When the buyer is a machine: how brands get built and measured in the agentic era
AI assistants build the shortlist without looking at your ads. What a machine can see in a brand, how to build for it.
Brands are built on feeling. That has been true since the first trademark. A brand is what a person believes about you before they check, and what they forgive you for after. Advertising exists to grow and reinforce that belief. Design makes it recognisable at every point of contact, so the feeling attaches to the logo, the packaging, the lobby, the website. Neither works on a machine.
That matters now because the machines are already the majority of the audience. Matthew Prince, the chief executive of Cloudflare, which carries a large share of the world’s web traffic, said in June that bots and AI agents now make up more than half the page requests on his network, the first time that has happened. He had expected the crossover at the end of 2027; it came about eighteen months early. He repeated the point on the Big Technology podcast this week and said he expects the ratio to keep widening. A growing share of buying decisions now start with a machine. Someone asks an assistant for a lakeside resort with a spa and a golf course, or an internet provider that does not throttle, or a warm island that is quiet in November. The assistant reads what it can find, weighs it, and hands back a shortlist of three. The person picks from the three. The nine brands that were not on the list never had a chance to make anyone feel anything.
The assistant did not look at the ads, notice the photography, or respond to the colour palette or the tone of the headline. Bots don’t care about art direction. It read text, compared facts, and checked who else agreed.
The wrong lesson
It is tempting to conclude that emotion no longer matters and that brand is now a data problem.
The assistant has no feelings, but it is built from what people wrote. Its recommendations lean on reviews, forum threads, press coverage, comparison articles and the tone of the language around a name. When guests love a place, they write about it. When customers feel looked after, they say so in public. The assistant reads all of that and treats it as evidence. So emotion still moves the outcome, one step removed. The feeling has to be turned into a record before a machine can use it.
Emotion also survives because for most purchases the machine shortlists and the person decides. The final choice among three is still made by a human looking at three websites, three photo sets, three prices. Creative and design still do their job there. What has changed is that they no longer get you into the room.
What a machine can see
If you want a machine to recommend you, you need to understand what it can and cannot perceive. It cannot see beauty, wit or confidence. It can see four things.
It can see whether you are legible. Can it find out what you are, what you sell, what it costs, where you are, what your hours and policies are, and is that information laid out in a form it can read without guessing? A resort whose rates live in a PDF and whose spa menu is an image is hard to read. A hard-to-read business is a risky recommendation, and the machine avoids risk.
It can see whether you are consistent. It reads your site, then reads Google, TripAdvisor, Wikipedia, the trade press and a dozen directories, and it compares. When your own site says one thing and three other sources say another, it does not pick your version. It marks you as unreliable and moves on. Most businesses have never audited this because until now nobody was checking.
It can see whether other people vouch for you. The machine trusts what others say about you over what you say about yourself, and it weights that by who is saying it. A hundred recent reviews, a mention in a national newspaper, a listing on a respected industry body’s site: these are the modern equivalents of a strong campaign. They are also the only place a customer’s feeling about you becomes visible to the machine.
It can see whether you are different in fact. A machine separates options by attributes. If two resorts are both “luxurious” and “welcoming” it cannot tell them apart and it will choose on price or proximity. If one has the only championship course within an hour of the airport, or the only fibre network in its town built and run by a local company, that is a fact the machine can hold onto and repeat. Distinctiveness has to be true, specific and rare. Tone will not do it.
The new work of building a brand
Put those four things together and the job of brand-building changes. It used to be deciding what you want people to feel, then making things that produce the feeling. It becomes deciding what you want to be known for, making it true, writing it down clearly, and getting other people to write it down too.
Some of this is technical and unglamorous: structured data on the website, facts that agree across every listing, and prices and availability that a machine can read. This is plumbing, and most marketing departments have never owned it.
Some of it is operational. Reviews come from experience, not from asking. Press coverage comes from doing something worth covering. The customer-service team, the front desk and the installer now have more to do with the brand than the ad agency does, because they produce the raw material the machine reads.
And some of it is old-fashioned positioning. You need a claim that is specific enough for a machine to repeat and true enough to survive a machine checking it. That is a harder discipline than a tagline.