Headings for AI: H1-H6 and the page outline

Ask ChatGPT, Perplexity, or Google's AI Mode a question and the answer is rarely built from a whole page. These systems tend to work from a passage, and retrieval pipelines commonly cut documents at heading boundaries. Google recommends clear headings for its AI search features, but no vendor publishes heading-based splitting rules. That changes what heading structure is for. It was never the ranking trick the checklists sold, and AI answer engines have given it a new job: your H1-H6 is the outline that machines and people both use to find the right part of your page. Get the outline right and each section carries a label that says what it contains.

What heading tags actually are

HTML has six heading levels. According to MDN's heading reference, the <h1> to <h6> elements "represent six levels of section headings", with <h1> "the highest section level" and <h6> "the lowest". The H1 is the page's main heading, the one that says what the whole page is about:

<h1>Main heading of the page</h1>

To be clear about terms: these are HTML tags that structure a web page, not the "Heading 1" style button in Google Docs or Word, and not the section headings of an APA or MLA paper. Together they form the page's outline, and that outline is machine-readable by design. MDN again: "Heading information can be used by user agents to construct a table of contents for a document automatically."

How AI answer engines read your headings

AI assistants do not read your page top to bottom the way a person would. They retrieve. When an engine needs material for an answer, it pulls from an index. Google documents retrieval at the page level; other RAG pipelines often split documents into smaller pieces before retrieval. Google's own documentation defines the technique: "Retrieval-augmented generation (RAG): A technique (also known as grounding) used to improve the quality, accuracy, and freshness of AI responses by relying on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index."

Where the chunks come from

Retrieval systems have to split long documents into pieces before they can index and fetch them, and heading boundaries are a standard place to cut. You can see this in the tooling such systems are built with. LangChain's header splitter, a widely used retrieval component, starts from the observation that "a markdown file is organized by headers" and states plainly: "This will split a markdown file by a specified set of headers". Crucially, "header information is retained in the metadata for each document." In plain language: your H2 and H3 boundaries are where a retrieval system tends to cut your page, and the heading text travels with each chunk as its label. A caveat that applies to everything in this paragraph: this is how retrieval systems are commonly built, not a per-vendor guarantee. Outside Google, no answer engine publishes its splitting rules.

Engine by engine

Google (AI Overviews and AI Mode): documented. Google is the one vendor that says anything concrete. Its AI features guidance recommends organizing content "by paragraphs and sections, along with headings that provide a clear structure to navigate content." It also heads off the tempting overcorrection: "There's no requirement to break your content into tiny pieces for AI to better understand it. Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users." Those two statements fit together cleanly. Do not hand-fragment your page into stub sections; write clear heading boundaries and let the system do the chunking. AI Overviews and AI Mode are Google Search features, so this guidance reaches them.

ChatGPT and OpenAI: not documented. OpenAI's crawler docs cover access, which bots fetch pages and how to allow or block them, and say nothing about how ChatGPT interprets a page's heading structure. So "ChatGPT uses your H1" is an inference from the retrieval mechanism above, not a published rule. The inference is reasonable, but it should be held as one.

Claude and Anthropic: not documented for web pages. Anthropic does write about structure, but in a different setting. Its prompt engineering docs note that "XML tags help Claude parse complex prompts unambiguously". That is about content you send to Claude, prompts and documents you paste in, where structure demonstrably helps the model parse what it is given. Anthropic documents nothing about how Claude's crawler reads your live page. The general lesson, that models handle clearly structured input better, is real; the specific claim about your website is not something Anthropic has made.

Perplexity: not documented. A common piece of GEO advice says to phrase your H2s and H3s as the questions people actually ask, so Perplexity can match a question to a section. No Perplexity documentation states this. The advice is plausible because a question-shaped heading makes a clear chunk label. Its basis is the retrieval mechanism, not a Perplexity statement.

What this means for your page

If retrieval cuts on headings, then each of your sections lives or dies on its own. A section under a clear, self-describing heading is a clean chunk: an engine can lift it, know what it is about from the label, and quote it with confidence. A wall of text under a vague heading like "More information" is the opposite, hard to cut, hard to label, hard to retrieve. The practical rule is simple: give every section a heading that actually describes what the section says, and make sure the section under it delivers on that label.

And for classic search?

The older question is whether headings make you rank, and the answer from the source is no. Google's SEO starter guide says that "from Google Search perspective, it doesn't matter if you're using them out of order" and that "There's also no magical, ideal amount of headings a given page should have." Even Moz's H1 guide, from the industry that spent years teaching H1 optimization, concedes the H1 is not a direct ranking factor.

