AI Prompt Engineering Best Practices

The CRISP Framework

• Context: Provide relevant background information about the task, industry, or situation
• Role: Define who or what the AI should act as (expert, assistant, analyst)
• Instructions: Clear, specific directions on what to do
• Scope: Boundaries and constraints (length, format, tone)
• Purpose: The ultimate goal or desired outcome

Tips

• Start prompts with 'You are a...' to establish role
• Use numbered lists for multi-step instructions
• Include examples of desired output format

Prompt Patterns That Work

• Chain of Thought: 'Think step by step and explain your reasoning...'
• Few-Shot Learning: Provide 2-3 examples before your actual request
• Persona Pattern: 'Act as a senior [role] with 20 years experience...'
• Template Pattern: 'Fill in the following template: [structure]...'
• Reflection Pattern: 'Review your response and identify any weaknesses...'

Examples

• Sales Email: You are a B2B sales expert. Write a cold email for [product] targeting [persona]. Keep it under 150 words. Focus on one pain point. End with a soft CTA.
• Data Analysis: Act as a data analyst. Review this dataset and: 1) Identify top 3 trends 2) Flag any anomalies 3) Suggest 2 actionable insights. Format as bullet points.

Common Mistakes to Avoid

• ❌ Being too vague: 'Write something about marketing'
• ❌ No context: Jumping straight to the ask without background
• ❌ Overloading: Asking for 10 things in one prompt
• ❌ Ignoring format: Not specifying desired output structure
• ❌ No iteration: Accepting first output without refinement

Quick Reference Cheat Sheet

• • Always specify the output format (JSON, markdown, bullets, etc.)
• • Use delimiters like ### or """ to separate sections
• • Ask the AI to 'ask clarifying questions' if unsure
• • Include 'Do not include...' for explicit exclusions
• • Test prompts with edge cases before deploying