What it measures

Before an AI system learns from a page or quotes it, a text extractor strips the page down to what it judges to be the main content. Navigation, footers and ads go. Sometimes the content goes with them.

This check runs the same kind of extractor on your page. It's trafilatura, the tool large AI training datasets such as FineWeb were built with. It then compares what the extractor keeps with your page's main text: everything on the page except the header, navigation, footer, sidebars, forms and cookie banners. The result is the share of your main text that survives.

Why it matters for AI

Extractors are tuned for articles. They look for paragraphs inside a recognizable content container, and they often skip layouts built from repeated blocks: topic and product cards, listicle entries, image captions. On one top-10 list we tested, the extractor kept the 115-word introduction and dropped all ten reviews, about 1,200 words. Text that doesn't make it through extraction can't be learned from or quoted, however good it is.

When extractors keep less than 20% of your main text, GenReady flags the page and scores your content on the full main text instead, so the content score reflects what you wrote. It also takes up to 10 points off the crawlability score, because most of what the page says never reaches these systems.

How to improve

1. Put each item (a review, a card, a product) in its own <article> or <section> with a heading and real <p> paragraphs, all inside <main>.
2. Link a card's title instead of wrapping the whole card in one link. Extractors treat text that sits mostly inside links as navigation.
3. Give list and category pages a short written introduction that says what the page holds.
4. Re-run the analysis after a change and compare the "How AI sees your site" section, which shows exactly what the extractor kept.