Practical guide
Follow the process step by step
Apply each stage in sequence. Document important assumptions, verify AI-generated information, and keep human approval at critical decision points.
- 1
Define the task and outcome
State what the model must do, what success looks like, and how the result will be used.
- 2
Supply relevant context
Provide background, audience information, source material, assumptions, terminology, and operating conditions.
- 3
Set constraints and boundaries
Specify scope, tone, length, prohibited behavior, evidence rules, formatting, and decision limits.
- 4
Provide examples when useful
Use representative examples to clarify expected structure, reasoning standards, terminology, and output quality.
- 5
Define evaluation criteria
Explain how accuracy, relevance, completeness, consistency, safety, and usability should be assessed.
- 6
Test and version the prompt
Run varied inputs, record failure patterns, improve instructions, and maintain controlled versions of reliable prompts.