What is an AI hallucination?
An AI hallucination is output that looks credible but is wrong or unsupported, such as an invented paper or a conclusion the cited material does not support.
NIST's Generative AI Profile uses the term confabulation and notes the common use of “hallucination.” Problems can include false statements, departure from the supplied information or contradictions within the same context.
A complete-looking citation can still fail a check
Suppose an answer gives a paper title, named authors and a publication year, but the paper cannot be found. The detail and confident tone are not evidence that the paper exists.
Trace the citation to the original work. Also check whether an existing paper actually supports the sentence attributed to it. A real URL can accompany a mistaken interpretation.
What do search and RAG change?
AI search and RAG can provide external material to an answer. They still require two checks: whether the source is sound and whether the generated answer represents it faithfully.
A language model's ability to write fluently does not establish either. The review of AI-generated sources considers what can happen when questionable material is repeatedly retrieved and cited.
Sources and scope
- NIST AI 600-1: Generative AI Profile, section 2.2 Confabulation
- Microsoft Support: information used to answer a Copilot prompt
Source-language entry records a source check on 2026-10-24. This adaptation preserves that provenance and does not represent a new audit of the cited pages.