Header Tags: How to Structure
a Page for Readers and Retrieval

Header tags are the six HTML elements (H1 through H6) that mark the headings on a page. These tags carry a rank of the heading within that page. H1 names the whole document, H2 opens a division of it, H3 opens a division of that, and so on down to H6. Browsers render them at descending sizes by default, which is why people often confuse them for text sizes, though the size is a style choice layered on top of the structure.

That structure decides how a page gets divided, and the divisions are what gets lifted out of it. An AI-generated response, a featured snippet, and a People Also Ask answer all return a section rather than a page. Heading tags draw the lines those systems cut along.

Getting the hierarchy right touches how the page reads, how it gets indexed, and how it gets pulled apart by the systems built on top of it. Headings sit alongside the other on-page SEO fundamentals: title tags, meta descriptions, internal links, and the rest of the elements a page owner controls directly. The rest of this guide covers what matters and what doesn’t.

Key Takeaways

• The tag level matters less than the words inside it. Moz’s published eight-week experiment from 2020, that switched blog headlines from H2 to H1, found no ranking change separable from noise, while rewriting a vague heading into a descriptive one changes which queries a page can answer.
• Heading order does not affect Google’s rankings, and it affects two things that do matter. Google says so directly. Screen readers navigate by heading rank, and retrieval pipelines use the heading path to label what each piece of your page is about.
• One H1 per page is a writing discipline. Google handles multiple H1s without trouble. The reason to use one is that a page with three top-level titles has not decided what it is about.
• Headings are where retrieval systems cut. Every section should still make sense if someone lifts it out of the page and reads it alone, because that is what snippet and answer engines do.
• A heading should say what the reader will find there. “Pricing” tells a reader nothing. “What the plan costs at ten seats” tells them whether to stop scrolling.
• The pronoun in your first sentence is a retrieval bug. A section that opens with “This works because…” loses its subject the moment it is extracted.
• Audit by consequence, not by crawler warning. A missing H1 and a site-wide duplicate H1 are worth your morning. A crawler flagging an H1 as four characters too long is not.

What are Header Tags?

Three separate things share the word “header” in HTML, and they do unrelated jobs. The confusion is widespread enough that searching for header tags returns developer documentation and SEO advice side by side, answering different questions.

The <head> is a metadata container. Nothing inside it appears on the page. It holds the title tag, the meta description, canonical tags, and script and stylesheet references.

The <header> is a container for introductory content. Per MDN, it usually holds a logo, a search form, an author byline, or navigation. Its meaning changes with position. As a direct child of <body> it becomes a banner landmark that assistive technology can jump to. Nested inside an <article> or <section> it has no landmark role and simply groups that section’s opening material. It cannot contain another <header> or a <footer>. It often contains a heading, and it is never a heading itself.

The <h1> through <h6> elements are the headings. When SEO writing says “header tags,” “heading tags,” or “H tags,” it means these six, and so does the rest of this article.

A developer told to “fix the header tags” may go and work on the site banner while the person who asked meant the H1 on 400 product pages. That’s at least one practical reason why getting this straight matters.

H1-H6 and What Each One Is For

The six levels describe depth. Each one opens a division of the level above it, and a page uses as many as its structure needs and no more.

Almost all web writing stops at H3, and that is where it should stop. A blog post, a landing page, a service page, or a product page rarely carries an argument that nests four levels deep. A page that does is usually a page that should have been two different pages.

H4 earns its place in technical documentation. API references, technical specifications, and long procedural guides have genuine four-level depth, running from a topic, to a set of operations, to one operation, to the parameters inside it. Compliance and legal documents have the same shape, so if you write such documents, use H4 without hesitation and do not read it as a warning sign.

H5 and H6 exist because the scale needed a bottom. Reaching for one is worth a pause to ask if the document is doing the work of three documents.

The levels carry no descending weight in search. An H3 does not count for less than an H2 because its number is higher, and burying a section one level deeper does not hide it. The rank just describes where a section sits in the outline,.

What Headings Do for a Page

Headings have four primary functions, and they are worth separating because the evidence behind each one is a bit different.

They let a reader scan. Nobody reads a 3,000-word page in order. They skim the headings, find the part that answers their question, and start reading there. A page with no signposts forces a choice between reading all of it and leaving.

