Mixed up: when AI models gave us a Polish company's prices

Anyone who offers AI visibility analyses should check their own website first. That is exactly what we did: we measured aivisible.ch with the same twelve AI models we query for our clients. The result was instructive, though not where we expected it.

What we measured

Technically, our website did well. Our own AIR score, AI Readability on a scale from 0 to 100, came out at 86 out of 100. AI crawlers can fetch the pages, the text is readable without JavaScript, and structured data and an LLM file are in place.

The interesting part was the models' answers. Several AI models did not describe AIVISIBLE as what we are. Instead, they mixed us up with other businesses that carry a similar name: aivisible.ai, aivisible.agency and, in one case, a reference to Tallinn, Estonia. None of these businesses is connected to us. They simply share a similar name and work in a related field.

The IBM Granite case

It became especially clear in a test on 1 November 2026. Asked about AIVISIBLE, IBM Granite answered with a phone number starting with +48, an email address and prices in Polish zloty. These details belong to the Polish provider aivisible.pl, an independent company we are not affiliated with.

For someone looking us up, the answer looked plausible. It was fluently written and named concrete contact details and concrete prices. It just was not correct. Anyone calling or writing would have reached a different company, and anyone comparing prices would have worked with figures that are not ours.

To be clear: the other companies did nothing wrong. They run their own offerings under their own domains. The mix-up does not happen on their side but inside the models.

Why AI models confuse by name, not by domain

To a person, aivisible.ch and aivisible.pl are obviously two different addresses. To a language model, that is far less obvious. Models learn from large amounts of text in which companies appear mainly by name and much less often by their full domain. In training, the name AIVISIBLE therefore stands for a blend of everything ever written under that word.

When asked about a company, a model assembles an answer from these fragments. Without a clear signal as to which AIVISIBLE is meant, it falls back on whatever is most strongly represented in its data: a phone number here, a price list there, a location from a third context. The result reads coherently but is a mosaic of different companies.

On top of that, many models do not search the web live. They answer from their training data. Even if your own website contains every detail correctly, that does not mean every model knows it or attributes it correctly. Models with web search can pick up new information faster; models without it only with a new training cut-off.

Similar names are common in business. Short, descriptive brands and names that exist in several countries under different domain endings are especially at risk. That applies not only to tech firms but equally to trades businesses, law firms or practices with common names.

What a mix-up costs a company

A wrong AI answer is not just an image problem. It has practical consequences:

• Prospects get the wrong contact details and reach a different company.
• Wrong prices in a foreign currency set expectations unrelated to your actual offering.
• Other firms' services are attributed to you, or your own services are missing entirely.
• A wrong location makes a company look less relevant for its own market, for example when a provider from the German-speaking DACH region is suddenly placed in the Baltics.

The tricky part: the company itself usually never notices. There is no error message and no warning. The wrong answer appears in a prospect's chat window and disappears again. The problem only becomes visible when you query the models deliberately and compare the answers with your own facts.

What we did

We did not treat the mix-up as a mishap but as a reason to apply our own recommendations consistently. Three measures came first.

First, a disambiguation sentence. Our legal notice and our LLM file now state explicitly: AIVISIBLE (aivisible.ch) is not affiliated with aivisible.ai, aivisible.agency or aivisible.pl. The sentence is factual, names the domains and leaves no room for interpretation. It gives models and search systems a clear signal that these are different companies.

Second, a fact block. We bundled our core details in one place: name, domain, registered office in Altendorf, Switzerland, email, services and prices in euros. The fact block serves two purposes. It is a reliable source for AI systems, and it is our reference against which we check every model answer.

Third, structured data. On the website we also describe AIVISIBLE in machine-readable form using Schema.org markup in JSON-LD: organisation, address, contact, offers. That way the correct facts are not only in the running text but in a form machines can evaluate unambiguously.

Whether and when individual models adopt these signals, we cannot guarantee. The models are operated by third parties. What we can do is make sure the correct information is clear, consistent and easy to find, and measure regularly whether the answers change.

Checklist: how to check your own company

• Ask several AI assistants directly about your company and note which contact details, prices and locations they give.
• Look for companies with similar names, including other domain endings and other countries.
• Record your core facts in a fact block: name, domain, address, contact, services, prices and currency.
• If there is a risk of confusion, add a factual disambiguation sentence to your legal notice and your LLM file.
• Provide Schema.org structured data, at least for organisation, address and contact.
• Make sure the same details are identical everywhere: website, business directories, profiles.
• Repeat the check at regular intervals, because models and their training data change.

Where does your company stand?

Whether AI crawlers can read your website and whether an LLM file and structured data are in place, our free Quick Check shows you in moments. It is a good first step to see where you stand.

In the next episode of this series we explain what GEO is and why a good Google ranking alone is no longer enough.