One H1 and a hierarchy: how retrieval chunks your page

Pages are too long to quote

A model answering a question does not paste your whole page into its context. A retrieval step splits the extracted content into chunks, scores each chunk against the question, and passes the best few along. The quote you get is one chunk, and the chunk carries whatever context the splitter gave it.

Splitters prefer to cut at headings, because a heading is a statement that a new idea starts here. A well-headed page yields chunks that each make sense alone: 'Refund policy: 30 days, unopened items only'. A page with one heading and forty paragraphs yields chunks cut at arbitrary lengths, so the passage about refunds may begin mid-sentence and lose the word 'unopened'.

What Qomvia checks

The Heading structure is usable check in the methodology counts <h1> and <h2> elements on a content page. Exactly one <h1> and at least one <h2> passes. One <h1> with no section headings is partial: the topic is clear but the page will be chunked blind. No <h1> fails. Several <h1> elements score low, because the main topic is ambiguous and the extractor has to pick one.

The multiple-H1 case is the most common on real sites and usually accidental: a logo wrapped in <h1>, a page builder that gives every section its own <h1>, or a component library that renders card titles as <h1>.

Rules that hold up

• One `<h1>`, and it names the page's subject. Not the site, not 'Welcome'. It should be a phrase someone could type as a question.
• `<h2>` for every section a reader might land on. If a section can answer a question on its own, give it a heading that states what it answers.
• Do not skip levels. <h2> then <h4> tells the splitter a level went missing. Use <h3> under <h2>.
• Style is not structure. A large bold <p> is a paragraph to an extractor. If it acts as a heading, mark it as one.
• Headings are not for layout. The logo, the nav label and 'Related posts' are not headings of the content. Use <p> or <span> and CSS.

[...document.querySelectorAll("h1,h2,h3,h4,h5,h6")]
  .map(h => h.tagName.toLowerCase() + "  " + h.textContent.trim().slice(0, 60))
  .join("\n")

Read the output as a table of contents. If it does not describe the page to a person who has not seen it, it will not describe it to a retrieval system either.

Headings, landmarks and FAQs

Headings do their work inside the block an extractor kept, which is why the landmarks article comes first. A FAQ section is the same idea taken to its conclusion: each question is a heading, each answer is one self-contained chunk, and FAQPage structured data makes the pairing explicit.

Questions

HTML5 allows multiple H1s in sectioning elements. Why penalise them?

The outline algorithm that would have made that meaningful was never implemented by browsers or extractors. In practice every reader treats the first H1 as the page topic and additional H1s as noise.

Should headings contain keywords?

They should contain the words a person would use to ask about that section. That is usually the same thing, without the stuffing.

Do headings inside a product description matter?

Yes. 'Materials', 'Dimensions', 'Care' as H2 or H3 make each specification a retrievable passage instead of a line in a wall of text.