Why can an answer sound right and still be made up?
Confabulation / hallucination
Generative AI can produce false or unsupported content with the appearance of a factual answer. Fluent language does not establish truth.
An everyday example
A fictional supplier summary includes a cancellation date that never appears in the supplied agreement.
Original illustration, not an observed participant case.
Where the idea stops
Not every error has the same cause. Bad source data, arithmetic mistakes and missing context need different repairs.
Something to try
Pick one consequential claim. Locate the exact supporting passage or record; mark it unsupported if you cannot.
Practise on a fictional case →Go deeper: the builder’s view
For builders: test grounded claims and unsupported additions separately. A disclaimer at the end does not validate the answer.
Source and interpretation
This is a plain-language interpretation with an applied example; it is not a quotation or an endorsement from the source.
NIST — Generative Artificial Intelligence Profile, section 2.2 (2024) ↗Catalog v0.1 · checked 2026-10-29