A hallucination is an answer from a large language model that is presented with confidence and is still untrue. A language model predicts which text is likely to come next. That usually produces fluent sentences, but fluent does not mean correct. OpenAI writes in its article Why language models hallucinate of 5 September 2025 that common training and evaluation reward guessing over admitting uncertainty.
Why does it matter?
An AI assistant can claim something about your business that is wrong: an incorrect price, a service you do not offer or an address from years ago. A customer who reads it has no way of knowing it is wrong. Live search helps: Google writes about grounding that it can reduce hallucinations. It does not rule them out.
Example
Suppose someone asks an AI assistant whether a hair salon in Breda is open on Sundays. The answer says yes, but the salon is closed on Sundays. Perhaps the assistant found no current opening hours, or took details from an outdated profile.
How do you check it?
Regularly ask the AI assistants your customers use the same questions about your business, and compare the answers with your own details. Make sure prices, services, address and opening hours are correct and match on your website and in your business profiles.
Common mistake
Publishing text written by an AI tool without checking it. Read in Can you publish AI-written content? What Google says why you should always review AI content yourself.