Semrush pulled clickstream data from almost 69 million Google Search sessions and found that 92 to 94% of Google AI Mode searches end without a single click to any website. Regular Google Search runs about 34% zero-click. With an AI Overview present, that climbs to about 43%. AI Mode sends almost no traffic anywhere, at more than double a standard AI Overview’s rate.
Hold that number next to the quarterly traffic slide, because it explains a gap more teams are living with every month. Organic traffic reads as decline while sales keeps hearing from buyers who arrive already convinced, certain they were told this vendor was the best fit. The recommendation happened inside an AI answer, and nothing in the analytics stack was pointed anywhere near it.
What an AI answer is built from
In McKinsey’s research on how people use AI-powered search, 44% now call it their primary, preferred source of insight, ahead of traditional search at 31%, brand or retailer websites at 9%, and review sites at 6%. People are ranking AI answers above a brand’s own website as the place to learn about that brand.
There’s a stranger finding underneath. The same research found a brand’s own site accounts for just 5 to 10% of the sources an AI answer draws from. The rest comes from publishers, affiliate content, forums, and reviews the brand didn’t write and can’t edit. The mix shifts by category. Consumer packaged goods answers pull roughly half their material from affiliate content, while ecommerce answers lean the other way, with about 80% coming from brand and retailer sites directly. A B2B healthcare or financial services brand likely sits closer to the CPG pattern than the ecommerce one, which means the buyer’s AI answer is more likely built from a third-party comparison than from the brand’s own product page.
So when a buyer never visits the website, the reason is simple. The answer they read pulled from somewhere else entirely, a review site, a comparison post, or a forum thread, somewhere no dashboard was ever pointed.
Visible in the exact channel nobody measures
This would matter less if marketing teams already had a way to see it. Most don’t. Goodfirms’ early-2026 survey of marketing and SEO practitioners found 89% of brands already get cited somewhere in AI-generated search results, while only 14% of marketers track the citations. And the money is moving faster than the measurement. Branch’s survey of enterprise marketing leaders found 65% are putting at least a quarter of their 2026 budget toward AI search optimization, budget flowing into a channel most teams can’t yet measure.
None of this makes the traffic dashboard wrong. It’s built to answer one question, where did this session come from, and a growing share of buyers never trigger that question at all.
The second dashboard
Three additions close part of the gap, and none requires replacing the analytics stack in place.
Track brand mentions the way a reputation channel gets tracked. A new class of tools monitors how often, and how favorably, a brand shows up inside ChatGPT, Perplexity, and Google’s AI surfaces, places Search Console never touches.
Watch branded search volume alongside organic traffic. A buyer who reads an AI recommendation and later searches the company name directly is the AI touchpoint surfacing a few steps removed from where it happened. No report will ever label it AI search. The timing is the only signal you get.
Ask sales what they’re hearing. One question on a discovery call, “how did you first hear about us,” captures a version of the journey no pixel ever will.
Built for a different search era
Complex industries feel this first. Healthcare and financial services buyers already research quietly, across long cycles, before they ever contact a vendor, and AI search extends that quiet phase further from anything a pixel can log. Getting into those answers is its own discipline with its own post. This one is about the proof problem that follows. Once a brand shows up in that research and gets recommended, most teams still have no way to show it happened.
The traffic slide can stay in the deck. It earns a partner built to answer the question the first slide was never designed to ask, whether AI recommends you, next to whether anyone clicked. Building that second dashboard is part of the discoverability practice now.
The recommendation happened. Your dashboard just wasn’t in the room.






