The verification loop is the habit that separates AI use from AI dependence. Every non-trivial AI output is verified against primary sources, checked for internal consistency, and stress-tested with adversarial prompts before it is trusted enough to act on.
- 01State the three verification steps.
- 02Design a personal verification rule.
- 03Recognize when the verification loop is being skipped.
The three steps
1. Source verification: every factual claim traced to a primary source you have read. 2. Consistency check: does the output contradict itself, or contradict adjacent output? 3. Adversarial prompt: 'What would someone who disagreed with this argue?' If the AI cannot answer, be skeptical.
Personal rules
Rule of thumb: the higher the stakes and the more novel the domain, the more verification. For low-stakes drafting, one pass. For a decision you will act on, all three. For public writing, all three plus outside review.
- Verification loops are habits, not decisions.
- The higher the stakes, the more paranoid the verification.