Knowing it wrong is worse. If AI knows nothing about a business, it simply doesn’t mention it, and the customer goes to whoever the system did find. That is a loss, but an honest one. If AI knows a business wrongly, however, it doesn’t stay silent: it draws a conclusion about the company and then uses that conclusion to decide who it should be recommended to and who it shouldn’t. The owner sees none of this, because nobody sends them a report on how the machine has understood them.
Doesn’t know you — stays silent. Knows you wrong — decides for you.
A wrong picture changes the recommendation, not just the description
A mistake in the street name or the opening hours is annoying, but it is easy to spot. A mistake about what the company does, who it works for and how big it is, is more dangerous. These are exactly the facts AI uses to answer a customer’s question “who should I go to?”, so a misunderstood specialism or scale leads to the wrong commercial conclusion.
We saw this happen to one Polish company. Google’s answers effectively assembled a different company in its place: it attributed offices in Malmö and Stockholm, large international clients, a big development team and a specialism in blockchain and Web3. It then drew the conclusion that followed logically from that picture: the company looked too large and too technical to be recommended to a small Polish service business. In other words, the error hit precisely the customers the company exists to serve.
Around the same time, other systems got it wrong in their own ways. One attributed another country’s projects to the company and decided it worked mainly in that market. Another described a different set of services and stated that small local businesses were not its clients. Three systems, three different invented companies — and none of them matched the real one.
When the owner saw these answers, the first reaction was not curiosity but worry: can a company really just be credited with something it doesn’t have? And if that is possible, an invention like this could do serious damage to its reputation.
When AI doesn’t know a business, it simply drops out of the answer
We know the second situation from our own work as well. In one measurement for a window service in Poznań, Gemini and Perplexity did not mention the company in their answers at all. For a customer asking them for advice, the business simply didn’t exist.
That is also a loss of customers, but with one important difference: nothing is distorted. The system has nothing to correct; it needs to learn, for the first time, who the company is, what it does and who it works for. So the task is clearer: to build the right picture from scratch, step by step.
Fixing a wrong picture takes longer than building from scratch
Once a wrong picture has formed, building the right one comes with an extra task: working out where the wrong one came from. As long as the source of the confusion is still there, the system will keep going back to it. That is why there is more work involved, and why it has to be checked: fix, then run a new measurement to see whether anything in the answers has changed.
This is why I see a wrong picture as the harder case. It seems to me that building without mistakes is always more rewarding than first untangling someone else’s mistake and only then building.
AI confuses companies because it matches data from different sources
So where does this kind of confusion come from? Systems have to decide which pieces of information from different places belong to one and the same real company. In the help pages for its Knowledge Graph, Google calls this process matching, or reconciliation, and describes separately where it breaks down: when there isn’t enough information to match, when one entity almost matches another, or when several records start matching the same entity.
Researchers who study this problem reach the same point. A 2026 survey in the journal Data & Knowledge Engineering describes entity matching as identifying which records from different sources refer to the same real-world entity. The authors show that it becomes especially difficult when sources are structured differently and describe the same thing in different ways. For local businesses this is common: the Google Business Profile, the website, social media and directories all tell the company’s story differently, and sometimes the company is described by people it has never even heard of.
After a month of work, no system confused the company any more
This story could have ended on a worrying note, but it ends differently. We worked on how the systems understand this company, and a month later we measured again. Every system described it correctly. Google no longer attributed Swedish offices to it and even pointed out that a company with a similar name in Sweden is a different firm.
That leads to the main takeaway for owners. A wrong picture is invisible until the company has been checked, but it can be fixed. The problem is not that the error exists; it is that nobody knows about it.
When this isn’t about you
If a business is very new and has almost no digital history yet, that is not just a weakness. It is also a chance to build that history correctly from the start: to explain consistently who you are, what you do, who you work for and what people should come to you for. The earlier a business starts building a consistent digital footprint, the lower the risk of later having to deal separately with a wrong picture that has already taken hold.
And an honest caveat: nobody can guarantee what a particular model will know about a company. What you can do is check what the systems are saying now and work with what they build their answers from.
Do you know which company ChatGPT, Gemini and Google see when they talk about yours — and is it really your company?
Sources: Google Search Console Help — Debugging entity matching issues; Data & Knowledge Engineering — Heterogeneity in entity matching: A survey and experimental analysis (2026).
To see whether your business already appears in AI answers, see the AI Visibility Audit
Want to check whether AI already recommends your business?
We start with a diagnosis of the company’s current visibility and real customer decision scenarios.
VIEW THE AUDIT