What is Hallucination?
Hallucination: When an AI model produces confident output that is false or not supported by its sources, such as invented facts, citations, or policy details.
A hallucination is output that sounds plausible but is wrong or made up. Models generate the most likely continuation of text, not verified truth, so when they lack information they can fill the gap with something that fits the pattern: a refund policy that does not exist, a citation to a paper nobody wrote, an order status the system never returned.
Hallucinations are reduced, not eliminated, by grounding the model in retrieved sources, instructing it to say when it does not know, requiring citations, constraining output with schemas, and checking answers against data before they reach a customer. Newer models hallucinate less than older ones, but no production system should assume a zero rate.
Why it matters when you build or buy AI
The question is not "does it hallucinate" but "what happens when it does." Design for it: a support bot should hand off when it cannot cite a source, and an agent that issues refunds should check amounts against the order system before acting. Any workflow where a wrong answer emails a customer or moves money needs a check or a human approval step.
Example
A support bot is asked about a warranty that the help center never mentions. Instead of inventing terms, it says it does not have that information and offers a human handoff.
Related terms
More in AI & LLMs
Book 20 minutes, or send a brief
A former CTO replies. You leave with a written plan and a fixed estimate.
Prefer LinkedIn? Connect with Harish.