Hallucination (AI)

Language models predict fluent text; they do not inherently verify truth. Hallucinations can invent citations, product features, or statistics. Grounding, retrieval, and human review reduce—but do not eliminate—this risk.

For brands, hallucinations can misrepresent offers or create false associations. Publishing clear, widely echoed facts and official docs helps models and retrieval systems stay accurate. Hallucination is the opposite problem of a desirable Citation.

Do not confuse user error or outdated training cutoffs with intentional deception; the term describes an output failure mode. See Grounding and Citation.