Your Next Customer Will Ask an AI About You

Last month I bought a pair of running shoes, and the way I bought them is the reason I am writing this.

I never saw a results page, a review roundup, or a “best running shoes 2026” listicle with affiliate links three paragraphs deep. I opened a chat assistant, typed that my knees had been hurting from running lately, and asked whether Hokas would be good for me. It said yes, more or less, and twenty minutes later the order was placed.

At the moment I typed that question, I could not have told you what made Hoka different from any other brand. Only afterward did I remember watching a video about marathon runners breaking the two-hour barrier, and noticing what was on the runner’s feet.

Somebody in marketing made that purchase happen, and they received nothing their analytics could capture even in principle: no click, no promo code, no referral source. As far as their dashboard is concerned, I do not exist.

This Is No Longer an Anecdote

Adobe has analyzed more than a trillion visits to U.S. retail sites, and the trend line is steep. Traffic from AI sources to those sites grew 393% year over year in the first quarter of 2026, following a 693% jump over the preceding holiday season.

The growth is the less interesting half of the story, because what reversed was the quality of that traffic. A year ago, visitors arriving from AI assistants converted 38% worse than every other channel. By March 2026, they converted 42% better, a record high and an eighty-point swing in twelve months.

That swing says more about the buyers than about the traffic, because people who arrive from an assistant have already decided what they want.

I should disclose that I run an influencer intelligence platform; we measure creators and audiences, not checkout carts, so this data comes from outside my own walls, but the direction is not subtle. Adobe’s consumer research found 39% of respondents had used AI for shopping and 85% said it improved the experience, and the same pattern now shows up well beyond retail.

The Machine Cannot Watch the Video

When an assistant researches a product, it reads reviews, spec sheets, forum threads and documentation. The twelve-minute video review that actually persuaded a human reaches the model as a transcript at best, and frequently not at all.

The tone of voice, the b-roll, and the face someone has trusted for three years never make the trip.

Researchers are only beginning to examine what happens in that gap. A 2026 study by Møller and colleagues found that people extracting information from video with AI assistance were measurably better off skipping the video, with accuracy improving by up to 35%. But when the assistant was fed false information, accuracy collapsed by 32% while participants’ confidence held steady at roughly 4.5 out of 5, which means people could not tell when the summary was wrong.

The most persuasive marketing format of the past decade is close to invisible to the system that now closes the sale, and the humans in the loop are poorly equipped to notice the substitution.

Influence Did Not Disappear, It Moved

I came out of this more bullish on creator marketing, not less, and the reason is in the exact wording of my own question.

I could have asked which running shoes are best for bad knees, which starts a fifteen-brand comparison in which the best spec sheet wins. Instead I asked about one brand by name, which bound the assistant: it was never going to show me alternatives, because I had not asked for any.

The brand inside the question skips the comparison entirely, while everyone else competes inside the answer.

That name came out of my fingers because runners I follow wear them. Somewhere over the previous year, creators put that brand in my head, so that when I finally sat down to buy, every competitor had been eliminated before the race started.

The click disappears but the influence does not; it moves upstream, into the question itself.

Communities Write the Evidence Layer

Being named is only half the win, because the assistant then checks its answer against text written by people.

This is where community stops being a soft metric and starts being infrastructure. An analysis of 30 million sources across ChatGPT, Google AI Mode, Gemini, Perplexity and AI Overviews found the most-cited domains were Reddit, YouTube, LinkedIn, Wikipedia and Forbes, with review platforms such as G2 and Yelp surfacing heavily on recommendation queries. An audience is no longer only influencing buyers; it is also authoring the corpus the model checks the answer against.

Gaming worked this out by accident: a game with tens of thousands of Steam reviews carries a richer machine-readable evidence layer than most consumer products will ever assemble deliberately.

The corollary is uncomfortable for brands’ own properties. Adobe’s audit of retail sites found product pages only 66% machine-readable, the worst-performing page type measured, and the one where the buying decision is supposed to happen.

What This Changes in Practice

None of this requires believing that search is dead. Brand recall becomes a performance metric rather than a soft one, because it determines which names get typed into the prompt at all. The text surface a brand and its community produce, meaning reviews, forum answers, documentation and structured product data, is now read by machines at least as often as by people, and should be maintained accordingly. Creator measurement has to account for influence that never touches a tracked link, which means holdout tests and brand-lift work rather than better pixels.

Attribution Was Always a Polite Fiction

This part should make marketers feel better rather than worse. Last-click measured proximity to checkout rather than influence, and AI did not so much break a working system as remove the pretense. What is left underneath is the asset that was always doing the work: being remembered.

A machine can compare and summarize trade-offs all day, but it cannot decide what a human walks in caring about. As more of the transaction is handled by software, human memory becomes the scarce input, and scarce things become more valuable.

Influencer marketing is not dying; it is about to be measured honestly for the first time.

I spent years as a content creator before I started building tools for them, so I will put it plainly: the brands that win the next decade will not be the ones that shout loudest or buy the most clicks, but the ones a real person remembers fondly enough to name out loud and back the claim up in text a machine can read.