Brands in the AI era

How to build positioning and brand preference when the one doing the choosing is Artificial Intelligence

Marketing is a war of perceptions

I say it often to my clients. It always has been, even in the years when we pretended it was a battle of products. We're convinced that the best product wins when in reality, the winner is whoever first occupies a position in the customer's mind. It's the fundamental law of positioning, and it's forty years old (have you ever read Al Ries?).

That law is still alive but for the first time in history, beyond the human mind, you also need to conquer that of its artificial extension. Between your brand and your customer, an intermediary has taken a seat: the AI assistant, the agent. Users are starting to search less and less, delegating to Claude, ChatGPT, Perplexity, or Grok the job of doing the work for them.

This means marketing now has two audiences (the consumer and the LLM assisting them), and the rules apply to both, with one ruthless detail: the machine punishes confusion far more than a person does. A blurry brand in a customer's mind is a weak brand. That same brand, in front of an LLM, simply doesn't exist.

The latent brand

There's a truth that managers struggle to accept: control over the brand is an illusion, and it always has been. The brand is what customers say about you, not what you say you are. The novelty is that now a third party, the machine, reconstructs that "who you are" from scratch every single time.

Vincenzo Cosenza (known to many simply as Vincos) coined the term "latent brand" in an article a few weeks ago, and I think he nailed the image. Your brand identity lives scattered across the traces you leave behind (website, reviews, articles, user-generated content, etc.) and takes shape only in the moment someone queries a model about you. Ask four chatbots to describe your company, and you'll get four different answers. None of them is the true one, but all of them are.

The brand is the statistical sum of the digital traces the machine aggregates in real time, brand book permitting. The consequence: stop broadcasting messages and start planting signals. Put in terms that the more technical among us can grasp: brands no longer control the output as they once did; they need to shift their focus to designing the input.

The brand that tries to mean everything ends up meaning nothing

Forty years ago, the rule went like this: better to be first than to be the best. First in the customer's mind, that is. Today, the translation of that rule is more brutal. Being first means being "the answer" not one of the many options the agent lists. When you ask an LLM, "What's the best X for Y," there's only one name that comes up first. And this brings to mind an evergreen that circulates in every marketing office: "focus" is the most neglected of all marketing laws. The brand that tries to mean everything ends up meaning nothing, both to people and to machines. In fact, the machine is merciless: faced with a generic brand, it rewards the specific one, the category leader, the first on the list.

Al Ries (again) wrote: "If you can't be first in a category, create a new category in which you can be first." Still true! In fact, it's truer today than ever. Because the agent doesn't think in terms of companies, it thinks in terms of categories: first, it understands what the user is looking for, then it finds who does it best. Owning a category means owning the first question, the one that comes before your name. And then (another wonderful implication) the name itself starts to carry weight again, in a way it hasn't for decades. For a model, your brand name is literally the key it uses to retrieve you from the pile. A generic, ambiguous name that resembles those of other competitors is a name the machine confuses or worse, discards. A distinctive name, on the other hand, is an asset that works even while you sleep. We've laughed for years at improbable brand names: today, that distinctiveness might turn out to be a far-sighted investment.

Then comes the companies' favorite temptation as they grow: expanding. Putting the brand name on everything, widening the product and service range, and claiming every possible shelf. In the agentic era, this becomes lethal: the more things you claim to be, the noisier the signal the machine receives, and a brand spread across ten categories doesn't own any of them firmly enough to come up first. The discipline of focus, which in the boardroom always feels like a concession, is in reality the only advantage the machine rewards.

Ask an AI assistant to describe your brand and compare it to your two main competitors, in three sentences. Those three sentences are your real positioning. Tone of voice, site aesthetics, and emotional campaigns, unfortunately, all of it evaporates in the summary. Only what has a clear, verifiable shape survives. If your brand's difference lives only in nuance, it's already lost: the machine throws nuance away. Differentiation today must be readable by an algorithm before it can be appreciated by a person.

Preparing for Share of Model

There's also a consequence that upends budget hierarchies decades in the making. The machine trusts what others say about you more than what you say about yourself. Your best traces are often signed by someone else: reviews, articles, mentions, various forms of content. Reputation builds brands; advertising defends them. In a world where a model reads you before a person does, this old provocation becomes a physical law. Earned reputation weighs more than bought messaging, because it's exactly what the model collects, cross-references, and returns. Investing everything in the perfect ad or the meticulously crafted headline, and nothing in reputation, means speaking beautifully to an audience that listens less and less.

For fifty years, we've measured share of voice. It's time to add what some are already calling "Share of Model": how often, how accurately, and with what sentiment does your brand appear in AI assistants' responses? One caveat: it's not a single number. There's no single model response about your brand. There is one for each intent. "Shoes for marathon runners," "Ethical shoes," "Brand controversies": three questions, three brands, one name. Share of Model is brand intelligence, not a simple data point, and it needs to be monitored with the same seriousness and consistency that once went into reading the press digest.

Before handing over the keys to the machines, it's worth avoiding a common misconception. In most cases, the agent surfaces a set of options but the final choice within that shortlist is still made by a human being. This means two games are being played simultaneously: getting onto the shortlist, where the machine rules with its evidence, and being chosen from the shortlist, where the person takes over with their emotions. Brands that optimize only for the machine build names that make the list, but don't get chosen. Brands that optimize only for emotion build beloved names that the machine never even puts on the list. You need both, and that may be the only true mindset shift this transformation asks of us.

The paradox of AI Transformation

For years, performance ate the brand budget alive, thanks to its measurability. The irony is that performance is precisely what AI automates first: bidding, creatives, targeting, attribution. And when everyone has access to the same automation, automation stops making a difference. What's left, the one thing that can still make a difference, the thing the machine can't automate for you or hand to your competitors, is the position your brand holds in the human mind. That's the paradox of AI Transformation. The better machines get at executing, the more what matters is the one thing they don't know how to do: decide who you are.

Every trace your brand leaves today, a piece of content, a review, a mention, will go on to compose the portrait a model paints of your brand two years from now, perhaps after freezing it in a training run whose calendar you don't control. The brand repays those who started early and had the patience to stay consistent. The bad news is that there are no shortcuts. The good news is that, for exactly that reason, the advantage of those who start now is defensible: your competitors can't make up for years of consistent signals they never planted, not in a week.

Three things to do to keep your brand from disappearing

To wrap up, let me suggest three moves in some ways straightforward, that are useful for keeping your brand from vanishing entirely.

First, audit how AI sees your brand today. Query the models about your brand, your category, and your competitors. Note what they say, what they get wrong, and who they mention. This is the new brand audit.

Second, own a category, not a showcase. Find the question to which you are the answer. If that question doesn't exist, the problem is one of positioning, not optimization.

Third, align your promise, your data, and your accessibility. Everything you claim to be must have a corresponding trace, and those traces must be machine-readable. An undocumented positioning is a positioning that doesn't exist.

Customers will keep buying with their emotions and justifying themselves with logic. But between their emotions and your brand, there is now an intermediary that has no emotions and reads only evidence. Branding over the coming years will have a dual, ancient task: building brands strong enough to move people emotionally, and clear enough to be chosen by machines.