Headings do feed one visible thing in search: the title link. Google's starter guide explains that the clickable title on a results page is generated from "the words inside the <title> element ... and other headings on the page." A clear H1 can therefore shape how your page is titled in search results, even though it does not move your position.

So there is no ranking boost waiting in heading tweaks. The value of headings is structure and clarity, and that value pays out everywhere at once: in AI answers, search-result titles, ordinary reading, and accessible navigation.

How to structure H1-H6 well

Start with one clear H1. MDN's guidance is direct: "A page should generally have a single <h1> element that describes the content of the page (similar to the document's <title> element)."

Under it, write descriptive subheads: read as a list on their own, they should form a coherent summary of the page, the automatic table of contents MDN describes.

Keep the levels in order. MDN: "Do not skip heading levels: always start from <h1>, followed by <h2> and so on." An H2 followed by an H4 tells software the outline has a hole in it.

Use headings for structure, never for looks. MDN again: "Do not use heading elements to resize text. Instead, use the CSS font-size property." The reverse mistake is just as common: styling a paragraph to look like a heading, which leaves nothing for a machine to see.

Accessibility is where these rules stop being advice and become a standard. MDN notes that "A common navigation technique for users of screen reading software is to quickly jump from heading to heading in order to determine the content of the page." The W3C's page structure tutorial makes the same point from the standards side: "Headings communicate the organization of the content on the page. Web browsers, plug-ins, and assistive technologies can use them to provide in-page navigation." WCAG success criteria 2.4.6 (descriptive headings and labels) and 2.4.10 (section headings) formalize it. A heading structure that works for a screen reader user is, not coincidentally, the same structure that works for a retrieval system.

Can a page have multiple H1s?

Two things are true here, and they are worth keeping separate. First: "using multiple <h1> elements on one page is allowed by the HTML standard (as long as they are not nested)", as MDN's heading documentation puts it, and Google's starter guide, quoted earlier, says there is no ideal amount of headings a page should have. A second H1 does not break your page.

Second: it is still not what you should do. The same MDN documentation says to "Prefer using only one <h1> per page and nest headings without skipping levels". The old pattern of nesting multiple H1s inside sections was once allowed by HTML5's outline algorithm; MDN notes: "However, this was never considered a best practice and is now non-conforming." The conclusion is not "multiple H1s are broken" but "one H1 is the cleaner signal": one unambiguous main heading tells readers, screen readers, and retrieval systems what the page is about, with nothing competing for that role.

H1 vs the title tag

The <title> tag and the H1 do similar jobs from different places. The title lives in the page's <head>: it shows in the browser tab and can become the clickable title in search results. The H1 lives on the page itself, as the heading visitors actually see. They can match word for word or differ, but they should always agree on the topic. Google names both the <title> text and the page's headings among the sources it uses to build the title link, so a title that says one thing and an H1 that says another is a page arguing with itself.

How to check your headings

You do not need any special tool for this. Open the page, view source or open your browser's devtools, and search for <h1. Confirm there is exactly one, and that it is a real <h1> element rather than a paragraph styled to look big.

Then run the outline test. Copy just the headings, in order, into a plain list. If the list reads as a sensible summary of the page, a reader skimming it and a retrieval system chunking it both get a clean map. If it reads as a jumble, the page's structure needs work before its prose does.

The red flags to look for: skipped levels (an H2 jumping to an H4), headings chosen for their font size rather than their place in the outline, and vague labels like "Introduction" or "Overview" that describe nothing and make poor chunk labels.

FAQ

Do AI engines like ChatGPT read your H1?

Google documents using headings to structure content for its AI features; ChatGPT, Claude, and Perplexity document nothing about it. Retrieval systems commonly split pages along heading boundaries, though, so a clear heading structure helps in practice even where no vendor has written it down.

Can a page have more than one H1?

Yes, MDN says multiple <h1> elements are allowed as long as they are not nested, and Google says there is no ideal amount of headings. Best practice is still one clear H1 that states what the page is about.

Do headings help SEO rankings?

Not directly; Google says heading order and amount do not matter for ranking. They help through structure and clarity, and they can feed the title Google shows in search results.

What is the difference between an H1 and the title tag?

The title tag lives in the page's <head> and shows in the browser tab and search results; the H1 is the main heading on the page itself. Both signal the page's topic and should agree.

Should headings be phrased as questions for AI?

It is common GEO advice and reasonable given how retrieval works, but no vendor documents it. A clear, descriptive heading, question or not, is what makes a section easy to retrieve and quote.