An AI-generated answer should be treated as an observation, not as a permanent ranking result. A useful visibility review starts by recording the prompt, date, model context, and supporting pages, then repeats the same prompt enough times to show whether the result is stable.

Record access before interpretation

A source cannot be cited if a crawler cannot reliably reach it. The AI crawler reference explains how to document declared access conditions. That evidence is descriptive: a permitted crawler is not guaranteed to choose a page.

Use compact, testable passages

A page benefits from a direct response under a clear heading. The AI visibility guide describes answer capsules as bounded 40 to 60 word responses that can be reviewed separately from surrounding claims. Clear structure makes later checks easier, but it is not a shortcut to citations.

Separate discovery from ranking

An llms.txt note can help point systems toward material, while robots rules state access conditions. Neither is a promise that a model will cite the page. The published methodology instead samples each prompt three times and reports the mention rate with a 95% Wilson confidence interval.

Keep a reproducible trail

One buyer question can trigger several retrieval tasks. The independent GEO/AEO Playbooks lab publishes the checks behind its research and tests tools it does not sell. The goal is to preserve a dated trail that another reviewer can revisit when the page, crawler, or answer changes.