How do you measure marketing attribution when discovery happens inside AI answers?

Stop trying to force last-click to explain arrivals it never saw, and measure 3 things instead. Share of answer: run the real buying questions of your category against ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews on a fixed cadence, and track whether you appear, how you are described, and who appears beside you. The visible shadow: branded search and direct traffic are what invisible recommendations look like from inside your analytics, so watch their correlation with the share-of-answer trend. And AI referrals where they exist, segmented separately, because a visitor arriving from an answer behaves like a shortlist finalist rather than a browser.

How do you make a product legible to AI models?

The unit of optimization moved from the page to the passage. Write self-contained claims that survive being quoted alone. Name your product and category consistently across the ecosystem, because retrieval is entity work. Publish original data worth citing, since models prefer sources that add facts to the pool. And build reputation beyond your own domain, because models weigh what others say about you more heavily than what you say about yourself.

Attribution After the Click: Measuring AI Search

Q: What is the difference between zero-click search and zero-visit discovery?

Zero-click meant an answer appeared above the links and the user did not need to click. Zero-visit is different in kind. An answer engine synthesizes a shortlist from passages it retrieved, the buyer evaluates inside the conversation, and if they reach your site at all they arrive pre-convinced, typing your name directly. The shortlist is set before the search bar. Influence happened and no event fired.

How do you measure attribution when discovery happens inside AI answers? You stop asking last-click to explain arrivals it never saw. An answer engine builds a shortlist from passages it retrieved, the buyer decides inside the conversation, and if they reach your site at all they arrive pre-convinced and type your name directly. Influence happened and no event fired.

Three measures replace it. Share of answer, meaning the real buying questions of your category run against ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews on a fixed cadence, tracking whether you appear and who appears beside you. The visible shadow, meaning branded search and direct traffic watched against that trend. And AI referrals segmented on their own, because someone arriving from an answer is a shortlist finalist rather than a browser.

Growth measurement was built on 1 physical event: the click. Someone saw, someone clicked, a parameter followed them, and the revenue report knew where they came from. Discovery is now moving inside LLM answers, and the click increasingly never happens. The model read your documentation months ago, formed its view, and recommended you or your competitor in a conversation you will never see.

Zero-click was the warning, zero-visit is the event

SEO already lived through answers appearing above the links. This is different in kind, not degree. An answer engine synthesizes a shortlist from passages it retrieved, the buyer evaluates inside the conversation, and if they arrive at your site at all, they arrive pre-convinced, typing your name directly. The shortlist is set before the search bar. Influence happened, and no event fired.

Adobe agreeing to buy Semrush for about $1.9 billion, framed explicitly around brand visibility in the agentic era, is the clearest price tag anyone has put on that shift.

What should you measure instead?

Share of answer. Take the real buying questions of your category, run them against ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews on a fixed cadence, and track whether you appear, how you are described, and who appears beside you. It is the new share of shelf, and it moves slowly enough to trend honestly.

The visible shadow. Branded search and direct traffic are what invisible recommendations look like from inside your analytics. Watch their correlation with your share-of-answer trend instead of forcing last-click to explain arrivals it never saw.

AI referrals where they exist. Some answer surfaces do send traffic. Segment it, because a visitor who arrives from an answer behaves like a shortlist finalist, not a browser. Adobe's holiday data put AI-referred visitors at a materially higher purchase completion rate than search visitors, which is the same qualified-arrival pattern I found in Friction Was Doing a Job.

How do you become legible to models?

The unit of optimization moved from the page to the passage. Self-contained claims that survive being quoted alone. Consistent naming of your product and category across the ecosystem, because retrieval is entity work. Original data worth citing, since models prefer sources that add facts to the pool. And a reputation beyond your own site, because the models weigh what others say about you more than what you say about yourself. Most of this is strategy wearing a technical costume.

The uncomfortable prerequisite is that the citation research keeps landing in the same place: the overwhelming majority of AI Overview citations come from pages already ranking in the organic top 10. Classic search visibility is the entry ticket, not the thing being replaced.

The agent makes it sharper

When the evaluator is an agent selecting from a tool registry, there is no impression at all, only the call or its absence. The measurement discipline and the activation discipline converge on the same requirement: be discoverable, be legible, work on the first attempt.

Honesty over precision

Some influence is now structurally unmeasurable, and the mature move is to say so. Pair the funnel you can still measure with a monitored proxy set you review on cadence, and retire the dashboards that claim a precision the click era took with it. The strategy side of this shift is distribution itself, which I covered in Distribution Is a Product Skill Now. The click is not coming back. The teams that admit it first get a head start on the measurement everyone will eventually need.