Rankfor.AI research · August 2026
We asked AI about 264 European companies.
For half of them it makes things up.
9,600 answers · 3 AI assistants · 21 markets · web search off, nothing looked up
FT1000Companies drawn from the FT 1000: Europe's Fastest Growing Companies 2026
The question
Does your company name change what AI says about you?
Peec AI showed that it does. Jan Ehrlinspiel published the study on 11 August 2026, and every invented name in it was a real English word.
Peec AI had no part in this study. We used the same word scale, so the numbers can be compared.
What Peec found
904 real brands, plus 80 invented name pairs. A name one point higher on the word scale is described 0.07 points more positively. The effect disappears once the model knows the brand.
What stayed open
Is it the meaning of the word, or the shape of it? Does any of it cross a language boundary?
Test one
We invented 15 companies so the name was the only thing to go on
Poison Inc. Freedom Inc. Misery Inc. None of them exist, so the AI has nothing to describe except the word.
We scored every word it came back with. The scale came from outside this study: 1,800 people rated 13,915 English words.
Murder sits near 1. Door and engine sit around 5. Vacation sits near 9.
3.1 points is the distance between how the scale rates "funeral" and how it rates "holiday". The name is the only difference.
Test two
Is it the meaning of the word, or the sound of it?
For every name we built a twin one letter away.
Poison Inc. and Poisom Inc. Freedom Inc. and Freedon Inc.
They sound almost identical. Only one is a word.
So the meaning moves the answer. Change one letter and most of the effect goes away, though what the last of it is, fifteen names could not settle.
Test three
A Polish name carries its Polish meaning into an English answer
Trucizna means poison. We asked in English about Trucizna Inc.
Trucizna Inc. → lethal, toxic, chemical, potent, dangerous
Poison Inc. → toxic, lethal, dangerous, potent, chemical
If your company name means something in your own language, one of these assistants reads it and one does not. Your buyer picks which.
What did not work
On real companies the effect is much weaker, and we cannot separate it
We ran the same test on 264 real European companies. The link between name and description is there but small, and it does not depend on whether the AI knows the company.
The reason is simple. Real names describe the business. Coffi Lab sells coffee. Baroni Home sells homeware. So a pleasant name and a pleasant business arrive together and no measurement can tell them apart.
We are reporting this because leaving it out would make the rest look stronger than it is.
What we found instead
Half of Europe's fastest-growing companies get put in the wrong industry
For every company we checked one thing that has a right answer: does the AI place it in the industry it actually operates in? We took the industry from the FT's own record, and we accepted any near-enough word.
Every one of them has real revenue and real staff. n = 175
- right industry in 90% of answers or more
- right industry in 50% to 89% of answers
- right industry in fewer than half the answers
Which companies
The assistants are reliable about software and guess at the rest
Split the same 175 companies by what they sell. Software companies land in the right industry in 86% of answers. Every other sector averages 64%, and a third of those companies land right in fewer than half.
Sector taken from the FT's own record. The full table for all 175 companies is published with the data.
Eight of the 175, from the company no answer places right to the one every answer does.
The part that should worry you
All three are wrong about a third of the time. One of them says so.
In the same 8,250 answers we counted two things: how often an assistant puts the company in the wrong industry, and how often it admits it has not heard of the company.
Clever is a Danish energy company. All three assistants describe it fluently and confidently. Two of them are reading the English word "clever" and guessing from there. Nothing in the answer tells the reader which is which.
This counts whether an assistant flags its own uncertainty. Claude is the least accurate of the three here. It is also the only one that tells you.
puts the company in the wrong industry
says it does not know the company
What to do
The question is whether the AI knows you
If you run a company
Until the AI knows you, it describes your name. Being known is the fix, and it is the only fix.
If you are reading an AI answer
Confidence is not knowledge. The assistants sound the same whether they know the company or are guessing from the letters.
Method
How this was measured, and what it cannot tell you
- 9,600 answers. 3 assistants, latest version of each, August 2026. Web search off, so nothing was looked up while answering.
- Every question asked for ten one-word descriptions plus one specific fact. We scored each word on the Warriner, Kuperman and Brysbaert (2013) list, the same 13,915-word scale the Peec study used.
- Every company was asked about ten times, because a single AI answer moves on its own. We measured that movement: 0.28 points on the 1-to-9 scale.
- Zero failed calls in 9,600. A failed call is recorded as a failure, never as an empty answer, because scoring a failure as "no result" is how measurement artefacts get published.
What this cannot tell you: whether an AI answer is true in detail, how any of this behaves with web search on, or whether it holds outside the 21 markets and 3 assistants tested.
Data, code and the full company list are open.
doi.org/10.5281/zenodo.21904654CC BY 4.0
Try it now
Ask AI about your own company.
The same question we asked about 264 companies takes thirty seconds to ask about yours.
See if AI recommends your brand →