D - Define Multiple Perspectives Instead of: "Write a marketing email" Use: "You are three experts: a behavioral psychologist, a direct response copywriter, and a data analyst. Collaborate to write..."
E - Establish Success Metrics Instead of: "Make it good" Use: "Optimize for 40% open rate, 12% CTR, include 3 psychological triggers"
P - Provide Context Layers Instead of: "For my business" Use: "Context: B2B SaaS, $200/mo product, targeting overworked founders, previous emails got 20% opens"
T - Task Breakdown Instead of: "Create campaign" Use: "Step 1: Identify pain points. Step 2: Create hook. Step 3: Build value. Step 4: Soft CTA"
H - Human Feedback Loop Instead of: Accept first output Use: "Rate your response 1-10 on clarity, persuasion, actionability, and factual accuracy. For anything below 8, improve it. If you made any factual claims you're not completely certain about, flag them as UNCERTAIN and explain why. Then provide enhanced version."
Real example from yesterday:
You are three experts working together:
1. A neuroscientist who understands attention
2. A viral content creator with 10M followers
3. A conversion optimizer from a Fortune 500
Context: Creating LinkedIn posts for AI consultants
Audience: CEOs scared of being left behind by AI
Previous posts: 2% engagement (need 10%+)
Task: Create post about ChatGPT replacing jobs
Step 1: Hook that stops scrolling
Step 2: Story they relate to
Step 3: Actionable insight
Step 4: Engaging question
Format: 200 words max, grade 6 reading level
After writing: Score yourself and improveResult: 14% engagement, 47 comments, 3 clients
What I learned after 1,000 prompts:
- Single-role prompts get generic outputs
- No metrics = no optimization
- Context dramatically improves relevance
- Breaking tasks prevents AI confusion
- Self-critique produces 10x better results
Quick test for you:
Take your worst ChatGPT output from this week. Run it through DEPTH. Post the before/after below.
Questions for the community:
- What frameworks are you using in 2025?
- Anyone found success with different structures?
- What's your biggest ChatGPT frustration right now?