← All glossary terms

AI & Web Glossary

AI hallucination

An AI hallucination is when a language model states something false with complete confidence — an invented fact, citation, or detail. It is a known failure mode that good system design works around, not a rare glitch.

Language models are trained to produce plausible text, and plausible is not the same as true. When a model doesn't know something, it doesn't naturally say 'I don't know'; it produces something that sounds right. An invented court case, a confident wrong price, a policy your company never had. That's a hallucination, and every language model does it.

The fix isn't hoping for a model that never errs; it's building so errors get caught or can't happen. Ground answers in your real documents through retrieval, so the model quotes rather than recalls. Require citations, so a human can check the source in one click. Keep a person approving anything consequential. Constrain what the system may answer at all. A well-built assistant says 'I don't have that information' precisely because it was engineered to.

Practical example: an assistant that quotes your actual returns policy, with a link to it, can't invent a 90-day window you don't offer. The same model with no grounding might. The difference isn't the AI; it's the system around it. This is most of what separates a production business tool from a demo.

Where this shows up in practice