glossary term · cross business
Hallucination
Learn what hallucination means through a plain-language definition, a synthetic business example, practical limits, and links to workflow guides.
Definition
In AI output, hallucination means generated content that presents an invented or unsupported claim as though it were factual. It can appear in a polished sentence, a source reference, or a statement that an action was completed. Reviewers must compare material claims with approved evidence and actual system state; confident wording is not evidence of truth.
Owner example
Synthetic example: Meeting notes say someone will check Friday availability. The draft says a Friday start is confirmed. The reviewer rejects the invented commitment and preserves the unresolved question.
Non-example and limitation
A clearly labeled hypothetical example is not pretending to report a fact. An explicit “unknown” is also different from supplying a guessed deadline. Even a source-backed statement can be wrong if the source itself is inaccurate.
Related terms
Evaluation tests for unsupported claims. Retrieval-augmented generation supplies evidence but does not eliminate the need to check it.
Deeper guide and practical link
Read checking follow-up commitments for the fuller explanation. Use the illustrative draft-testing workflow to apply the distinction to a bounded task.
Sources and scope
This is an original editorial definition with a synthetic example, not a report of a tested implementation or business result. Suggest a correction.
Change notes
Foundation draft prepared October 20, 2026. No public publication or successful live-tool test is claimed.
Next learning step: Check AI-generated commitments before you send a follow-up