Retrievability is the probability that an AI engine selects your page into the shortlist of sources it retrieves for a given query, the gate you must pass before anything else can happen. In a retrieval-augmented answer, an engine first fetches a set of candidate documents, then reranks them, then synthesises and cites from what survives. Retrievability is only that first stage: were you pulled in at all?
It matters because it is the stage most GEO advice skips. Metrics like Position-Adjusted Word Count measure prominence, how much of your wording an answer uses once you are already in the context, which is a different thing entirely. A July 2026 critical survey of 45 GEO studies (Olivier Martinez, arXiv 2607.14035) found that citation-oriented rewrites can raise prominence while lowering retrievability, in one test cutting a page's top-10 presence after reranking by about 16%. Optimizing to be quotable can make you harder to fetch.
Read retrievability as the bottleneck, not the finish line. Topical relevance and clean, crawlable structure are its most reproducible levers; clever formatting tuned for the prominence stage is not. Because no engine reports retrievability directly, the practical proxy is your measured cited-pages mix and citation coverage per engine over time, watched for whether a change helped you get selected, not just quoted.