Header Hierarchy for AEO: Why H2 and H3 Choices Drive Citation Rate
AI models use header hierarchy to navigate articles and extract relevant sections. Well-structured headers drive measurable citation rate improvements.
Key Highlights
- AI models use header hierarchy (H1, H2, H3) to navigate articles and decide which sections to extract for which queries
- Cite-worthy header hierarchy follows three rules: one H1 per page, descriptive H2s that map to buyer questions, and H3s only when the H2 has genuinely sub-divisible content
- The most common header mistake is overusing H2 for visual styling, which fragments the article structure and reduces citation extraction reliability
- Brands that audit and restructure their article header hierarchy typically see citation rate gains within four weeks on the rewritten articles
How AI models use header hierarchy
When an AI model extracts content from a long-form article, it does not read the whole article. It uses the header hierarchy to navigate to the most relevant section for the query. A buyer asking "how do I measure AEO ROI" gets answered by extraction from the section under the H2 "How do I measure AEO ROI" if such a header exists.
This makes header hierarchy a navigation aid for AI, not just a visual style. Articles with clean, query-aligned headers earn citations on a wider query set than articles with the same content under generic or styled-only headers.
The pattern is observable in OnlyAEO's measurement work: equivalent content under different header hierarchies produces measurably different citation rates. Header hierarchy is a small lever with disproportionate impact.
The three rules of cite-worthy header hierarchy
| Rule | What it looks like | Citation effect |
|---|---|---|
| One H1 per page | The article title is the only H1 | Anchors the page's primary topic |
| Descriptive H2s | "How do I measure AEO ROI" not "Measurement" | Lets AI match section to query |
| H3 only when sub-divisible | H3 used to split a genuinely multi-part H2 section | Provides extractable sub-sections without over-fragmenting |
Articles following all three earn citation lift over articles violating any of them. The rules are simple, but most published content violates at least one.
The one H1 rule
Every article should have exactly one H1, which is the article title. The H1 anchors the page's primary topic for AI extraction and confirms the page's role to crawlers.
The common mistake is multiple H1s, often introduced when a CMS treats every "headline" element as H1. Multiple H1s fragment the topic anchoring and reduce extraction reliability.
The fix is CMS configuration plus author discipline. Configure the CMS so the article title is H1 and section headers default to H2. Train authors to use H2, H3, H4 for sections, not H1 again.
Descriptive H2s that map to questions
The single highest-leverage header rule is making H2s descriptive and question-aligned where appropriate.
A section about measuring AEO ROI can have either of these H2s:
Generic: "Measurement"
Descriptive: "How do I measure AEO ROI"
The descriptive version earns citations on the natural-language query directly. The generic version may earn citations through inference, but at meaningfully lower rates. Specificity wins.
Question-form H2s work particularly well for definition queries, how-to queries, and yes-no queries. They work less well for narrative explainers ("Why this matters"). Use question form where the section answers a question. Use declarative form where the section explains an argument.
H3 only when sub-divisible
H3 is overused in most published content. Authors often use H3 to add visual variety or to break up dense paragraphs. The AEO consequence is over-fragmentation: each H3 section becomes a smaller extractable surface that competes with siblings for citation.
The right rule: use H3 only when an H2 section has genuinely sub-divisible content (e.g., three named sub-mechanisms within a broader mechanism, or three named tiers within a broader pricing discussion). When the H2 section is a coherent single argument, no H3 is needed.
A useful test: if removing the H3 and replacing it with bold text or an inline label would not lose anything, the H3 should not exist. Use H3 for genuine sub-structure, not for visual styling.
What over-styled headers do to AEO
A common CMS template applies stylistic emphasis through heading levels (an H2 to make text large, an H3 to make text medium, an H4 to make text small). This treats headers as visual primitives rather than as document structure.
The AEO consequence is that AI models cannot reliably parse the document. A page with an H2 followed by an H4 followed by an H2 has broken hierarchy that confuses extraction. AI models either fall back to text-only extraction (losing the structural signal) or skip the page entirely on ambiguous queries.
The fix is to separate styling from semantics. Use CSS classes for visual variation. Use heading levels for document structure only.
The header content itself
Beyond hierarchy, the content of each header matters. Three patterns work consistently.
The first is keyword inclusion. Each H2 should include the relevant category or topic keyword. "How AEO ROI is measured" outperforms "How it is measured" because the keyword anchors the section to the topic.
The second is parallel construction. H2s within an article should follow a consistent grammatical form (all questions, or all noun phrases, or all imperative). Mixed forms confuse extraction.
The third is appropriate length. H2s of three to ten words extract best. Longer headers get truncated or partially extracted. Shorter headers lose specificity.
Schema and headers
Headers do not require explicit schema markup. The document structure (h2, h3 HTML tags) is the structure AI models use. However, articles with Article schema benefit from clean header hierarchy because Article schema reinforces the article's structural role and headers are the primary navigation within that structure.
The combination of Article schema plus clean header hierarchy produces the most reliable extraction. Either alone produces less.
A four-week header audit and restructure
Week one: identify the top 20 published articles by citation share or pipeline value. Audit each for header hierarchy compliance with the three rules.
Week two: rewrite headers on the top five articles. Apply the three rules. Add Article schema if absent.
Week three: continue the rewrite through the next 10 articles. Two to three per day is a realistic pace.
Week four: republish, rebaseline citation share, and identify the next batch of articles for the same treatment.
Get your free AI visibility audit
OnlyAEO will audit your top articles for header hierarchy compliance, identify the citation gaps, and return a prioritized restructure plan in one week. No commitment.
Get Your Free AuditFrequently Asked Questions
Should every H2 be in question form?+
How many H2s should a typical article have?+
Does the title tag need to match the H1?+
Can table of contents help with AEO?+
How does our CMS template affect header hierarchy?+
OnlyAEO
Expert insights on Answer Engine Optimization and AI visibility strategy.
Related Articles
How to Write a Definition Section That AI Models Quote Verbatim
AI models quote definition sections more often than any other surface on a page. This guide explains the pattern that gets your definition cited verbatim instead of paraphrased.
Read article
Do AI Models Cite Paywalled Content? What the Evidence Shows
Paywalls, crawler access, and licensing shape whether AI models can cite you. Here is what the evidence shows and what brands should keep open.
Read article
How AI Models Choose Which Source to Cite
AI models pick sources using entity clarity, structured answers, corroboration, freshness, and authority. Here are the real signals that earn citations.
Read article