Definition
Hallucinations happen when a model fills gaps with likely-sounding text rather than verified facts. They are especially risky because the output can be polished and confident.
How it works
Generative models predict likely tokens. If the prompt lacks reliable context or the task asks for unavailable facts, the model may produce an answer that matches language patterns but not reality.
Why it matters at work
Hallucinations create legal, compliance, brand, and operational risk. Teams need verification habits before AI-generated work reaches customers, regulators, or internal decision makers.
Workplace example
A finance analyst asks AI for a customer statistic. The model invents a number, so the analyst checks the source system before adding it to an executive deck.
Frequently Asked Questions
How do you prevent AI hallucinations?
You reduce hallucinations with clear prompts, trusted source context, RAG, citations, evaluation rubrics, and human review for high-risk work. You cannot eliminate them entirely.