Why "zero-click" is the wrong metric: measuring influence, not visits
The most quoted number in AI search is how rarely anyone clicks. It deserves the attention: the rates are real and they are enormous. But the panic it produces points at the wrong problem. The clicks that disappeared were never the value. They were a proxy for it, and in AI search the proxy has broken in both directions at once: fewer visits, and visits that are worth far more each.
How bad the click math actually is
The cross-engine picture was assembled in a May 2026 benchmark that compiled zero-click rates per engine, and the spread is tight enough to treat as a property of the category rather than a quirk of any one product:
Layer Pew Research Center's browsing-panel data on top: when an AI summary is present on Google, visits with a click on any result drop from 15% to 8%, and clicks on the citations inside the summary run at roughly 1% of visits. Classic search was already majority zero-click. AI search is zero-click by design.
The half of the story the panic leaves out
Here is the number that never makes the doom threads. Search Engine Land's August analysis of client-side analytics found LLM referral traffic converting at 20%, the highest of any tactic in their dataset and 61% higher than paid search. The explanation is structural, not magical: by the time someone clicks out of an AI answer, the model has already done the comparison, the shortlist and most of the objection handling. What lands on your site is not a browser. It is a decision looking for a checkout.
"Fewer visits, worth more each. Any measurement plan that only counts the first half of that sentence will conclude the channel is dying while it quietly closes deals."
The same pattern shows up in behavioral work: Scrunch's prompt-to-purchase analysis of millions of search events found AI touching decisions well before any site visit registers, and Semrush's ROI guidance now treats AI referral traffic as the last, smallest visible slice of the influence it is trying to measure.
What to measure instead
The working answer emerging across the industry is a stack, not a single number. Search Engine Land's five-layer framework is a reasonable template: track whether engines can access your content, whether they cite you, what share of answers in your category you occupy, what that does to branded search and direct traffic, and only then what the referral clicks convert at. Citation share is the leading indicator in that stack. Clicks are the trailing one. Most dashboards have it exactly backwards.
What to do this week
Split "AI traffic" out of your referral reports and measure its conversion rate against paid search. If your numbers rhyme with the 20% figure, that is the argument for the budget conversation. Then set up a monthly citation-share check on your ten most valuable prompts, because that number moves before traffic does. The visits are not coming back. The influence never left.