They tell a search engine what each part covers. Google indexes passages as well as whole pages, which lets a single section surface for a query the rest of the page never addresses. Headings are the clearest signal of where one passage ends and the next begins.

They create extraction targets. A question posed as a heading with a direct answer underneath is the shape that featured snippets, PAA boxes, and AI answers pull from. More on this below, because it is where the interesting work is.

They give screen reader users a table of contents. That one is a verifiable fact about software rather than a claim about algorithms, and it gets its own section further down.

The ranking question, answered honestly

Two well-sourced claims about headings and rankings contradict each other, and both circulate freely.

On one side, John Mueller has described heading elements as a really strong signal that helps Google understand what a page covers. On the other, a published eight-week experiment switched every post headline on a large blog from H2 to H1 and measured a ranking difference indistinguishable from noise.

Both results hold, because they measure different things. Changing which tag wraps a heading changes nothing, since Google can work out what the headline of a blog post is without the markup telling it. Changing what the heading says changes the topics the page is understood to cover, and that can change which queries it surfaces for.

The conclusion for your Monday morning is that promoting H2s to H1s across your blog is busywork. Rewriting twelve headings on a page sitting at position nine is not.

Accessibility is the effect you can verify

Screen readers build a list of every heading on a page and let the user pull it up and jump straight to any of them. In NVDA and JAWS it is the elements list. In VoiceOver it is the rotor. For someone who cannot see the layout, that list is the page’s navigation.

The list is built from heading rank. When a page goes H1, then H2, then jumps to H4, the software reports a level with no parent, and the user has to guess whether they have moved into a subsection or missed something. When a designer bolds a line of 18px text instead of using a heading element, that section does not appear in the list at all.

This is the one heading effect you can test in half a minute, on your own machine, with no argument about correlation.

The Heading Hierarchy Rules That Matter

Three hierarchy rules get repeated everywhere. Two are worth following, and the reasoning usually attached to them is wrong in a way that matters once you have to decide what to fix first.

There must be only one H1

In 2019, Google has said plainly that multiple H1s cause no problem for Search. The old advice treating a second H1 as a technical fault is out of date.

Use one anyway. A page with three top-level headings is usually a page that has not settled on a single topic, and that indecision shows up in the copy long before it shows up in any report. The H1 should name what the page is about, once, at the top of the content.

The related question is whether the H1 should match the title tag. They should say the same thing without being identical. A title tag has to survive truncation in a results page, so it stays tight and often carries the brand. An H1 has the full width of the page and can be longer and warmer. Our guide to title tag optimization covers where the two diverge and why the title is the one worth testing.

Skipped levels are an accessibility problem

The reasoning usually attached to this rule is backwards. Google’s SEO Starter Guide addresses heading order directly in its section on SEO myths: “Having your headings in semantic order is fantastic for screen readers, but from Google Search perspective, it doesn’t matter if you’re using them out of order.” The same passage adds that there is no ideal number of headings for a page.

So the rule survives and the justification changes. Keep your levels in order because a broken rank sequence breaks the navigation index described above, and because a hierarchy that makes sense to software usually reflects an outline that makes sense to a reader. Do not keep them in order because you expect a ranking change, and do not spend a sprint on skipped levels while the H1s on your money pages are missing.

Pick the level, then style it

The most common heading fault has nothing to do with SEO. It is a writer picking H3 because H2 looks too big in the editor.

Pick the level that describes the relationship, then change the size in CSS. A section sitting beneath your third H2 is an H3 whether it looks right or not. Teams that hand this to a designer once and settle it in the stylesheet stop having the argument.

The same applies to the reverse fault. Bolded text is not a heading. It looks like one to a sighted reader and is invisible to everything else.

How many headings are too many?

There is no number, and Google says so. A usable habit is that a reader should not scroll more than a screen or two without a signpost, which works out around every 250 to 350 words for most writing. Denser than that and the page starts to feel like a slide deck. Thinner and it turns into a wall.

How to Write a Heading Worth Reading

The mechanics above take an afternoon to fix. What the headings say is the part that repays attention.

