How to Review AI Output Before You Use It

A polished AI response can still contain an incorrect date, unsupported claim, missing requirement or subtle change in meaning. Review is not a sign that the tool failed; it is part of a sensible workflow whenever accuracy matters.

1. Check against the source

Compare important claims, numbers, names, dates and commitments with the original material.

2. Check instruction coverage

Use the original task as a checklist. Mark each required point as present, missing or changed.

3. Look for invented detail

Watch for confident statements that are not supported by the source. If the assistant was asked to infer something, distinguish inference from fact.

4. Check audience and tone

A technically correct answer can still be unsuitable for the reader. Remove jargon, unnecessary certainty or wording that sounds unlike the organization.

5. Run a final usability pass

Make sure the output is easy to act on: clear headings, readable formatting, one obvious next step and no accidental placeholders.

Example: reviewing a customer email

Before sending an AI-rewritten email, compare every promised date and price with the source message, confirm the requested action is still present, and check that the tone does not imply a refund, guarantee or commitment that was never approved.

Review the draft below against the source facts.

Return four sections: 1) factual changes, 2) missing requirements, 3) unsupported claims, 4) recommended edits. Do not rewrite the draft yet.

SOURCE:
[paste source]

DRAFT:
[paste AI draft]

Common mistakes

• Reviewing only grammar and spelling.
• Assuming fluent wording means factual accuracy.
• Checking the draft without keeping the original source visible.
• Making important decisions based on an AI answer without independent verification.

FAQ

Should every AI response be reviewed the same way?

No. The review depth should match the consequences of being wrong. A brainstorming list needs less checking than a customer promise or financial summary.

Can AI review its own output?

A second review prompt can find some issues, but it should not replace checking important facts against a trustworthy source.