What is hallucination (in AI)?
When an AI states something that sounds plausible but isn't true.
Hallucination (in AI), explained
On a business call, a hallucinated price or policy is a real problem. Grounding answers in a knowledge base, limiting topics and escalating unknown questions all reduce the risk.
The safest voice agents are told exactly what they may answer, look up facts instead of recalling them, and say "let me connect you to the team" when they don't know. Reviewing call transcripts regularly catches mistakes early.
See how these ideas come together in Chitti, Manolle's AI voice agent.
Related terms
- GuardrailsRules and checks that keep an AI agent within safe, approved behaviour.
- Retrieval-augmented generation (RAG)A technique where an AI looks up relevant information before answering, instead of relying only on what it learned in training.
- Knowledge baseThe business information a voice agent answers from, such as prices, timings, policies and FAQs.
See an AI voice agent at work on your calls.
See the complete system live: marketing, lead capture, AI voice calling, sales conversion, and analytics.