Module 3 of 6
The Iteration Loop
How to read AI output as feedback. When to refine, when to redirect, when to restart.
The Mistake Most People Make
When the output isn't quite right, most people start completely over. They throw away the conversation and write a brand new prompt from scratch.
Why this is wrong:
You're throwing away useful context. The AI already understands the domain, the general direction, the constraints. Starting over means re-teaching all of that.
The Better Approach
Prompt → Output → Diagnose → Adjust → Better Output
Iteration is a conversation, not a slot machine
When output isn't right, pause and diagnose: Why didn't this work? Did the AI misunderstand? Is it close but needs tweaking? Is it completely off base?
The Three Moves
REDIRECT
Use when:
AI misunderstood your intent
What to say:
"That's not quite it. I meant [X]."
Example:
You asked for 'simple pricing table' and got a complex chart. Say: 'Not quite — I meant a 3-column comparison with just plan names, prices, and top features.'
REFINE
Use when:
Output is mostly right but needs tweaking
What to say:
"Good, but adjust [specific thing]."
Example:
You got a good draft but the tone is too formal. Say: 'This is close. Make it more conversational — like you're talking to a friend over coffee.'
RESTART
Use when:
Completely off track
What to say:
"Start fresh with better prompt"
Example:
You asked vaguely and got something useless. Don't iterate on garbage. Write a new prompt using the 5 levers.
Decision Heuristic: Which Move?
Use this quick decision tree when output isn't perfect:
| Output Quality | Move | Example |
|---|---|---|
| 70%+ right | REFINE | Good structure, wrong tone |
| 50-70% right | REDIRECT | Wrong focus, but salvageable |
| Directionally wrong | REDIRECT | Misunderstood the core ask |
| Completely off | RESTART | Not even close to useful |
Real Iteration Example
First Attempt
You: "Write an email to my team about the new project timeline."
AI: [Generic, formal email that doesn't match your voice]
Redirect (clarify intent)
You: "Not quite. We're behind schedule and I need to be honest about that without creating panic. Tone should be: 'Here's what happened, here's the new plan, we've got this.'"
AI: [Better email, right tone, but too long]
Refine (small adjustment)
You: "Good. Cut it to 3 paragraphs max. Keep the directness."
AI: [Perfect. Ready to send.]
Result
Two iterations, 90 seconds total. You got exactly what you needed by diagnosing quality and making the right move each time.
Try This
Your challenge: Practice diagnosing output
Next time you get imperfect output from AI, pause before starting over. Ask yourself:
1. What percentage is this right? (Be honest)
2. Why didn't it work perfectly? (Misunderstanding? Missing context? Wrong tone?)
3. Which move fits? (Redirect, Refine, or Restart)
Success looks like:
You get to a great result in 2-3 iterations instead of starting from scratch 5 times.
Key Takeaways
Don't start over when output is imperfect. Iteration preserves useful context.
Diagnose first: What percentage right? Why didn't it work? Which move fits?
REDIRECT when AI misunderstood. REFINE when it's close. RESTART when it's hopeless.
The decision table is your friend: 70%+ = Refine, 50-70% = Redirect, <50% = Restart.
Great results usually take 2-3 iterations. That's normal and much faster than starting over.