Iterative prompt improvement

Improve prompts based on measurable outcomes.

Useful techniques include:

• structured input
• explicit output schemas
• examples
• iterative refinement
• feedback loops
• reasoning-oriented instructions when appropriate

Instead of:

> Analyze this.

Use:

Customer:
Issue:
Priority:
Relevant policy:
Required output:

Define expected fields:

{
  "category": "...",
  "priority": "...",
  "reason": "..."
}

Prompt
 ↓
Model
 ↓
Evaluate
 ↓
Identify failure
 ↓
Modify prompt
 ↓
Retest

Prompt iteration should be evidence-driven.

Do not continuously change prompts without a controlled evaluation set.

---