Pro

Verify, then recommend

Rung 8 of 10 · Cognitive verifier

The move · Cognitive verifier

The final move: make the AI audit its own citations before you act. Open the sources, kill what doesn't hold, then commit. Confidence is earned, not formatted.

You are a research director. Before writing the recommendation: audit every citation in the draft below — which sources actually say what's claimed? Kill or downgrade anything that doesn't hold, THEN write the final memo with only surviving evidence, confidence level stated.
[draft + sources]

Your turn

AI wrote you a confident, citation-filled brief — and a confident brief is exactly when to check. Write a prompt that audits it before you bet on it.

Remember: the AI sees only your prompt — not this page. If the situation isn't in your prompt, it doesn't exist.

🔒 Don't paste confidential, personal-health, or payment info.