Your next buyer is going to research you in a chat window before they ever load your homepage. They will open ChatGPT, or Google's new AI Mode, type in the job they are trying to get done rather than your company name, and read the paragraph a model writes back about who solves it. By the time they reach the site you rebuilt, they have already narrowed the field to the two or three companies the machine described most clearly, and you either made that cut or you did not. The homepage was never going to be the first impression this year. It is the second one, and most brands are still spending the entire budget on it.

This is the part of the AI shift brand people keep underreacting to. On May 19, Google announced what it called a new era for search and made AI Mode the default path for anything past a simple lookup, putting a model-written answer where the blue links used to sit. The next day, Klarna shipped a shopping app inside ChatGPT that pulls live products, prices, and stock straight into the conversation. Two announcements one day apart, pointing at the same shift: the first surface a buyer touches is no longer a page you designed, it is a paragraph a model wrote about you while you were asleep.

Your buyer now meets a machine's description of you before they ever reach your homepage, and that machine does not grade how you look. It grades whether your surfaces agree with each other well enough to describe you at all.

The first impression moved and nobody updated the brief

For fifteen years the job was clear. Get them to the homepage, win the first five seconds, control the story from there. Every brand decision downstream assumed the website was the front door and you were standing in it. That assumption is quietly breaking. The industry numbers are soft and self-serving, but the direction is not in dispute: most product research now begins inside an assistant, and a 6sense figure that gets passed around constantly puts large language models in the purchase journey for the overwhelming majority of B2B buyers. Your category is being explained to your buyer by something that is not you, before your art-directed front door has even loaded.

The uncomfortable part is that you do not get to design this surface. You put the budget into the hero section, the typography, the scroll animation, and the thing forming the first impression is a sentence you never wrote and cannot lay out. The question stops being how does my brand look and becomes what does it say when a machine reads everything about me at once and compresses it into three lines.

The model is not judging your taste, it is checking your story for contradictions

Here is the mechanism, because it changes what you do about it. When an answer engine gets your company wrong, it is almost never a random glitch but a retrieval failure. The model went looking for a clear, consistent account of who you are, could not find one it trusted, and built the most confident-sounding version it could out of whatever it could reach. So it lifts the one-liner off your homepage, an older and slightly different one off your LinkedIn, a third version from an old press release, a stale descriptor from Crunchbase, and a founder bio on X that still describes a company you no longer quite are. Then it averages them, or worse, picks the wrong one and states it with total confidence. That is where the damaging outcomes live: wrong founders, wrong headquarters, your history braided together with a competitor's because the model could not find the line where one of you ended and the other began.

Consistency used to be the boring part of brand, the slide nobody argued about in the readout. It just got promoted. When a model is the first reader, the consistency of your surfaces quietly becomes the input that decides whether you can be represented accurately at all. This is the credibility gap meeting a new and far less forgiving reader. For years the gap that cost you was the distance between what your company actually was and what a human felt when they landed on your site. Now there is a second gap sitting underneath it, the distance between what you say in one place and what you say in another, and a machine reads that distance as confusion and prices it straight into how it describes you.

The one-word name that helped you raise is working against you

There is a specific own goal worth naming here, because the people most exposed to it believe they made the safe choice. For the last few years the fashionable move in B2B and AI has been the clean dictionary-word name. Linear, Arc, Lattice, Mistral. It looks like a peer to every other funded company and sits nicely inside a lowercase wordmark. The trouble is that a dictionary word is exactly the condition the AEO crowd flags as hallucination-prone, because the model now has to disambiguate your company from the ordinary noun and from every adjacent startup that grabbed a word off the same shelf. You named yourself after a concept the model already holds ten thousand confident associations with, then gave it almost no structured signal to tell you apart. Naming was always strategy rather than decoration, and it just picked up a second job: a name is now also a retrieval instruction, and a vague one quietly tells the machine to guess.

What to actually do on a Tuesday

Open ChatGPT and type the sentence your buyer would type, not your name but the job they are trying to get done, something like best tools for X, or who handles Y for a company like mine, or alternatives to the obvious leader in your category. Read what it says about you, if it says anything at all, and notice whether the description is actually yours or just the average of your category wearing your logo. Then open five tabs and put your homepage, your LinkedIn company page, your Crunchbase entry, your last press release, and your founder's bio side by side, and read the single sentence each uses to say what you do and who it is for. Wherever those five disagree is exactly where the model is guessing, at the moment your buyer is forming a first opinion. Fix the disagreement before you touch the logo. The most valuable branding work in your company this quarter is very likely getting five surfaces you already own to say the same true thing.

We build every site this way now, and it turned from a technical footnote into part of the pitch. One descriptive headline per page, the named entities spelled the same way everywhere, the structure clean enough that a model can read it without guessing. The whole job of the studio has always been closing the gap between what a company is and what its surfaces say, and the surface that matters most this year is the one being read by something that cannot ask a follow-up question.

Run the test this week, then reply and send me the sentence the model gave back for your company. I will tell you whether it is describing you, or just describing your category with your name attached to it.