Your AI Shoppers Are Showing Up in Your Google Numbers

Only 8.8% of the traffic an AI recommendation produces arrives looking like AI traffic.

That figure comes from Similarweb, which tracked what happens to a brand after an assistant recommends it. Visits jumped 2.5x within seven days. But 55.9% of that downstream traffic came in as branded search, someone typing the company’s name into Google, and only 8.8% came through a direct link from the AI. Scot Wingo, who runs the agentic commerce newsletter Retailgentic, published the breakdown on August 9.

Read that next to what Shopify told analysts two weeks ago and a confusing quarter starts making sense.

What Shopify reported

On its August 5 earnings call, Shopify said AI-driven traffic and orders tripled YoY in Q2, and that Google traffic did not fall to pay for it. President Harley Finkelstein told analysts that AI has become a “complement to search, rather than a substitute for it.”

His supporting number was about the old channel, not the new one:

“Traditional search sessions are up 1.3x over the past two years, holding roughly a third of all storefront sessions”

That’s the opposite of the story most marketing teams have been told, and the opposite of what Getty reported a week ago when it blamed AI answers for new customers drying up. Both can be true. Getty sells a commodity that an AI answer can replace. A merchant sells a specific product an AI has to send you to.

One more Shopify figure explains the mechanics. Finkelstein said “half of all AI-referred sessions are landing directly on a product description page,” which he put at 2.5 times the rate for traditional search. The shopper arrives having already done the comparing.

Why the dashboard gets it wrong

Put the two datasets together and the measurement problem is obvious.

An assistant recommends three brands. The shopper reads the answer, then does what people have done for 20 years: opens a new tab and searches the name of the one they liked. Google serves the brand’s own listing. The click lands in your analytics as organic search, or as branded search if you segment that far.

The AI never touched your numbers. It did all the work.

So the marketing team looks at a flat AI referral line and concludes AI isn’t sending traffic yet. Meanwhile branded search is climbing and gets filed as brand awareness, or as a nice result from last quarter’s campaign. We’ve made the point before that AI visibility is not referral traffic. Similarweb’s split puts a ratio on it: for every visit that arrives tagged as AI, roughly six arrive wearing Google’s clothes.

What this changes about your reporting

The practical consequence is that the two numbers most teams watch are both misleading on their own.

Your AI referral traffic understates AI’s effect, badly. It captures the minority of shoppers who click straight through from an answer. Treat it as a floor, not a measurement.

Your branded search volume is now partly an AI metric. If it’s rising without a campaign, an assistant may be doing your top-of-funnel work. That’s worth knowing before you credit it to something else, and before you cut whatever is feeding it.

Shopify’s other Q2 detail sharpens where this matters: 75% of AI-attributed purchases happened outside its top 100 categories. Niche and specific beat head terms, which is the same shape we found in ChatGPT’s retrieval stack, where a small index favors pages that answer a narrow question cleanly.

Where the evidence is thinner than it looks

Two honest limits.

The conversion figures that get quoted alongside all this are from a different quarter. Shopify’s numbers on AI-referred shoppers converting nearly 50% better and carrying 14% higher order values come from its own Q1 2026 data, published in May, not from the Q2 call. Trade coverage has been blending the two. They’re separate windows and separate claims.

And Shopify has an obvious interest in the conclusion that AI expands commerce rather than concentrating it. A platform whose merchants are mostly small benefits enormously if assistants favor the long tail. The finding may still be right. It isn’t disinterested, and Similarweb’s branded-search split is the more useful number precisely because Similarweb isn’t selling storefronts.

What to do Monday

Pull two lines and put them on one chart: branded search sessions and AI referral sessions, monthly, for the last year.

If branded search is growing faster than your campaign spend explains, and AI referrals are growing at all, you’re seeing the same effect Similarweb measured, and your AI number is the small visible part of it. Report them together or you’ll keep underrating a channel while quietly depending on it.

File it under AI, content and brand. The awkward part is that the channel doing the most work leaves the least evidence behind.