What Is AI Hallucination?
When a language model generates fluent, confident content that is factually wrong or entirely invented.
Language models are trained to produce plausible text, not verified truth. When asked something outside their knowledge, they tend to fill the gap with something that sounds right — an invented API parameter, a fabricated refund policy, a fee schedule that never existed. The fluency is what makes it dangerous: wrong answers read exactly like right ones.
In customer-facing deployments, hallucination is the primary risk. A support assistant that invents policy creates liability; a docs assistant that invents endpoints creates tickets.
Mitigation is architectural, not cosmetic: ground generation in retrieved source content (RAG), require citations so claims are verifiable, score confidence, and refuse to answer when the knowledge genuinely is not there.
How Kelu uses this
Kelu answers only from your indexed content, cites the sources used, and applies confidence scoring — when your sources don't cover a question, it says it doesn't know instead of guessing, and in a helpdesk the ticket goes to your team. Unanswerable questions are logged as knowledge gaps so the missing content gets written.
Related terms
RAG (Retrieval-Augmented Generation)
An AI architecture that retrieves relevant documents from a trusted knowledge source and passes them to a language model, so answers are grounded in real content instead of the model's training data.
Read definitionAbstention
When an AI assistant declines to answer because its sources don't cover the question, and says so plainly instead of writing a confident guess.
Read definitionGrounding
Tying an AI answer to specific source content, so each claim comes from retrieved passages rather than the model's memory and can be checked against them.
Read definitionCitations
Links from an AI answer back to the specific source documents used to generate it, allowing users to verify every claim.
Read definitionEval (LLM Evaluation)
A systematic method for measuring the quality of AI-generated answers — often using a second model as a judge — so quality is tracked rather than assumed.
Read definitionDeflection Rate
The percentage of support questions resolved by self-service — typically an AI assistant — without creating a ticket or reaching a human agent.
Read definition
From definition to deployment
Connect a knowledge source and see cited AI answers on your own content in minutes — free, no credit card required.
Free to start · No credit card required