Descriptive beats categorical

Compare two versions of the same section heading. “Pricing” is a category label. “What the plan costs at ten seats” is a description of what the reader will find. The second works as a scanning target, as a passage boundary, and as a snippet candidate. The first works as a filing label.

The test is whether someone could act on the heading without reading the paragraph under it. If the answer is no, the heading is a label.

Length is about scanning

Around 60-70 characters is the range most guides settle on, and the reason is legibility rather than any limit Google applies. Headings wrap, so there is no truncation cliff the way there is with a title tag. A heading that runs to two and a half lines has stopped being a signpost and started being a sentence.

Front-load the part that carries meaning, because a scanning reader takes in the first three words of a heading and decides from those.

Put the term where it helps, once

Your target term belongs in the H1. Related terms, variations, and the questions people type belong in the H2s and H3s that cover those subtopics.

What does not work is repeating the same term in nine consecutive headings. It reads as machine-written to a person and adds nothing for a search engine, which understood the topic from the first two instances. Our guide to keyword research covers pulling the question phrasings worth building headings around.

Ask the question, then answer it

Headings phrased as questions match how people search, and they set up the answer-first structure that extraction systems reward. “How long should a heading be” and “Can I use two H1s” are headings a query can match directly.

The limit is tone. A page where every heading is a question reads like a quiz. Mix question headings into a structure that is otherwise declarative, and save them for the sections where the reader genuinely arrives with a question.

Read only the headings

Copy every heading out of your draft into a blank document and read them in order, with none of the body copy.

If that outline tells a coherent story, the structure is sound. If it reads as a list of loosely related nouns, the page has sections rather than an argument, and no amount of work on the paragraphs will fix that. This takes two minutes and catches more structural problems than any crawler will.

Headings as Retrieval Units

With the invention and widespread use of LLMs, the heading structure matters more now than it did five years ago.

Search engines and AI systems do not only evaluate whole pages. They break documents into passages, score the passages, and return the ones that answer the query. Google’s passage indexing does this inside Search. The retrieval systems behind AI answers do a version of the same thing, splitting documents into chunks before matching them to a question. In both cases, heading boundaries are the most reliable place to cut, because they are the only structural marker that dependably signals a change of subject.

So the heading structure is less a summary of your page than a set of instructions for how it gets taken apart.

Every section should survive being lifted

The working test: take any section of your page, paste it into a blank document with its heading, and read it as though you had never seen the rest of the article.

Does it still make sense? Does it name the thing it is talking about? Would someone arriving at this section cold get a usable answer? Sections that pass are the ones that can be quoted in an AI answer or pulled into a snippet. Sections that fail are the ones that depended on context they no longer have.

This is a decent editing test for its own sake. A section that cannot stand alone is often a section that has not decided what it is for.

The pronoun problem

The most common way a section fails that test is the opening sentence.

“This approach works because it keeps the crawl budget down.” Which approach? The one named in the previous section, which is not in the chunk. “It is also the faster option.” Faster than the choice described two sections earlier, which the reader of this chunk never saw.

Anaphora, the habit of pointing back at something already mentioned, is normal writing and it reads well in sequence. It falls apart the moment a section is extracted, because the referent stays behind on the page. The fix takes one word. Name the subject in the first sentence under every heading, then use pronouns freely for the rest of the section.

Answer first, then expand

Under a heading that poses a question, give the answer in the first 40 to 60 words. Then explain, qualify, and give the exceptions.

This is the shape a paragraph snippet is pulled in, and the shape a retrieval system finds easiest to score. Burying the answer in the fourth paragraph, after three paragraphs of context, gives an extraction system nothing to take and gives a reader a reason to leave.

The objection is that answering immediately removes the reason to keep reading. It does not. Readers who get a direct answer stay for the qualifications more often than readers made to hunt for it, and the ones who leave satisfied were never going to read the whole section anyway.

Mirror how the question gets asked

Queries typed into an assistant are longer and more conversational than queries typed into a search box. People ask full questions there, in full sentences.

A heading that mirrors that phrasing has a better chance of being matched to it. “Can I use more than one H1 on a page” is closer to how someone asks than “H1 quantity.” Whether that visibility is holding is worth watching alongside your rankings, and our guide to search visibility covers the metrics that move first.

Heading Patterns by Page Type

The job changes with the template. Blog-post heading habits applied to a product page produce a page that reads like an essay about a product.

