Attribution

Attribution is the business of working out where a visitor came from before they arrived. Traditionally it was straightforward: a click from a search result carried a small note saying which page had sent it, and analytics software filed the visit accordingly. That note is what almost all measurement quietly depends on.

AI answers break the arrangement in several ways at once. Most often the person simply reads your name in an answer and never clicks at all — one study of Perplexity found that only twelve to eighteen percent of citations produced a visit, which means that the large majority of times you are named, nothing is recorded anywhere. When people do arrive from an assistant's mobile app, the visit usually turns up stripped of any label and is filed as direct, indistinguishable from someone typing your address from memory. Google's AI Mode removes source information from its links by design.

The consequence is that somewhere between a third and two thirds of AI-driven visits arrive with no origin attached. Dashboards sold to measure AI visibility are therefore reporting confident numbers about something they largely cannot observe. The figures are not fabricated; they are samples and estimates wearing the clothes of precision.

This is not an argument for measuring nothing. It is an argument for keeping your certainty proportionate. Your own website — whether it says plainly what you do, whether a machine can read it at all — you can inspect directly and change this afternoon. How often an assistant names you is, at best, glimpsed. Spending heavily to watch the foggy half while the clear half goes unattended is the common and expensive mistake.

The human translation is word of mouth, which has always had this shape. You never learned how many people were recommended to you at a dinner you did not attend. It mattered enormously and it was never countable, and the correct response was to be worth recommending.