How to measure AI visibility

We start with server and CDN access logs, which record when OAI-SearchBot, PerplexityBot, or Googlebot crawls a page. That’s first-party and verifiable. From there we layer in GA4’s AI Assistant channel, share-of-voice tracking, and self-reported attribution, each weighted by how much it can prove.

Because a single score hides which parts of it are provable and which parts are estimates. Share-of-voice tools sample a fraction of possible prompts, and a citation has well under a 1 in 100 chance of reappearing identically across runs. Collapsing that into one number makes a volatile, partial signal look like a settled fact. We show the layers instead, so you can see what’s solid and what’s directional.

No, and any agency who tells you they can is overclaiming. Isolating incrementality, whether a customer would have converted anyway, is the hardest open problem in marketing measurement, and there’s no validated method for doing it in AI search yet. We report what the layers can show and leave out the number nobody can currently prove.

Being crawled means a platform’s bot read the page, which shows up in server logs. Being cited means the platform used that page in an answer it gave someone. A page can be crawled constantly and cited rarely, since most retrieved content is never surfaced in a final response. Crawl logs prove the first. Share-of-voice tools give a directional read on the second.

GA4’s AI Assistant channel depends on the referring platform sending attributable signals with the visit. It’s documented to be blind to Google AI Overviews entirely, and it likely undercounts Perplexity as well, since not every AI-originated visit arrives with a trackable referral. It’s real data, we just don’t treat it as the full picture.