Part 4 of 4

How to Iterate on Outputs

#prompt-engineering #ai #tutorial

1. Start with Evaluation Before Changes

  • Review the output against your original objective
  • Check alignment with task instructions and constraints
  • Identify specific gaps (missing detail, incorrect tone, unclear structure)

2. Isolate What Needs Improvement

  • Focus on one dimension at a time (clarity, depth, structure, tone)
  • Avoid changing multiple variables in a single iteration
  • Clearly define what “better” looks like for the next version

3. Refine the Prompt, Not Just the Output

  • Update Task Instructions to make expectations more explicit
  • Adjust Constraints (length, tone, scope) to guide improvements
  • Add or trim Context if accuracy or relevance needs improvement

4. Use Targeted Follow-up Instructions

  • Request specific improvements instead of full rewrites
  • Example directions:
    • Expand a section with more detail
    • Simplify language for a different proficiency level
    • Restructure content into clearer sections

5. Compare Versions Systematically

  • Evaluate the new output against the previous version
  • Check if the intended improvement is achieved
  • Retain changes that improve clarity and usefulness

6. Reinforce What Works

  • Keep effective instructions and constraints unchanged
  • Build on sections that already meet expectations
  • Use successful patterns for future prompts

7. Iterate Until Criteria Are Met

  • Continue refining until the output satisfies:
    • Objective alignment
    • Instruction completeness
    • Constraint adherence
  • Treat iteration as a controlled, step-by-step process