What is AI hallucination?
An AI hallucination is when a language model states something false or invented, such as a policy, figure or source, as confidently as a true fact.
Why it matters for a business
A hallucinated refund policy or contract term can cost you money or trust, and it looks just like a correct answer.
More
It happens because the model predicts likely text rather than looking facts up.
You cannot remove the risk entirely, but you can shrink it: ground answers in your documents, limit the topics, and send anything uncertain to a person.
Why it happens
The model produces plausible next words; when it lacks the fact, plausible and true can differ.
How to reduce it
Answer from retrieved documents with citations, let the model say it does not know, and test with tricky questions.
Measure, then trust
Keep a test set of real questions with known answers and rerun it after every prompt or model change.
- InputA question, a message, a document
- ContextYour data, found by search
- Language modelReads, reasons and writes
- ResultAn answer, or an action in your tools
What it costs
Starting prices of our services on this topic, in your currency.
AI hallucination in practice
When an AI model states something false or invented as confidently as if it were a fact.
Example
- A travel company in London tests an assistant that answers questions about bookings.
- Asked about a cancellation fee that is not in its documents, the first version invents one.
- The fix: answers only from the booking terms, a citation under each reply, and ‘I’ll check with the team’ when nothing matches.
Which term is it?
A large language model: an AI model trained on huge amounts of text to read and write language.

From the founder
I want AI that does a real job inside your tools, with a person checking what matters. We test it on your own data before you rely on it.
Anatoly MikhailovFounder of Pazl







