APPLIED GUIDE
How to verify an AI answer
Verification starts by separating claims from fluent wording. Identify what can be checked, return to current primary sources, reproduce calculations, look for omitted constraints, and decide whether the answer is suitable for the actual audience and consequence.
Recommended process
Verification starts by separating claims from fluent wording. Identify what can be checked, return to current primary sources, reproduce calculations, look for omitted constraints, and decide whether the answer is suitable for the actual audience and consequence.
- Extract each factual claim and requested decision.
- Open the original source and confirm date, scope, and wording.
- Recalculate numbers and test edge cases.
Review checklist
Use this checklist before accepting the output or turning it into an action.
- Look for missing alternatives, affected people, and uncertainty.
- Record what was checked and who approved use.
CONCRETE EXAMPLE
Observable result
An AI answer says a policy permits an action. The reviewer opens the current policy, finds an exception omitted by the model, and corrects the recommendation before use.
- Extract each factual claim and requested decision.
- Open the original source and confirm date, scope, and wording.
- Recalculate numbers and test edge cases.
PRIMARY SOURCES
Check the basis for this guide.
NIST · 2023
AI Risk Management Framework 1.0
NIST · 2024
Generative AI Profile — NIST AI 600-1
Frequently asked questions
Is asking the model for sources enough?
No. Open each source and verify that it exists, is current, and supports the exact claim.
Which work task should use AI first?
Choose a frequent, reversible task with clear quality criteria and low consequences if the first draft is wrong.
AI Literacy at Work: Prompt, Verify, Protect
Use generative AI for a real work task with a repeatable verification and privacy checklist.
View the free course