Kantar has started running its brand trackers twice. Once on live respondents, the way it always has. And once on large language models, asking them the same questions about the same category.
For one client, a category leader, the two readings split clean down the middle. With people, the brand sat at number one. Inside the machine, in the same category, it came back sixth.
That gap is the new reality the market is only starting to find words for. A brand can own a consumer’s memory and still be invisible to the system that increasingly stands between that consumer and the purchase.
Gonzalo Fuentes, Kantar’s CEO of Insights for EMEA, calls the metric “brand power,” the probability that you get chosen, and his point is that it now has two values and they no longer agree. For decades marketing did one job: win the human. Now there is a second job, predispose the algorithm, and most brands have not noticed they are failing at it.
The scale is what makes the blind spot expensive. Kantar’s own Marketing Trends 2026 report finds that 24% of AI users already use AI shopping assistants for purchase decisions, and three-quarters of people who use AI assistants regularly seek out AI-driven recommendations.
McKinsey puts a number on where that leads. By 2030, on its research, agentic commerce could orchestrate three to five trillion dollars of global sales, and 38% of European consumers already use generative AI to research products and decide what to buy. If even part of that holds, the unit of marketing changes. You are no longer competing only for a place in a human’s consideration set. You are competing for a place in a model’s output.
Why simply being in the results stopped working
The first instinct of a brand that discovers it is missing from AI answers is to treat it like old SEO. Show up more, get crawled more, climb the ranking. That is the wrong model of what is happening. The machine is not running a popularity contest.
When an LLM ranks options inside a category today, it leans on rational signals: price, product benefits, composition, how the thing compares on paper. A brand whose strength lives in emotion and memory, the exact asset that wins with humans, can sink precisely where it is strongest. Kantar’s own framing of the shift is blunt: salience alone won’t make you algorithmically preferred.
The most telling thing about how the people running real brands describe this is that they find the problem in symptoms, not in reports. Eduardo Barbaro, Chief Omnichannel and Digital Officer at Tiffany, noticed in his own data that web traffic was sliding down, not a collapse but enough to make him uncomfortable, and that the visitors who did arrive were spending less time before deciding.
Which meant discovery was happening somewhere he could not see. He went and looked at the AI platforms and asked a sharper question than most brands ask. Not whether Tiffany appeared, but how Tiffany was described.
He reduced it to three questions a brand now has to answer about itself inside a model. Is it visible at all? When a customer asks for advice, does the AI suggest it? And the important one: does the AI understand what the brand stands for?
For an engagement ring, the specifications are not the product. The product is the emotional weight, and a model that compresses a Tiffany ring into carat, cut and price has lost the thing the brand sells.
There is a paradox buried here that matters well beyond luxury. Luxury spent decades building mystique, deliberately withholding, creating desire through what it refused to explain. The machine is a literalist. It wants the thing solved.
Reconciling those two logics is not a content fix. It is a question about whether a brand can stay legible to an algorithm without becoming generic to a human. McKinsey, looking at the same problem in luxury specifically, frames it as a race to rebuild the “front door” of retail before someone else defines that layer for you. And that question is now in front of everyone, not only the jewelers.
Trust as the one signal you cannot fabricate
If rational specs are easy to copy, the open question is what will distinguish brands in agent-mediated commerce at all. Sarah Taieb, who runs eBay in France, bets on trust, and a marketplace built on resale has more riding on it than most. Buying a standardized product, you compare price, features, delivery, and any seller will do.
Ask an AI assistant for the best refurbished iPhone under a certain price in the best condition for the money, or an authentic vintage watch, and the model has to do something harder. It has to be confident the item is real and the seller reliable, because it is now staking its own recommendation on that confidence.
eBay’s answer is to build the trust into the infrastructure, with programs like Authenticity Guarantee, where high-value items in categories like watches, jewelry and sneakers are physically inspected by experts before reaching the buyer. The argument that AI is less a technology than a trust engine is convenient for a resale platform.
It is also probably right, and that is the uncomfortable part for brands without the infrastructure: trust cannot be written into a product card, it has to be built over years.
Derya Matras, VP of Meta’s Global Business Group for EMEA, translates the same thing into economics. In a world where making a competent ad or a passable piece of content costs almost nothing, the cheap stuff stops differentiating and the expensive thing, real distinctiveness, becomes the whole game.
Meta has data on what its own automation buys: advertisers using Advantage+ AI targeting see roughly 22% higher return on ad spend than those who don’t. Her recipe for standing out comes down to three things: be where the customer already spends time rather than on a site they visit once a month; know the customer well enough to anticipate intent; and deliver the brand promise consistently enough that it survives contact with the algorithm.
What the machine actually reads
The most underrated takeaway here concerns the limit of optimizing for AI, not the optimizing itself. Fuentes closed on a thought the others kept circling: how a brand looks inside the machine is increasingly determined by real human experience.
The more people genuinely engage with a product, an event, a live stream, the more raw material the models have to learn what the brand means. The thing that looks least like AI marketing, a person watching another person open a booster pack of trading cards on a live shopping stream, may turn out to be exactly what teaches the algorithm who you are.
Kantar’s report lands in the same place: if the model doesn’t know you, it won’t choose you, so the brand’s job is to be present in the content models actually learn from.
Which leaves a question the market has not closed. If models learn what brands mean from the texture of human experience, and if that experience is the one input a competitor cannot cheaply clone, then the algorithm may be won by the brands that invested least in gaming it.
The McKinsey and Kantar framing is that winners will rebuild their workflow around AI rather than bolt it on. Fair enough.
The harder version of that point is that the rebuild may have to start somewhere unfashionable.
Not in the model, but in whatever real thing the model is eventually going to read.
