Your brand already has an identity inside AI. Is it the right one?

What is AI already saying about you?

Something. The fact that you have not asked does not mean it is not answering your customers.

Most companies treat AI visibility as something that will begin at some point, once they decide to deal with it. In practice it began already. Every time somebody asks ChatGPT, Gemini or Google’s AI Mode about your sector, your city or your category, the system decides whether to name you and in what words. That decision does not wait until you are ready.

So the question is not whether you will acquire an identity inside AI. You have one. The question is who wrote it. If you did not, it was written by the reviews, the directories, an article from 2019 and whatever else the engine found.

Where does that identity come from?

Two layers, and they do not change at the same speed.

The first is what the model learned during training. That knowledge has an expiry date: the model knows your company as it was when its data was gathered, not as it is today. If you have since changed your name, added services or moved, the model may not know.

The second layer is what it finds when it searches live. More and more systems look up sources at the moment of the question and build the answer on them. That information is more recent, but it depends on what the system finds and which sources it trusts most.

When the two layers disagree, the answer can blend them. When both have gaps, the system can fill them with something plausible that is not true. That is how a company acquires services it does not offer, opening hours that changed two years ago, or a category it does not belong to.

Which wrong identities show up most often?

Five, and they rarely appear on their own.

• The old one. The company before the rebrand, before the change of ownership, or before it changed direction. It is the most common, because old descriptions sit for years in directories, press releases and job ads that nobody remembers to update.
• The narrow one. The company is described through a single service, usually the one mentioned most often online. A diagnostic laboratory offering a full range of tests becomes “the one for blood work”. A hotel group with three very different properties becomes just one of them.
• The generic one. The category is right, but every point of difference is missing. “A hotel in Halkidiki.” “A consulting firm in Thessaloniki.” Accurate, and no reason for anyone to choose you.
• The borrowed one. The description comes from somebody else: a booking platform that classified you by its own criteria, an aggregator with a copied description, or a company with a similar name whose details have been mixed up with yours.
• The absent one. That is an identity too. When a system answers questions in your sector without ever naming you, it has effectively decided you do not belong in the answer.

Why does a wrong identity not correct itself?

Because it tends to confirm itself.

Take a hotel positioned as a quiet retreat for couples, while the platforms and the older reviews describe it as family-friendly. When a parent asks AI about family hotels, the hotel comes up. Families arrive and write reviews about what a good time the children had. Next time the system searches, it finds still more evidence that the hotel is family-friendly. The couple looking for quiet never found it.

A wrong identity works as a loop: it brings in the customers who match it, and they reinforce it with what they write. The longer it goes untouched, the harder it becomes to change.

Waiting, then, is not a neutral choice. Every month without intervention adds evidence in favour of the description that already exists.

Why does getting it wrong cost more than it appears to?

Because the customer does not separate what AI said from what you said.

When a system says you offer a service you do not, or puts your prices at a different level, the customer arrives with the wrong expectation. When that expectation is not met, the disappointment lands on you rather than on the machine, even though the company never supplied the information.

Wrong positioning costs in quieter ways too. It puts you up against companies that are not your real competitors, often on price instead of value. It produces enquiries your sales team has to correct before the real conversation can start. And it brings in customers who do not fit what you do best, so they stay a shorter time, pay less, and write reviews about something other than what you want to be known for.

None of that shows up in a traffic report. It shows up in the margin.

How do you find out which identity you have been given?

By asking what your customers ask, not only “what is my company”.

There are two kinds of question and you need both. The first is about you: what the company is, what it does, who for. Those questions show you your description. The second is about the category: which laboratory, which hotel, which consultant should I choose for what I need. Those questions show whether you appear in the answer at all, and who appears in your place.

The second kind matters more, because that is where decisions are made. A customer who already knows your name rarely asks AI who you are. They ask what to choose.

One question is not a measurement, though. These systems answer differently on every run, and a single answer has nothing to be compared against. What you need is a fixed set of questions, drawn from the language of your actual customers, run repeatedly against each system separately. That is what an AI visibility audit does, and every serious piece of work starts there.

What can you change, and what can you not?

You cannot change what a model remembers. You can change what it finds.

There is no way to reach into the model and correct a description. Anyone promising you that is telling you something untrue. What you can change are the sources the answer is built from — both today, when the system searches live, and tomorrow, when its next version is trained.

Some sources you control completely: the website, the Google Business Profile, your social profiles, the careers page. There, correcting things is a matter of weeks.

Some you influence without controlling: directories, booking and appointment platforms, partners who write about you. There you request corrections, update details, and give everyone the same description to use.

And some you do not control at all: reviews, articles, forum threads. There the only thing you can do is deliver the experience you want described, and ask for reviews from the customers who match your positioning. Those are also the slowest to change, which is why they determine when the difference shows up in the answers.

Where do you start?

By comparing two sentences.

The first is the one you want said about you: what you are, who for, why you. The second is the one AI systems say today, properly measured. The distance between them is the brief. It tells you what has to change, in which sources, and in what order.

If you do not have the first sentence yet, start there. Without it you have nothing to compare the second against, and the correcting is done blind.