exceed 80 characters (warning)
Long H3s get truncated in AI citations and in jump-link UIs.
18. H3s do not introduce new top-level topics (blocker)
If an H3 starts a topic that is not a subset of its parent H2, promote it to an H2.
19. At most 5-7 H3s per H2 (warning)
More than 7 H3s under a single H2 signals that the H2 should be split into multiple H2 sections.
H4 rules (20-22)
20. H4s are reserved for supporting detail (warning)
Use H4 for examples, edge cases, parameter descriptions, or step micro-headings — not for new ideas.
21. H4s only ever live under an H3 (blocker)
Skipping from H2 to H4 breaks the hierarchy and confuses screen readers and AI parsers alike.
22. Avoid H5 and H6 (warning)
If you need H5 or H6, the article is probably trying to do too much. Split it, or convert deep sub-points into a list.
Hierarchy and global rules (23-25)
23. No skipped levels anywhere (blocker)
The sequence H1 → H2 → H3 → H4 is strict on the way down. Going H2 → H4 is invalid.
24. Heading order matches the table of contents (blocker)
Your generated TOC should be a verbatim list of H2s and H3s in document order. If it is not, the markup is wrong.
25. Every heading text is unique within the page (warning)
Unique headings give every section its own anchor and citation surface. Repeated text collapses anchors and reduces citation precision.
Quick reference table
| Level | Purpose | Length | Style | Required? |
|---|---|---|---|---|
| H1 | Page's main topic | 40-70 chars | Sentence case, focus keyword | Exactly 1 |
| H2 | Major sections / user questions | ≤ 80 chars | Question-form preferred | 3+ recommended |
| H3 | Sub-answers to H2 | ≤ 80 chars | Narrower questions | Optional |
| H4 | Supporting detail | ≤ 70 chars | Descriptive | Optional |
Why heading hierarchy drives AEO
Answer engines do not read pages linearly. They chunk content by heading and rank each chunk for extractability. Three behaviours make hierarchy critical:
- Passage ranking. Google, Perplexity, and Bing all rank individual passages, then bubble the strongest passage into the answer. A passage is delimited by its surrounding headings.
- Self-contained extraction. LLMs prefer chunks that make sense without the surrounding article. Question-style H2s plus a 40-60 word direct answer maximise this.
- Topical mapping. The H1-H4 tree tells the engine what the page is about and how its sub-topics relate. A flat or skipped hierarchy flattens this map.
For a deeper dive, see the answer-first content framework and the AI summary block pattern. For pillar context, start at the AEO hub.
Common mistakes to avoid
- Using bold text instead of an H2 or H3. Bold is not a heading; AI parsers ignore it as structure.
- Stacking three H2s with no body content between them as a "table of contents" substitute.
- Repeating the same H2 ("Benefits", "Overview") in multiple places.
- Putting the answer in paragraph three instead of paragraph one of a section.
- Treating headings as design elements (font size + bold) instead of semantic tags.
FAQ
Q: How many H2s should an AEO-optimised article have?
Aim for 4-8 H2s for a 1,200-2,500 word article. Fewer than 3 usually means the article is too thin or under-structured for extraction; more than 10 usually means the H2s should be merged or the article split into two pieces.
Q: Should every H2 be phrased as a question?
Not every one, but at least 60% should be question-form or directly query-mirroring. Mixing question H2s with descriptive H2s (e.g., "Quick reference table", "Common mistakes to avoid") reads naturally and still gives engines plenty of question-shaped extraction surfaces.
Q: Does H1 need to match the tag exactly?
They should be identical or near-identical. Small variations (e.g., adding a brand suffix to
Q: Can I skip H3 and go straight from H2 to H4?
No. Skipping levels breaks the hierarchy for screen readers, accessibility audits, and AI parsers. Always demote or restructure so the levels are contiguous.
Q: Does heading hierarchy still matter if I use schema markup?
Yes. Schema (FAQPage, HowTo, Article) is complementary, not a replacement. Schema tells engines what the page is; heading hierarchy tells them how to read and extract it. Strong AEO uses both.
Related Articles
AEO Anchor Text Phrasing Reference
Reference for AEO anchor text phrasing: how AI engines verbalize citations with 'according to', brand-stem patterns, and reporting-verb selection.
AEO Answer Block Schema Specification: A Markup Standard for Extractable AI Answers
A vendor-neutral specification for an AEO answer block schema using Schema.org Answer plus JSON-LD so generative engines can reliably extract and cite atomic answers.
AEO Callout Box Extraction Patterns for AI Snippet Optimization
A framework for callout boxes (note, warning, tip, info, danger) that AI engines extract cleanly: types, ARIA roles, summary-first patterns, and MDX vs raw HTML.