AI & Research
What Is LLM Extractability?
How much of a page's meaning survives when it is read by a language model pipeline rather than rendered in a browser.
Definition
LLM extractability describes how completely and accurately a language model system can recover a page's content and structure from what it actually fetches. It is distinct from whether a page is indexed. A page can rank well in a traditional search engine and still be poorly extractable, usually because its content depends on JavaScript or its structure carries no semantic signal.
Where content gets lost
Most extractability failures fall into a small number of categories, and each one is measurable by comparing the raw HTTP response with the rendered DOM.
- Content that exists only after client side JavaScript executes
- Text rendered inside images or canvas elements with no text alternative
- Primary content buried below navigation, interstitials and boilerplate
- Flat markup with no heading hierarchy, so passages have no boundaries
- Content behind interactions such as tabs and accordions that are not in the DOM until clicked
- Blocked resources or bot rules that stop the fetch before it starts
Why it differs from classic SEO
Google renders JavaScript before indexing. Many of the crawlers that gather content for language model systems do not. This means the effective content of a page can be very different depending on which reader arrives, and the raw HTML response becomes the conservative case worth optimising for.
How to measure it
Fetch the page twice, once as a plain HTTP request with no JavaScript execution and once through a headless browser with scripts enabled. Diff the two. The content present in the second but missing from the first is the content at risk. Scoring that gap across dimensions such as text volume, headings, links and structured data gives a repeatable measure.
A page that a plain HTTP fetch cannot read is a page that many AI systems will never quote, no matter how well it ranks in a browser based index.