Polp
← Back to glossary
AI quality and safety

AI hallucination

A confident-looking AI output that is unsupported, fabricated, or inconsistent with the available evidence.

Business example

An assistant invents a contract renewal date instead of saying that the date is absent from the connected documents.

How it works

  1. 1.A model predicts plausible language.
  2. 2.Missing or ambiguous context increases uncertainty.
  3. 3.Grounding, citations, and abstention reduce business risk.

Common misconceptions

A fluent answer is probably correct.
Fluency is not evidence; important answers need verifiable sources.

Related concepts

Retrieval-Augmented Generation (RAG)Knowledge base

Related reading

The new AI SaaS metric: answer quality, not document count

Sources

Reviewed:

Reviewed by: Javier Chulvi Bernad · LLM Engineer · Madrid