.png)
Module 2.2
Verify before you rely on it
AI is fast and often right — which is exactly why a wrong answer can slip through unquestioned. An output can look correct and be wrong. It can cite sources that do not exist. It can omit the one fact that changes the answer.
- Identify the specific claim or output you are relying on.
- Identify the consequence if it is wrong.
- Check the claim against a primary source you already trust.
- Ask the AI what is uncertain, what assumptions were made, and what is missing.
- Run the same query with different wording and compare outputs. Agreement across rewordings narrows the odds of an obvious error — it does not confirm the answer is true. Only the primary-source check (step 3) does that.
- Record the human decision, not just the AI output.
A quick judgment check before you rely on it
- Is my source for step 3 actually independent — not the same AI, and not the material the AI drew from?
- Am I trusting this because I checked it, or because it sounds fluent and confident?
- Would I put my own name to this as my decision, not the AI’s?
- If it is wrong and no one catches it, who besides me is affected?
Where verification would have flipped the decision.
In 2023 a lawyer filed a court brief citing several cases an AI had supplied. The citations were fluent, correctly formatted, and confidently worded — and the cases did not exist. One primary-source check (step 3), looking up a single citation in a court database, would have stopped the filing. Nobody checked; opposing counsel and the judge did.
Source: Mata v. Avianca, Inc. (S.D.N.Y. 2023). Full record in the Appendix.
Plausible output is not validated output. Build your verification step before the result reaches a decision — not after something breaks. The difference is invisible until it costs you, and it applies to any high-stakes use.