Retrieval
Whether your own pages can be found and quoted.
Updated October 20, 2026
Presence tells you where you stand. Retrieval tells you why. It works on your own pages: can a retriever find this page, and if it does, is there a passage on it worth quoting?
Two different failures
A page can fail by being invisible, or by being visible and useless. The first is discovery: nothing links to it, it renders only in JavaScript, it is blocked. The second is extractability: the answer to the question is spread over six paragraphs and a table, so there is no chunk a retriever can lift.
What a score is
A page is scored against one fan-out query at a time. Its whole text is chunked, every chunk is scored, and the best one is kept, on two numbers: relevance, target 105, and n-gram overlap, target 3. Clearing both is Ready, one short is Borderline, neither is Weak. A score is a diagnosis rather than a ranking, and it moves when you change the page rather than when the market does.
The order to work in
Fix discovery first, because a page nothing can reach cannot be rewritten into relevance. Then fix the chunk that should be the answer. Then re-score, and only then look for movement in presence, which lags by a crawl cycle on every assistant.
Concepts rather than clicks: writing pages a model can extract.