Glossary
AI hallucination
An AI hallucination is output from an AI language model that sounds plausible and convincing but is factually wrong or entirely made up. For small and medium-sized businesses this matters because such errors can slip unnoticed into quotes, customer replies, or internal research. The most effective countermeasure is grounding answers in verified sources.
Last updated: 2026-08-03
Language models generate text by predicting the statistically most likely continuation – they do not check whether a statement is true. When a model lacks knowledge about a question, it fills the gap with plausible but invented content: fabricated citations, wrong figures, blended facts, or details that never existed. What makes this especially treacherous is the confident tone – hallucinated answers sound just as convincing as correct ones. Hallucinations are not a flaw of any single vendor but an inherent property of generative AI: they can be reduced, but never fully eliminated.
In everyday business, this creates concrete risks: a customer reply with incorrect product details, a contract summary citing a clause that isn't in the document, or an internal answer about a policy that never existed. Reusing such output unchecked risks bad decisions and lost customer trust. A practical rule of thumb: AI answers without a traceable source should always be reviewed by a knowledgeable person before they are used – the higher the stakes, the more thorough the review.
chifty Knowledge helps ground answers in approved information. Sources support review but cannot exclude incorrect or unsupported statements. When evidence is missing, clarify the issue before making a binding reply. Check whether the cited passage actually supports the specific claim.