Two patterns to avoid across all of them. Do not use the site or brand name as an H1, which wastes the one heading that describes the page. And do not let template furniture, a “Related products” or “Newsletter” heading, sit at a higher rank than the content headings, because the outline then says the newsletter is a peer of the page topic.

How to Audit Your Headings

Crawl the site, then fix in order of consequence rather than in the order the tool flagged things.

What to fix first

1. Missing H1. The page has no stated topic. Fix these first and fix them all.
2. Template boilerplate as the H1. The site name, a tagline, or “Welcome” sitting in the H1 slot across a template. One change fixes every page using it.
3. The same H1 on hundreds of pages. Usually a category or filter template that never received a dynamic value. Same fix, same payoff.
4. Label headings on pages that get impressions but no clicks. Pull those pages from Search Console, open them, and read the headings alone. This is where rewriting pays.
5. Skipped levels. Real, worth fixing, and an accessibility fix rather than a ranking one. Batch it.
6. Multiple H1s. Do these last. It is a tidiness issue on a page that otherwise works.

The tools worth using

Screaming Frog exports H1 and H2 columns with filters for missing, duplicate, and multiple, which covers items one through three and six in a single crawl. For a single page, a browser extension like HeadingsMap renders the outline as a tree and makes a skipped level obvious at a glance. Chrome’s accessibility tree in DevTools shows the same structure without an install.

For the one check nobody runs, turn on VoiceOver or NVDA and open the heading list on your most important page. Ten minutes of that is worth more than a spreadsheet of warnings. Our technical SEO guide covers fitting a crawl like this into a wider audit.

What to ignore

Crawlers flag heading issues that are not issues. An H1 over 70 characters is a style note. A page with eleven H2s is fine if it has eleven sections. An H1 that does not contain the exact target keyword is not an error, since the heading has to describe the page, and matching phrasing is a means to that rather than the goal.

Resist the site-wide heading rewrite. Tie the work to pages that already draw impressions, treat it as a batch, and record what you changed so the result means something later. A recurring content audit is the natural home for it, since headings go stale alongside the content beneath them.

Common Heading Mistakes

• Labels where descriptions belong. “Benefits,” “Features,” “Overview.” Each one tells the reader that something is coming without telling them what.
• Openers that delay the subject. “Understanding How Heading Structure Works” spends four words before it says anything. “Heading Structure, Explained” says the same and gets out of the way.
• The same keyword in every heading. It signals effort rather than relevance, and it makes the outline unreadable.
• Bold text standing in for a heading. Invisible to the heading list, invisible to passage segmentation, and it usually happens because the H3 style looked wrong.
• Headings chosen by size. Covered above and worth repeating, because it is the most common of all of these and the easiest to settle once.
• An H1 copied verbatim from the title tag. Not harmful, and a missed opportunity. The title tag is written for a crowded results page. The H1 is read by someone who has already clicked and has more room to work with.
• Navigation headings above the content in the DOM. A sidebar H2 that appears before the H1 in source order puts the outline in the wrong sequence for anyone reading the page structurally.

Conclusion

Heading tags have been discussed for twenty years as a markup question, which is why the advice about them has calcified into rules about counting H1s and avoiding skipped levels. Those rules are fine. They are also the least interesting thing about headings, and following them perfectly buys you close to nothing.

What pays is treating the heading structure as an argument. The outline is the page’s reasoning made visible, and if it does not hold together when read on its own, no amount of markup hygiene will save the page. That was always true for a reader scanning for the part they needed, and it is more true now that snippet engines and AI systems cut pages along heading lines and hand a single section to someone who will never see the rest.

The question worth asking is whether each section of your page could be torn out, read cold by a stranger, and still be worth something. Write headings that describe rather than label, name your subject in the first sentence beneath them, and read your outline alone before you publish. The markup rules take an afternoon. The rest is writing.

Work With Us

Heading structure is the kind of work that gets deferred because nothing breaks when you skip it, then quietly caps what a page can do. The Growth Spice handles content, link building, and SEO for teams that would rather have the structural work done properly than added to a backlog. If your pages are ranking below what the content deserves, get in touch and we will take a look.

Frequently Asked Questions (FAQ)