Your outline is the machine-readable summary of the page
Strip a page to its headings and you should still be able to tell what it covers. That skeleton is what assistive technology navigates by, and increasingly it is what an answer engine uses to decide which passage answers a question - heading as question, the text below it as answer.
What breaks that
- Skipped levels. H2 straight to H4 leaves a hole. The H4 now belongs to nothing, and the tree an extractor builds is wrong.
- Multiple H1s. Legal in HTML5, still ambiguous. One page, one statement of what it is about.
- Thin sections. Under about 40 words and the heading is decoration - it promises an answer the section does not contain.
- Bloated sections. Over 400 words and the answer is buried somewhere in the middle, which is exactly what does not get extracted.
- Headings used for styling. An H3 chosen because it looked the right size is invisible to a reader and misleading to a parser.
The 120-180 word band
That range is what our GEO scorer measures section depth against. It is enough to answer a question properly and short enough that the answer stays findable. It is a guide, not a rule - but a page where most sections sit far outside it usually has a structural problem rather than a length problem.
Raw HTML only
This reads what your server returns, before JavaScript. If a page shows no headings here but plenty in a browser, that gap is the finding: AI crawlers do not run JavaScript, so the outline they build is the empty one.
For the full picture, run the AI visibility checker, which scores structure alongside nine other dimensions.