Measuring AI search when the referrer never arrives
Short answer
There are two separate things to measure and neither produces a clean number. Visibility is measured by running a fixed prompt set several times across ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews on a schedule, recording whether you were named, how you were described, what was cited and who appeared instead. Referral traffic is measured in analytics and is structurally undercounted, since Similarweb's July 2026 analysis puts the share of AI referral sessions arriving with no referrer at 35 to 70 percent, and those land in Direct where nothing can recover them. Report the spread rather than averaging it away, because one tidy number for something this unstable is a presentation choice rather than a measurement.
There is no Search Console for ChatGPT. What you can measure honestly, what lands in Direct, and why two visibility tools never agree.
Every AI visibility tool gives you a different number for the same brand, and the reason is not that one of them is broken. Models answer differently each time they are asked, most dashboards ask once, every vendor runs its own prompt list, and ChatGPT, Perplexity and Google AI Overviews retrieve from different indexes, so a blended score hides the variance it was built out of. NovaTechRay reports share of answer with the spread attached and still cannot tell a client which booking came from an assistant, because the guest often reads the name in the answer and then picks up the phone. That gap can be narrowed and not closed. We would rather say so than sell a dashboard that implies we have solved attribution when nobody has.