“AI traffic? Basically zero for us.” I hear this a lot lately.
You open Google Analytics, and referrals from ChatGPT don’t even reach 1% of your traffic. And yet somebody upstairs keeps saying “we need an AI strategy.”
Annoying, right? I stared at that same dashboard and thought “fine, we can wait.”
But that sub-1% number might not be evidence that AI isn’t working.
Because there’s data suggesting the AI effect shows up somewhere else entirely — in a column your referral report doesn’t label “AI.”
So let’s dig into it, using a third-party consumer panel study.
AI Gets Used Most Before Anyone Buys Anything
The study is Similarweb’s June 2026 look at how US consumers shop, based on their consumer panel (the study).
The first interesting bit is when people reach for AI.
At the discovery stage — “what should I buy, what are my options” — 35.0% of people said AI tools were the most useful thing they used. Search engines got 13.6%.
At the research stage it’s still lopsided: AI at 32.9%, search at 15%. More than double.
But at the price-comparison stage — “where do I actually buy this” — it evens out: AI at 24.3%, search at 22.1%.
So AI dominates the fuzzy, early “what’s even out there” phase. Which happens to be exactly where your gets formed — the mental shortlist of brands worth looking at.
So Why Is The Referral Number Under 1%?
Which brings us back to the opening puzzle. If AI owns the front of the funnel, why does it send you almost no traffic?
The answer is boring: AI mostly resolves things inside the conversation. Nobody clicks a link because there’s often no link worth clicking.
And it’s not that people stopped using the tools. Visits to AI platforms grew 28.6% in the year from 2025 to 2026.
Referral traffic from those platforms to websites? Flat.
People take the name the AI gave them and come back for it later, through some other door.
The Effect Books Itself Under Branded Search
So where’s the other door?
Per the study, among people who got a recommendation from an AI, 40% went and ran an additional Google search, and 28% went straight to the brand’s site.
is just that — searching for the company or product by name.
And here’s the decisive number: 55.9% of AI-influenced visits arrived via branded search.
For visits with no AI influence, it was 40.4%. That’s a real gap.
So the AI’s contribution never lands in the “AI” row of your traffic report. It gets quietly booked under “branded search” and “direct” — and erases the trail.
It’s like pointing your only security camera at the front door while customers stream in through the side entrance.
A Recommended Brand Got 2.5x The Visits, No Link Required
That’s the “hard to see” half. Now the “how big is it” half — the strongest number in the study.
People who saw ChatGPT recommend a brand were 2.5 times more likely to visit that brand than its competitor — with no direct link and no prior visit to the site.
And visitors who came off an AI recommendation viewed about 2x the pages and stayed about 2x as long as regular visitors.
Just being named lifted both the volume and the quality of visits. That’s a strange sentence to type, but there it is.
One caveat, though, and it’s worth sitting with. Similarweb sells AI-visibility measurement. A company that profits when “AI visibility matters” is the company telling you AI visibility matters — so read the exact figures with a discount.
Still, the early-funnel usage pattern and the branded-search spillover point the same direction, and that direction holds up.
Conclusion: Don’t Size Your AI Bet By The Referral Number
So what does this actually change? One thing, mostly: a sub-1% AI referral number is not the same claim as “AI isn’t working.”
The effect lives in branded search and direct — outside the frame your last-click report draws. Which means the fix isn’t to squint harder at the AI row. It’s to watch a different row.
I think it comes down to two moves:
- Track branded search growth as a success metric alongside (not instead of) AI referral counts
- Keep measuring how the AI names you — which category it puts you in, and what it says
Getting named by an AI looks less glamorous than getting clicked. But that unglamorous exposure is what comes back later as 2.5x the visits.
Maybe I’m wrong and the panel skews somewhere odd — the sample size isn’t public, after all. But “measure it before you dismiss it” beats “dismiss it because the dashboard’s quiet.” Start by counting how you’re talked about now.
Sources
- Similarweb, “How AI Is Changing the Consumer Buying Journey” (2026 AI Brand Visibility Index / AI Consumer Journey), June 30 2026, similarweb.com (US Market Research Panel, January 2026 data. By journey stage, “most useful tool”: discovery AI 35.0% vs search 13.6%; research 32.9% vs 15%; price/purchase 24.3% vs 22.1%. AI referral traffic under 1% of organic; AI platform visits up 28.6% in a year while referrals stayed flat. 55.9% of AI-influenced visits came via branded search vs 40.4% without AI influence; after an AI recommendation 40% searched Google and 28% went direct. A ChatGPT-recommended brand drew 2.5x the visits of its competitor with no direct link or prior visit, and AI-referred visitors roughly doubled pages viewed and time on site. Caveats: Similarweb sells GEO measurement, and the panel’s exact sample size is not disclosed.)
