AI Iteration Workflow Card

Use this after the first response. The goal is to turn a rough answer into useful output without restarting the conversation.

1. Critique

Use when the answer is close but specific parts are wrong.

Prompt:

Keep [specific part that works]. Change [specific part that does not work]. The problem is [concrete issue]. Produce a revised version only.

2. Narrow

Use when one part matters more than the rest.

Prompt:

Focus only on [section / option / risk]. Expand it into [format]. Include [details required]. Ignore the rest for now.

3. Pivot

Use when the material is useful but the audience, format, or angle is wrong.

Prompt:

Use the same substance, but rewrite it for [audience] as [format]. Optimize for [decision / action / clarity]. Keep [constraints].

4. Stress-Test

Use when the answer sounds plausible but might be weak.

Prompt:

Stress-test your answer. What is most uncertain, what would a strong critic challenge, and which claims should I verify before using this?

5. Verify

Use before decisions, publishing, customer messages, policies, or code changes.

Checklist:

• Which factual claims need a source?
• What assumptions did the model add?
• What is the consequence if this is wrong?
• What edge case is missing?
• Can I act on this as written?

Stop Or Restart

Stay in the same thread when the answer is mostly on target and you can name what needs to change.

Restart when the conversation has drifted, become too long, or you want a fresh test of the same prompt.