Heading Structure (H1–H6): The Rules for SEO, Accessibility and LLM Answers
Headings are the cheapest structural improvement in on-page work: no new content, no redesign, an hour of your time. Done well they help search engines, screen-reader users, and the AI systems that increasingly summarize and cite web pages. Done badly they quietly undermine all three. Here are the rules, the reasoning behind each, and the mistakes that survive even on professional sites.
The one rule that matters most: exactly one H1
Every page gets exactly one H1 — the page's single main topic. Two H1s split the page's semantic identity: search engines get two competing "this is what the page is about" signals, and screen readers building the page outline see two equal-level sections where there is really one page. The practical test: if you cannot pick one heading as the page's topic, you have a page-splitting problem, not a heading problem.
The common failure modes: a site banner or logo set as H1 (it is navigation, not content), a "Welcome to [Brand]" H1 that says nothing about the actual page, and the H1 hidden in CSS while a smaller styled heading says the real thing. The H1 should be the heading a person would say if asked "what is this page about?" in one line.
Hierarchy: never skip a level
The document outline is the contract. An H3 must sit under an H2 that sits under the H1. Skipping from H1 to H3 produces an outline with a missing rung — assistive technology users navigating by headings land on a section with no parent, and crawlers building the page structure get the same broken tree. The fix is not to "fix the level numbers" — it is to fix the actual structure: either the section deserves a parent (add the H2) or it is not a sub-section (demote it).
Heading levels exist to express subordination. If two headings are at the same level, they are siblings. If one is a level deeper, it is a child. Write the outline in plain text before you write the body — the outline is the article.
Keywords in headings: the current, defensible position
The old "stuff the keyword in every H2" era is over — and it is worth saying plainly, because a lot of outdated advice still circulates. The defensible position in 2026:
- H1: include the primary topic phrasing naturally. "Merge PDF Online Free" is a real H1; "PDF Merge — PDF Merge Tool — Merge PDFs Online Free Merge" is not a page, it is a list.
- H2s: the main sections, phrased as questions or outcomes where the audience searches that way. "How to compress without visible quality loss" is a better H2 than "Image Compression Techniques."
- Beyond that: stop. A heading is a navigation label for the reader first. Keyword intent is served by the page's coverage and the H1; the deeper headings should sound like a table of contents a human would write.
Why LLMs make headings matter more, not less
Answer engines and summarization systems pull structure from HTML the same way crawlers do: headings define the sections a model can cite. A page with a clean H1 → H2 → H3 outline is directly quotable — "According to the 'Fix the dimensions first' section…" — while a page with five H1s and no sub-structure offers the model nothing but a paragraph soup. You do not write "for AI." You write with structure because it is good for humans, and the AI readers inherit the benefit. The two audiences reward the same behaviour.
A worked structure for a 1,000-word article
H1 How to Compress an Image Without Losing Visible Quality ├─ H2 What "compression" actually does ├─ H2 Step 1 — fix the dimensions ├─ H2 Step 2 — pick the format for the image type │ └─ H3 (only if you genuinely need a sub-case) ├─ H2 Step 3 — set the quality, watch the size curve ├─ H2 Verify before you publish └─ H2 FAQ
Notice the shape: the H2s are the reader's path (understand → steps → verify), and the H3 appears only where a real sub-case exists. No level is used for styling. No heading is a keyword list. The outline reads as the article's skeleton — because it is.
The accessibility side (this is not optional)
Screen-reader users navigate long pages by heading — they do not read every paragraph, they jump section to section. A page with no headings, or one heading, is effectively unreadable to them at length; it is a wall of text with no way to scan. Semantic headings (real <h2> elements, not bolded paragraphs) are also what keyboard users and outline tools rely on. So "heading structure" is not an SEO box — it is the document's table of contents for a real class of readers. WCAG treats heading structure as a requirement, and it is one of the few that is trivial to satisfy correctly.
Five mistakes to check on every page
- Two or more H1s — including one in the header/logo.
- No H1 at all — the page starts at H2, common on template-based CMS pages.
- Level skips — H1 → H3 with no H2 parent.
- Headings as styling — an H4 used because it looks the right size; use CSS for size, headings for structure.
- Empty or orphan headings — an H2 with nothing under it before the next heading; the outline promises a section that does not exist.
The reference
The MDN documentation on the heading elements — MDN: section and heading semantics — is the clearest short statement of how heading levels define the document outline and why level-skips break it.
FAQ
- Can I have an H1 in my site header? No — the header belongs to every page, so an H1 there duplicates on every page and dilutes the per-page H1. Brand and nav are not headings; they are navigation.
- How many H2s per page is "too many"? There is no hard limit — ten H2s on a 3,000-word guide is normal. The constraint is that each one should be a real section a reader would navigate to.
- Do heading tags affect rankings directly? They contribute to how the page is understood and to the user/assistive experience that rankings reflect. Treat them as structure, not as a keyword slot.
- Where do I audit my existing pages? View source (or inspect) and read the outline top to bottom. The Alt Text Checker is the natural companion audit — headings and alt text are the two semantic signals that take an hour to fix and pay for years.