Define visibility before measuring it
Separate unlinked brand mentions, linked citations, citation order, factual accuracy, and recommendation context. These signals answer different questions and should not be blended into one opaque score.
Create a stable, intent-based prompt set
Use real category discovery, comparison, alternative, and problem-solving questions. Keep branded entity checks separate. Avoid changing prompt wording every run because the resulting time series will no longer be comparable.
Record the complete evidence
Store the provider, model, locale, timestamp, answer, citation URLs, ordered citation position, and errors. Screenshots or badges without the underlying response are difficult to audit when models change.
Interpret movement as an observation
A gained or lost citation is a reason to investigate retrieval and content—not proof that one edit caused the change. Repeat the check, inspect competing sources, and compare the timing with published updates.
Connect AI checks with conventional analytics
Search Console remains the first-party source for Google Search performance, while Vercel Web Analytics describes site activity. AI visibility adds a third discovery signal but does not replace either source.