Prompt EngineeringUpdated 2026-07-20
The Universal Prompt Formula
The one formula that works for almost every task: Role + Task + Context + Format + Constraints.
The Formula
- ROLE — Who is the AI? (e.g. "You are a senior editor")
- TASK — What should it do? Start with a strong verb (write, analyze, generate).
- CONTEXT — What does it need to know? Audience, background, purpose.
- FORMAT — How should the output look? Bullets, table, JSON, length.
- CONSTRAINTS — What are the limits? Tone, exclusions, boundaries.
One-Line Template
- "You are a {role}. {task} about {topic} for {audience}. Output as {format}. Use {tone} and avoid {exclusions}. Target {length}."
Quality Boosters
- Add 1-3 examples of ideal output (few-shot).
- Add "Think step by step" for reasoning tasks.
- Add "Only answer from the provided context" to prevent hallucination.
- Add "Ask clarifying questions first" for ambiguous requests.
Quick Wins
- Never accept the first draft — iterate once minimum.
- Feed failures back into the prompt with a fix instruction.
- Test one prompt across two models; pick the winner.