You get:
- the same prompt producing different outputs on different days
- new team members unable to understand or modify your prompts
- legacy prompts that no one wants to touch because they’re fragile
- vague instructions that the model interprets differently each time
- prompts that work for you but fail for others (different assumptions)
But clear prompts follow structure:
- role: who is the model acting as?
- task: what exactly should it do?
- context: what information does it need?
- constraints: what can’t it do?
- format: what should the output look like?
- examples: show, don’t just tell
Without rewriting, vague prompts stay vague.
This prompt transforms messy prompts into production-ready clarity.
Assume the role of a prompt engineer who specializes in clarity.
Your task is to rewrite a vague prompt into a clear, structured, production-ready prompt.
Generate:
1. ORIGINAL PROMPT ASSESSMENT
- What works (preserve this)
- What's unclear (specific issues)
2. REWRITTEN PROMPT
- Use structure: ROLE + TASK + CONTEXT + CONSTRAINTS + FORMAT + EXAMPLES
- Add XML or markdown tags for structure
- Include placeholders in {{brackets}} for variable inputs
3. ADDED ELEMENTS (explain what you added and why)
- Role definition
- Output format specification
- Examples (few-shot)
- Negative constraints (what NOT to do)
- Edge case handling
4. BEFORE/AFTER COMPARISON
- Show side-by-side
5. TEST SUGGESTIONS
- 3 test inputs to verify the rewritten prompt works
INPUTS:
Original prompt:
[PASTE THE VAGUE/CONFUSING PROMPT]
Intended task (if not clear from original):
[E.G., "Extract action items from meeting notes"]
Model to optimize for:
[GPT-4 / CLAUDE / GEMINI / MULTI-MODEL]
Tone/style preference:
[PROFESSIONAL / CONCISE / CONVERSATIONAL / TECHNICAL]
RULES:
- Preserve the original intent (don't change what it does)
- Add structure without adding unnecessary length
- Every added element must have a justification
- Use anti-examples (what NOT to do) when helpful
- If the original prompt has good examples, keep them; if not, generate plausible ones
- Run this on every prompt that produces inconsistent outputs — clarity is usually the fix.
- Use the rewritten prompt as a template for standardizing prompts across your team.
- Test the rewritten prompt on the “test suggestions” before deploying.
- Keep the original prompt assessment — it helps you learn what not to write next time.
- For legacy prompts, run this before any major modification — understand it first.
Original prompt:
“Look at this customer feedback and tell me if they’re mad and what we should do about it. Keep it brief.”
Intended task:
“Classify customer feedback sentiment and suggest an appropriate response”
Model to optimize for:
“GPT-4”
Tone/style preference:
“Professional — for customer support team use”
This framework improves outcomes by forcing:
- role assignment (who is the model?)
- task specification (what exactly to do?)
- constraint listing (what NOT to do?)
- format definition (what should output look like?)
- example inclusion (show, don’t just tell)
Great prompt rewriting doesn’t change what the prompt does — it makes what it does reliable and repeatable.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.