What is hallucination, and how do we reduce the risk?

A language model produces what appears to be a plausible answer. It does not distinguish between knowing and guessing, and the guess appears just as confident.

Ground the answers in your own sources

An agent that only answers based on your own knowledge articles hallucinates far less than one that answers freely. This is the single most important measure.

Tell it to say when it doesn't know

State it explicitly in the instructions. Otherwise, it will fill in the gaps itself, and that is precisely where errors arise.

Show the source

Answers that reference the article they are based on allow users to verify the information themselves. This does not shift responsibility, but it helps catch errors early.

Human in the loop where it matters

If the answer is going to a customer or could affect an agreement, someone should review it first. For internal lookups, the risk is lower and the review can be less stringent.