Definition
AI Hallucination
An AI hallucination is when an AI model produces information that sounds confident and plausible but is false or not supported by its sources.
Large language models generate text by predicting likely words, not by looking facts up in a database. When they lack reliable information, they can invent details, such as a wrong price, a non-existent policy or a fake citation, while sounding certain.
Businesses reduce hallucinations by grounding AI in their own approved information (for example with retrieval-augmented generation), telling it to say when it doesn't know, showing sources, and having people check important outputs.
An AI hallucination in practice
A customer asks a website chatbot about a refund policy. Without access to the real policy, the bot invents a 60-day refund window. Connecting the bot to the actual policy document and requiring it to cite its source prevents the mistake.
Related terms
Common questions about AI hallucinations
Why do AI models hallucinate?
Because they generate the most likely text based on patterns, rather than checking facts. Gaps in their information are filled with plausible-sounding guesses.
Can hallucinations be eliminated completely?
Not completely, but they can be greatly reduced with grounding in approved sources, clear instructions, testing and human review.
How does RAG reduce hallucinations?
Retrieval-augmented generation gives the model relevant passages from trusted documents before it answers, so responses are based on real information.
Should businesses still use AI if it can hallucinate?
Yes, with the right design and checks. Many tasks benefit from AI as long as important outputs are verified.
