STR-04
Long stretch of text with no structure
What the check measures
The page is cut into runs of text and the words in each run are counted. The finding is raised when at least one run exceeds 600 words, roughly two and a half pages without a single break. The threshold lives in config/ under rule_ as max_ and comes from the product specification; the finding text prints the measured run length next to it.
Heading levels are irrelevant. The check does not care whether a run is ended by an H2 or an H4, only whether there is more text between two adjacent dividers (or from the last one to the end of the page) than the threshold allows. A page with no dividers at all is one run from top to bottom.
A run is ended by an h1 to h6 heading, by the question of a collapsible list (<summary>) and by the term of a definition list (<dt>). Twelve collapsible questions or a glossary with fourteen entries are therefore not treated as undivided text, even though neither has a heading. Site chrome, that is navigation, header and footer, is not counted into the run either. Both live in config/ and both were added on 2026-09-19, when production data showed nine findings out of thirteen reporting “continuous text with no subheading” on content that was in fact divided.
What the check does not do:
- It recognises no structure that has no markup of its own. Paragraphs, bullet lists, tables and horizontal rules do not count. A well-
paragraphed 700-word passage still gets the finding. - It does not judge whether a subheading belongs there. A long continuous argument may be written correctly, and chopping it up with headings would make it worse.
- It ignores language. In languages that glue words together, the same amount of text counts as fewer words.
The finding is page-level and informational, not a warning. It costs no score at all. For a check that measures only length, that is as it should be.
How strong the evidence is
We recommend it because it does no harm or has some other benefit, but we promise nothing about whether it makes language models cite you. Nobody has demonstrated that yet.
We have no evidence that breaking long text up with subheadings improves your chances of being cited. No operator declares it and we know of no measurement comparing the two. Hence the “effect not demonstrated” class.
The reasoning behind the recommendation runs as follows, and it is worth knowing so you can form your own view: when an assistant has to answer a specific question, it needs to lift a chunk that stands on its own out of your page. Text where two thousand words roll on without a single break offers no such boundary. It is a hypothesis derived from how long documents are generally processed, not a measurement of your case.
What is documented in this area concerns the substance of the text, not its layout. The Princeton experiment (KDD 2024, 10,000 queries) measured gains in visibility from quotations (+41 %), concrete statistics (+33 %) and links to sources (+28 %). Section length was not among the factors measured at all.
The practical conclusion: treat this finding as an editorial note, not a technical defect. A long airless block of text mainly puts people off, which is a reason in itself, just not the reason you bought an audit for. If the text is written so that a continuous argument cannot be split, ignore the finding. That is why it is informational.
How to fix it
First read the page and ask whether that long run has internal structure you simply never marked up. Usually it does. Then all you need is subheadings where the breaks already are:
- A subheading every 200 to 400 words reads comfortably. It is not a rule, it is an observation.
- A subheading should name what is under it, not just separate text decoratively. “How long processing takes” is a subheading; “Next” is not.
- Keep the level matched to the nesting, not to the font size; otherwise you will manufacture a
STR-01.
When the text genuinely has no internal structure because it is one continuous argument, a different edit helps. It often works better than artificial headings:
- Pull a summary to the top. Three bullet points of conclusions ahead of a long argument do more for comprehension than five subheadings in the middle.
- Break enumerations out into a list. A sentence listing six things is almost always a list that decided not to be one.
- Consider splitting into two pages. When the text covers two subjects, it is two pages, and each can then be found and linked on its own.
And one thing not to do: do not insert subheadings just to clear the finding. A page with seven headings reading “Continued” is worse than one without them, and we will not praise anyone for it.
What the report says about it
Finding description
The longest continuous stretch of text on this page is [words] words; our threshold is [threshold]. We count stretches between dividing elements, which are headings h1 to h6, the questions of a collapsible list (`<summary>`) and the terms of a definition list (`<dt>`). Content divided some other way, such as twelve collapsible questions or a list of glossary entries, is therefore not treated as undivided. Navigation, header and footer are not counted into the stretch. A long undivided block is harder for readers and for AI systems to work through, and it is harder to pick out the answer to a specific question; we have no measurement to back that up.
Recommendation
Split the long stretch into smaller chunks with their own subheadings (H2/H3), each covering one idea or sub-topic. Adding subheadings just to make the finding go away is pointless: a continuous argument that cannot be split is fine as it is.
Sources
- Princeton: GEO: Generative Engine Optimization (arXiv 2311.09735, KDD 2024) (accessed 2026-08-19)
- Google Search Central: SEO Starter Guide (accessed 2026-09-12)
Text verified 2026-09-12