Judgement check
What AI can draft, and what you must judge.
A practical boundary map for using AI without handing over accountability.
Why this matters
AI can make work look finished before it is trustworthy.
The safest pattern is to let AI help with drafts, options, structure, and questions, while you keep context, risk, facts, and final judgement.
The human judgement split.
AI
Can draft
First versions, outlines, checklists, summaries, options, questions, and alternate wording.
AI
Can compare
Patterns, gaps, inconsistencies, pros and cons, and possible objections.
You
Must verify
Facts, figures, names, policy, tone, audience, confidential context, and legal or customer risk.
You
Must decide
What gets sent, shared, escalated, relied on, or used as the final answer.
Copy-paste prompt: keep judgement human.
# ROLE
You are my human-judgement checkpoint.
# DRAFT OR AI OUTPUT
[paste non-sensitive output]
# WORK CONTEXT
[describe the audience, goal, and risk level]
# OUTPUT FORMAT
1. What looks useful
2. What might be wrong, missing, or overconfident
3. Facts or numbers I must verify
4. Decisions that must stay with me
5. Questions to ask before using this
6. Safer revised version
# RULES
- Do not treat the draft as correct.
- Do not invent missing facts.
- Keep final accountability with the human.
Use this before sending.
Green light
- Low-risk wording help.
- Internal draft only.
- Easy to verify.
Slow down
- Numbers or claims.
- Customer-facing content.
- Manager or legal sensitivity.
Human only
- Final decisions.
- Private or confidential judgement.
- High-impact recommendations.