AI won’t fix a brand that doesn’t know what it is

What does a language model do with your brand?

It summarises it. If it cannot find one version, it makes one itself.

Somebody asks ChatGPT “which hotel in Halkidiki for couples”, or Gemini “a good diagnostic laboratory in Thessaloniki”. The answer does not come from a ranked list. It is a synthesis of what your website says, along with the reviews, the directories, the articles and your profiles on the platforms. The engine is looking for a pattern: which word recurs, and which attribute more than one source confirms.

When the pattern is clear, the description comes out clear. When it is not, the engine does what a hurried reader would do and generalises. “A boutique hotel with a focus on wellness” becomes simply “a hotel in Halkidiki”. A specialism that took years to build disappears inside one sentence.

Why does technical optimisation not solve it?

Because it amplifies whatever it finds. It does not decide what deserves amplifying.

All the technical work is needed: structured data, proper access for crawlers, content that answers real questions. But it all does the same thing. It carries your message to the engine more reliably.

If the message is vague, it carries vagueness — only more reliably.

I see it often in projects that reach us after somebody has already “done AEO”. The schema is correct and declares the company a business consultancy. The homepage talks about digital transformation. The service pages read like an agency. LinkedIn says “software solutions”. Technically nothing is wrong. Strategically, there are four companies sharing a name.

“How does the engine find me” is a technical question. “What does it find when it gets there” is a positioning question, and it has to be answered first.

How do you know your brand does not know what it is?

From the distance between what you say and what others say about you.

An identity is rarely missing altogether. Usually a company has several at once. Certain signs come up again and again:

• The descriptions on the site, the Google Business Profile and LinkedIn were written in different years, by different people, for different reasons.
• The sales team describes the company differently from the website. Often they are right, because they know what customers actually buy.
• The homepage lists services instead of saying what result the company delivers. When twelve services carry equal weight, none of them is the main one.
• The reviews praise something the company mentions nowhere about itself.
• There was a rebrand, but the directories, the old press releases and the job ads still carry the old description.

That last one is worth dwelling on. Job ads and careers pages are among the most neglected texts a company produces, and among the most public. An ad about “a dynamic, fast-growing business in the sector” tells the engine nothing. Employer branding belongs to the same entity as the brand. A model does not separate what you write for customers from what you write for candidates.

Why does vagueness cost more in AI than in Google?

Because in Google the user chooses, and in AI the engine chooses.

In classic search results, a vague company could survive. It appeared among the ten links, and the user clicked, read and judged for themselves. Clarity helped, but it was not a precondition.

In an AI answer, the judgement has been made before the user sees it. The engine puts forward two or three names and writes a sentence for each explaining why. To recommend you, it has to be able to write that sentence with confidence. If it cannot, it will write it for a competitor it can describe.

That changes who wins. Not necessarily the largest, or whoever has the biggest budget. Often it is whoever is easiest to describe. For smaller companies it is one of the few times size does not decide the outcome. Clarity does.

Where does the work start?

With one sentence the whole company says the same way.

Before any optimisation, we ask the client’s senior people to answer four questions in writing, each of them separately:

• What are we, in one sentence?
• Who for?
• Why us and not the next one?
• What are we not?

The fourth is the hardest and the most useful. A company that cannot say what it does not do will be described as a company that does everything. And nobody recommends a company that does everything.

In the first round the answers almost never match. That is not a failure. It is the project’s first finding, and usually its most valuable. If three executives describe the company three ways, we cannot expect a language model to settle on one.

After the sentence comes the vocabulary: which word we use for our category and which ones we avoid. “Diagnostic laboratory” or “diagnostic centre”? “Resort” or “hotel”? It looks like a detail. For an engine looking for repetition, it decides whether it sees one entity or two.

Only then does the message go everywhere, in the same sentence and the same words: on the homepage, the service pages, the structured data, the Google Business Profile, the directories, the careers page, the executive biographies and the press releases.

So where does the technical work come in?

Immediately after. It matters no less, but without a clear message it has nothing to carry.

At PHOENIX VERUS we have divided the work this way deliberately. I take positioning and message. Taxiarchis Vafeas measures how AI systems describe the company before anything changes, and fixes what stops the engines reaching the content. The two run in parallel, but in order. First we take the baseline, then we lock the positioning, so every optimisation that follows is aimed at something specific.

The order matters for a simple reason. Fix the technical work first and then change the positioning, and you will do the technical work twice. Change the positioning without a baseline, and you will never learn whether the change worked.

What can you check yourself this week?

Whether you are described the same way everywhere. It takes an hour.

Open a document and copy your company description word for word from five places: the homepage, the Google Business Profile, the LinkedIn page, the most recent job ad, and the most important directory or platform in your sector.

Put the descriptions one under the other. If it takes you more than ten seconds to see that they are talking about the same company, the engine will have the same problem.

You can also ask an AI system “what is [your company]?”. You will get an indication rather than a measurement, because these systems answer differently each time and one answer has nothing to be compared against. But if the indication does not resemble the sentence you would want to hear, you know where to start.