Hallucination
Also called AI hallucination, Confabulation
A hallucination is an output from an AI model that sounds confident and plausible but is false or unsupported by its sources, such as an invented fact, quote, price, policy or link. It happens because language models generate likely-sounding text and have no built-in way to know when they lack information.
Example
Asked about a bereavement discount, a chatbot describes a refund rule that the company never offered, complete with a deadline.
Why it matters for a website chatbot
On a business website, a hallucinated policy can become a promise you're held to. Grounding answers in your content reduces the risk. ChatBento is instructed to answer only from your content and to say plainly when it doesn't know, never guessing prices, policies, dates, links or features.
See it in practice: Pre-sales questions
Related terms
- GroundingGrounding means tying an AI model's answer to a specific, trusted source of information, such as retrieved documents, a database or search results, so its claims can be traced and checked. A grounded answer says only what its sources support, and ideally shows which source each part came from.
- GuardrailsGuardrails are rules and checks placed around an AI system to keep its behavior within safe, intended limits. They can restrict topics, block harmful or off-brand output, require answers to come from approved sources, filter inputs for attacks, and send uncertain or sensitive cases to a person.
- Fallback messageA fallback message is what a chatbot says when it can't understand a request or doesn't have an answer. Good fallbacks admit the limit plainly and offer a next step, such as rephrasing, a link or contact with a person, instead of guessing or repeating the same unhelpful reply.
- Retrieval-augmented generation (RAG)Retrieval-augmented generation (RAG) is a technique in which an AI system first retrieves relevant passages from a chosen set of documents, then gives them to a language model to write its answer. It lets the model answer from current, specific information it was never trained on, and makes answers easier to check.