The Iteration Guide — Frameworks for Refining AI Outputs
A practical guide to iterative prompting — techniques, frameworks, and templates for bouncing AI outputs to perfection.
The Iteration Guide 🎯
Every great AI output started as an okay one.
The RICE Framework for Iterative Prompting
The most effective iteration method in 2026 follows four steps:
R — React
Read the output. What is good? What is wrong? What is missing? Be specific in your assessment before you respond.
I — Instruct
Tell the AI exactly what to change. "Make it better" is useless. "Make the introduction more concise, add specific statistics, and change the tone from academic to conversational" is actionable.
C — Constrain
Add boundaries that were missing from your original prompt. Length limits, format requirements, audience specifications, exclusions.
E — Expand
Ask for more depth in areas that worked well. If paragraph three was excellent, ask for that level of detail everywhere.
Five Iteration Patterns That Work
1. The Funnel
Start broad, get narrow. Each iteration zooms in on specifics.
| Iteration | Prompt Pattern | Example |
|---|---|---|
| 1 | "Give me an overview of X" | "Give me an overview of email marketing trends in 2026" |
| 2 | "Focus on [best section]" | "Expand on the AI personalisation section" |
| 3 | "Add specifics to [area]" | "Add statistics and tool recommendations" |
| 4 | "Format for [use case]" | "Reformat as a client-ready presentation outline" |
2. The Pivot
When the output is wrong direction, not wrong quality. Do not iterate on bad foundations — redirect.
"That's focused on B2B. I need this for B2C consumer brands. Same depth, different angle."
3. The Critic
Ask the AI to critique its own work before you refine.
"Review what you just wrote. What are the three weakest points? What claims need better evidence? What would a skeptic challenge?"
Then: "Now rewrite it addressing those weaknesses."
4. The Exemplar
Show the AI what good looks like.
"Here is an example of the tone and depth I want: [paste example]. Now rewrite your previous response to match this style."
5. The Additive
Do not rewrite — layer improvements on top.
"Keep everything but add: (1) a comparison table, (2) a specific case study, (3) a contrarian perspective."
When to Stop Iterating
The law of diminishing returns applies. Most outputs peak at 3-5 iterations. Signs you should stop:
- The changes between iterations are cosmetic, not substantive
- You are rearranging rather than improving
- The AI starts looping back to previous versions
- You have spent more time iterating than the output is worth
The goal is not perfection. The goal is good enough for your purpose, and that almost always arrives by iteration 4.
Iteration by Use Case
| Use Case | Typical Iterations | Key Refinement Focus |
|---|---|---|
| Blog post / article | 3-4 | Tone, structure, evidence |
| Code generation | 4-6 | Edge cases, error handling, naming |
| Email drafting | 2-3 | Tone, call-to-action, length |
| Data analysis | 3-5 | Accuracy, methodology, visualisation |
| Creative writing | 5-8 | Voice, pacing, imagery |
| Business strategy | 3-4 | Assumptions, risks, specificity |
| Technical documentation | 4-5 | Accuracy, completeness, clarity |
Common Iteration Mistakes
| Mistake | Why It Fails | Better Approach |
|---|---|---|
| "Make it better" | AI cannot read your mind | Specify exactly what to improve |
| Rewriting the full prompt each time | Loses conversation context | Build on previous exchanges |
| Iterating on fundamentally wrong outputs | Polishing a wrong answer | Pivot instead of refine |
| Never stopping | Perfectionism wastes time | Set an iteration budget (3-5 max) |
| Only iterating content | Missing format/structure issues | Iterate on format first, content second |