For two years the knock on AI search has been that it does not send traffic. The model answers the question, the user never clicks, and your analytics show nothing. New data complicates that story in a way every marketer should sit with.
SparkToro published research from Similarweb this week measuring what happens after an AI tool recommends a brand. The finding is not subtle. The mention moves people.
The Numbers Behind the Mention
When an AI assistant recommended Capital One for a credit card question, users were 14.2 percent more likely to visit Capital One afterward. American Express mentions showed a 7.2 percent lift in subsequent visits. These are not impressions. They are people arriving.
The more telling detail is where they arrive from. When a visit was influenced by an AI recommendation, search activity dropped roughly 15 percent and direct visits rose. People were not Googling the brand to double-check. They were typing the address straight in, or tapping through, because the AI already did the convincing.
And it happened fast. The study measured the effect inside a seven-day window. That is not slow brand building. That is a recommendation translating into a visit while the intent is still warm.
Read those three facts together and the shape is clear. An AI mention is starting to behave less like a citation and more like a referral from someone the buyer trusts. The search step, the one marketers have optimized for two decades, gets skipped.
Why This Changes the Math on Visibility
I have argued for a while that AI mentions are not the same as AI trust, and that still holds. A mention is not a sale. But this data adds the missing half. A mention is now a measurable nudge toward the door, and the nudge is large enough to show up in traffic logs.
That reframes the cost of being absent. If your competitor gets named when a buyer asks an AI which provider to use, they are not just getting visibility. They are getting a 7 to 14 percent edge in who shows up next, before you have any chance to compete on the page. The answer is the new top of funnel, and it closes faster than search ever did.
From building GEOflux, the pattern I see matches the data. Brands that get recommended inside AI answers report traffic they cannot fully explain in their old attribution, because the path no longer runs through a clean search query. It runs through a conversation the brand never saw and cannot replay.
There is a caveat worth keeping honest. Similarweb studied large, established brands, Capital One, American Express, Sephora, in finance, travel, and beauty. Whether a smaller brand gets the same lift from an AI mention is an open question. The mechanism should travel. The size of the effect might not.
What to Actually Do About It
There is a strategic read hiding in the direction of that traffic. If AI mentions push people to type your name directly, then the brands with the strongest existing recognition compound the fastest. The model recommends them, the user already knows them, and the visit closes without friction. A weaker brand gets recommended and the user still hesitates, still checks, still drifts. The mention rewards the reputation you already built, which means the real work starts long before an AI ever names you. That is the uncomfortable part for challengers, and the quiet advantage for anyone who invested early.
Start by measuring whether you get mentioned at all. Pick the ten questions a buyer would ask an AI before choosing your category, and read the answers. If your brand is missing, that absence is your real ranking, and no amount of paid search makes up for being left out of the recommendation itself.
Then watch your direct traffic differently. A rise in direct visits with no matching campaign is no longer just brand awareness doing its slow work. It may be AI recommendations converting, and you want to know which questions are driving it. I wrote before that search does not build your brand, your reputation does, and this is the same idea with a traffic number attached. The reputation the model repeats is now showing up in your logs.
The practical work has not changed, but the stakes have. Earn mentions on the sources the model reads. Make claims specific enough that an AI repeats them word for word. Keep your positioning consistent enough that every answer lands on the same description of you.
The old complaint was that AI search gives you nothing back. The data just retired that excuse. AI mentions move real traffic, on a seven-day clock, and they do it by skipping the search box you spent years trying to win. The brands that get this will stop treating AI visibility as a vanity metric and start treating it as the demand channel it is becoming.