AI Output Verification Checklist
Use this before acting on an AI answer.
Before entering material into an AI tool, use a public or invented example. Share confidential, client, or personal material only when the exact tool, account, and data type are approved and you have permission. Removing names alone may not protect it.
Quick Classification
| Output type | Default treatment |
|---|---|
| Draft, rewrite, outline, brainstorm | Edit for quality and fit |
| Summary of provided source | Check coverage, unreadable sections, every consequential claim, number, and quote against the source; keep page or section references |
| Names, dates, numbers, quotes, citations | Verify before use |
| Legal, medical, financial, safety, HR, or customer-impacting advice | Do not act on the model alone; use a qualified professional and current official authority appropriate to the decision |
| Current events, prices, product details, or local rules | Use current search or official references |
Verification Steps
- Highlight every specific claim.
- Identify the source that should prove each claim.
- Check the primary source where possible.
- Remove or qualify claims you cannot verify.
- Keep links or citations with the final output.
- For supplied documents, record which pages or sections were checked and any content the tool could not read.
Go / No-Go
- Go: for a lower-stakes use, the relevant checks are complete and the source actually supports the final wording.
- Hold: claims are plausible but unsourced.
- Escalate: the output affects customers, money, health, legal rights, employment, safety, or regulated data. A citation is not a substitute for qualified review.