There is a version of the customer journey that leaves no trace in your analytics. Someone asks a model a question, gets an answer with two or three brands in it, forms an opinion, and either types your name into a browser a week later or never thinks about you again.
Nothing in that sequence produces a referrer. If they do arrive eventually, they arrive as direct traffic, which is where unexplained things have always gone to hide.
Marketers have lived with dark traffic for years — the podcast mention, the message in a group chat, the recommendation over lunch. The response was always to accept a fuzzy edge on attribution and get on with it.
What is different here is the volume and the position. This is not a fuzzy edge; it is a step that sits directly before consideration, happening at scale, and it is one where a machine is doing the summarising rather than a friend.
If you cannot measure the click, measure whether you were in the room. That is what visibility tracking is actually for: not proving conversion, but establishing whether your brand exists at the moment the shortlist is drawn up.
A brand named in the answer has a chance of being chosen. A brand absent from the answer has no chance at all, and no amount of downstream optimisation fixes that.
Traffic data does still help, just not as proof. If the questions you track start naming you more often, and direct traffic and branded search rise over the same period, that is not attribution — but it is consistent, and consistency across independent signals is how most brand measurement has always worked.
The mistake is demanding a standard of proof from this channel that nobody ever demanded from brand advertising.
The practical consequence is that some pages should be written to be summarised rather than to be visited. A page that states its facts plainly, early, and without hedging is a page a model can lift from cleanly.
That is an uncomfortable thing to optimise for, because the payoff never shows up in that page's own traffic. It shows up as your name appearing in an answer somewhere you will never see.
If the click is not available, the substitutes are all indirect, and each is weak on its own.
Branded search volume is the closest thing to a proxy — people who encountered you somewhere unmeasured and went looking afterwards. Direct traffic is noisier but moves for the same reason. Both are blunt, both are influenced by everything else you do, and neither will isolate this channel.
The more honest instrument is the answer itself. Whether your brand is named, where in the answer, and in what terms, is directly observable, repeatable, and does not depend on inferring anything from traffic that may never arrive.
The risk in a channel without clicks is that it gets measured by whatever number happens to be available, and then that number gets optimised.
Citation counts are the usual candidate, because they are countable. But a brand can accumulate citations while consistently being framed as the second choice, and a report showing that count rising quarter on quarter would look like progress throughout.
Pick the measure that matches the outcome you care about. If the outcome is being chosen, the measure is being named first and described well — not being linked.
Contents
Models group brands their own way, and the set they put you in shapes every answer about you.
Read →AI answers are replacing search clicks, and the tools to measure it were priced for enterprises. So we built one for the rest of us.
Read →Not your homepage. A working theory of the sources that shape what AI answers say about a brand, and what that means for where you spend effort.
Read →Tell us your domain and we'll send you a visibility report — free, no card.