A critical review looks for weak points, not nicer wording. AI drafts can sound complete while missing evidence, context or risk markers.
Run a second pass that checks facts, assumptions, tone risks and approval needs.
Start with a narrow boundary: which website, space, file, recipient or decision is affected? This makes the task reviewable instead of turning it into a broad catch-all request.
A useful work order is: “Critically review this answer for false claims, unsupported assumptions, missing data, tone risks and approval needs.” For important cases, add that uncertainties must be marked visibly instead of being filled in silently.
Pay special attention to risk, review duty, privacy and later findability. These points decide whether the result is only useful for the moment or can be found, checked and continued by the team later.
Do not repeat the same drafting prompt and call it review.
The final output improves because it was challenged before use.