Stage 1: access
Confirm the assistant crawlers you want are allowed, that pages return real HTML, and that response times are low. Silent blocks in robots.txt, WAF rules, and bot-protection defaults are the most common cause of total absence.
Stage 2: retrieval
Retrieval is passage-based. Break content into labelled, self-contained chunks of 80-200 words with descriptive headings so the right span can be found without reading the whole page.
Stage 3: synthesis
Synthesis keeps what it cannot regenerate. Original numbers, dated tests, named examples, and explicit trade-offs survive; generic advice is merged into the consensus and loses its attribution.
Entity strength decides naming
Engines cite sources they can identify. Consistent organization data, a real About page, named authors, and third-party mentions determine whether you are named or paraphrased anonymously.
Corroboration compounds
The same claim appearing on independent sites raises confidence. Digital PR, documentation, community answers, and review platforms all feed this, and none require paid tooling.
Run a fixed measurement loop
One frozen prompt set, three runs per prompt, one schedule, one log. Everything else in AI visibility reporting is anecdote.