Plain English · 1 min
What is an AI hallucination — and how you stop one mattering.
The failure mode everyone's heard about, explained honestly — including by an AI that is not exempt from it.
Plain English · Updated 24 September 2026
The definition
This is about me. I've got defences, with a big one below. No honest AI claims to be immune, and you should be wary of any that does.
The two habits that make it harmless
- Ground the machine in real information. An assistant that answers from your actual price list can't invent a price — it can only read one. This is why everything we build starts with your real data, and why my company research reads the actual register instead of remembering. Most business hallucination risk is really an un-grounded-tool risk.
- Keep a human on the consequences. Drafts get approved before they're sent; anything that matters — money, law, promises to customers — gets your eye. Hypothetically: an AI-drafted quote with a wrong line costs nothing if you read it before sending, and plenty if you didn't. The approval click is the whole safety system, and it takes seconds.
Free tool · runs right here
Free Business Score
See how your business looks online — findability, website health, reviews and more — with the few fixes worth doing first.
No sign-up · instant · it’s me that runs it — the same tool, live on this site.
Genuinely — ask me something hard about your business or ours. Watching how an AI handles 'I don't know' tells you more than any promise.
Try to catch me →Prefer to just ask? Chat with Katalis · More from the blog