AI hallucination describes an answer that sounds credible but contains false or unsupported information. NIST’s Generative AI Profile calls this risk confabulation and includes confidently stated erroneous content. Fluency should therefore be kept separate from factual verification. Source: NIST: Generative AI Profile, AI 600-1

A travel example

Imagine an assistant invents a hotel transfer time because its approved material does not contain one. A useful response would identify the missing information and offer a suitable next step. This example illustrates an evaluation case, not an incident involving a particular product.

How to test for it

Ask questions with known answers, questions with no supported answer and questions whose premise is deliberately wrong. Compare each answer with its cited source. Record whether the assistant corrected the premise, admitted the gap or supplied an unsupported detail.

Keep a correction record

We recommend recording the question, answer, source and action taken. Repeat the test after a source or configuration changes. This gives the team evidence of how a problem was addressed instead of relying on a broad claim that an assistant is accurate.

Written and updated by Travelgenix. Sources checked on 16 October 2026. Worked examples and checklists are editorial illustrations and recommendations, not customer results.