AI hallucination, explained
Language models predict likely text. When they don't have the fact, they can produce something that looks right, such as a price, a policy or a date that doesn't exist.
For a business, a hallucinated refund policy or price is a real problem. The fixes are to ground answers in the business's own content, show sources, and have the bot say it doesn't know instead of guessing.
Testing with real customer questions and reviewing conversations catches the remaining gaps.
In AgenBase
AgenBase agents answer only from your content, never invent prices or policies, and say so when they can't find something, offering the contact form instead on paid plans, where unanswered questions are also listed in analytics.