Build an AI Verification Engine

Step 1 — Capture the original output

Keep the AI-generated answer intact. Verification is harder when the evidence trail has already been rewritten.

Step 2 — Extract checkable claims

Separate factual claims from advice, opinion and stylistic language. Focus verification effort where error matters.

Step 3 — Classify risk

Prioritize claims involving money, health, law, security, deadlines, current events, quotations and important decisions.

Step 4 — Demand provenance

For consequential claims, identify the primary or authoritative source needed to verify them.

Step 5 — Check reasoning

A set of true facts can still lead to a bad conclusion. Ask whether the conclusion follows from the evidence.

Step 6 — Resolve contradictions

When sources disagree, preserve the disagreement and investigate why—date, scope, methodology or definitions may differ.

Step 7 — Human sign-off

The final reviewer decides whether the evidence is sufficient for the intended use. Verification is an approval process, not a confidence score.

Copy the complete blueprint

Customize the bracketed details for your situation, then test it on a real but low-risk task before depending on it.

Act as a verification layer, not a second writer.

Review the AI-generated output I provide and:
1. Extract every factual claim that could be wrong or outdated.
2. Mark each claim: VERIFIED, NEEDS SOURCE, OPINION, ASSUMPTION, or HIGH-RISK.
3. Identify missing context and contradictions.
4. Tell me which claims require a primary source.
5. Check whether the conclusion actually follows from the evidence.
6. List the three most important things a human must verify before using the output.

Do not silently rewrite questionable claims as if they were true.

The Human Edge Check