Prompt Engineering: The CRAFT Framework for Better AI Results

Why most people get bad results from AI

The quality of your AI output is almost entirely determined by the quality of your prompt. Most people write prompts like a Google search query — short and vague. AI performs much better when given context, role, format and constraints.

The CRAFT framework for prompts:

C — Context: What is the situation? What does AI need to know?

"I run a 12-person UK digital marketing agency serving B2B SaaS companies..."

R — Role: Who should the AI be?

"Act as a senior copywriter with 15 years of B2B SaaS experience..."

A — Action: What specifically do you want?

"Write a LinkedIn post announcing our new case study..."

F — Format: How should the output be structured?

"Output as 3 options. Each option should be 150-200 words with a hook, body and CTA..."

T — Tone: What style/voice?

"Professional but conversational. Confident without being salesy..."

Examples — Bad vs Good prompts:

❌ Bad: "Write me a cold email" ✅ Good: "You are an expert B2B sales copywriter. Write a cold outreach email for a UK software development agency targeting HR Directors at companies with 50-200 employees. The email should address the pain point of HR teams spending too much time on manual processes. Tone: professional and direct. Length: 150 words max. Include a clear CTA to book a 20-minute call."

Prompt templates for business:

• Research: "Summarise the key challenges facing [industry] in 2025. Structure as: 3 main challenges, evidence/examples, implications for business strategy."
• Ideation: "Give me 10 ideas for [goal]. For each idea, rate feasibility (1-5) and potential impact (1-5)."
• Editing: "Improve this text for clarity and professionalism. Keep the core message. Output the improved version only: [paste text]"