Most of the writing about AI in business assumes the risk is hallucination. The model makes something up about you, prospects believe it, your reputation takes the hit. It's a tidy story. It's also, looking at the data from real recommendation queries, almost entirely wrong.
What the data actually says
Across 669 AI mentions of Northern Irish businesses across ChatGPT, Claude, Gemini and Perplexity, the facts the assistants stated were right roughly 99% of the time. One material error in the entire dataset. A handful of minor slips (a wrong founding year, a misremembered service). Overwhelmingly, when AI talked about a business, it got the business right.
If hallucination were the real risk, that number would be 70%, 80%, somewhere worrying. It's not. It's 99%. Which makes the actual finding much harder to explain away.
The real risk is being skipped, not described badly
The same scan ran 1,296 answer-engine queries across 36 NI businesses. On open buyer questions where the business name wasn't in the prompt, a business got mentioned 19.2% of the time. Four times out of five, the assistant reached for someone else.
That's the uncomfortable part. The risk isn't that AI describes you badly. It's that, when a stranger asks AI for a recommendation in your category, four times out of five you're not in the answer. You're not being misrepresented. You're not being represented at all.
Why this is harder to fix than hallucination
Hallucinations have a satisfying fix. Write the correct version somewhere AI can read it. Tighten your structured data. Get your facts onto Wikipedia or Wikidata. Push the right answer high enough that it surfaces. There's a thing to do and a way to know you've done it.
Absence has no equivalent fix. You can't write your way into being mentioned. You can't tell AI to consider you. You can only make yourself harder to leave out, by showing up consistently in the sources AI already trusts. That's the slow work of footprint, not the quick work of metadata.
What to do about it
First, find out which one you have. Run your business across the four major AI assistants on the questions your customers actually ask. If you appear and the description is wrong, you have a small problem with a clear fix. If you don't appear at all, you have a bigger problem with a slower fix, and you've spotted it before most of your competitors will.
Then prioritise based on which gap is biting. Most businesses we see have absence problems, not accuracy ones. The work is getting cited, listed and reviewed in the places AI looks for your category, until the signal accumulates enough to flip the answer.
Read the full research for the accuracy figures by sector and the source families the assistants leaned on across hospitality, accountancy and agencies.
See where you stand
Run the free AI visibility check. It tells you in about a minute whether the problem you have is being described badly, or not being described at all.