Verifying an AI answer means checking the facts, figures, sources, and citations before using it, because a model can produce falsehoods with confidence. The level of checking depends on the stakes: the greater the consequence of an error, the more the verification must rely on reliable, independent sources.
Example
Before sending a report, cross-check every figure cited by the AI against the source document.