FREE TOOL · NOTHING LIKE IT IN FRENCH

GEO extractability score

An AI does not cite a page, it cites a passage. This tool measures six criteria that decide whether yours can be reused — and says which to fix first.


GEO extractability score

Extractability score

0/100


Why the passage matters more than the page

A generative engine does not copy a whole page into its answer. It extracts a fragment of a few sentences, the one that most directly answers the question asked. That fragment is what gets cited, so that is what you should be writing.

In practice this changes how you write: a thousand-word wall of text with no subheadings offers the engine a single indivisible block, hard to reuse. The same content cut into standalone eighty-word passages, each under a heading that poses a question, multiplies the points of entry.

The six criteria, and why these

Length, because too short carries nothing and too long has to be split by the engine. Standalone sense, because a fragment starting with “it” or “this” loses its meaning the moment it leaves its context. The question-shaped heading, because it is the most direct match to a query phrased in natural language.

Verifiable figures, because a generated answer favours what can be checked. Entity naming, because an AI cites the name present in the text, not the one in the logo. And sentence length, because beyond thirty-five words an extract becomes impossible to quote without cutting.

What this tool does not measure

It scores form, not truth or authority. A perfectly structured passage saying something false will score well and will never be cited for long. Content accuracy and domain reputation remain decisive.

It also does not check whether AI crawlers can reach your page. Flawless text on a client-rendered site, or one whose robots.txt blocks GPTBot, stays invisible. That is the technical layer, and it comes before the writing.


Common questions

Where do these criteria come from?

From observing what generative engines actually reuse, and how they split a document. They are not rules published by OpenAI or Google — none of them documents its extraction. Treat the score as a review grid, not a verdict.

Does a score of 100 guarantee a citation?

No. It means form is no longer the obstacle. Domain authority, content accuracy and technical accessibility weigh at least as much, and are outside this tool.

Should every passage score 100?

No. Some passages are narrative and are not meant to be cited. Focus the effort on passages answering a real customer question — those are the ones that can appear in an AI answer.

Is my text sent anywhere?

No. Everything is analysed in your browser, with no network request. You can paste unpublished content into it.


Form is fixed. Whether AI engines read you is another question.

A perfectly extractable passage on a page GPTBot cannot reach will never be cited. Crawler access, server rendering, structured data: that is what AI Visibility covers.