agent-skills · PromptZap Team

7 Cross-Model A/B-Tested Prompt Techniques

Compilation of six months of A/B testing on Claude, GPT-4, and Gemini covering persona stacking, anti-examples, role-reversal QA, and emotional priming techniques.

Framework

standard

Model targets

claude, chatgpt, gemini

Prompt text

Apply these 7 A/B-tested prompt engineering techniques that consistently improve output quality across Claude, GPT-4, and Gemini:

# Technique 1: Persona + Constraint Stacking
Combine expert persona with explicit constraints:
"You are a [EXPERT ROLE] with [SPECIFIC CONSTRAINT]. [TASK]."
Example: "You are a senior security engineer who never suggests solutions without explaining the threat model first."

# Technique 2: Anti-Examples
Show what NOT to do:
"Generate [OUTPUT]. DO NOT: [bad example 1], [bad example 2]."

# Technique 3: Role-Reversal QA
Make the AI question YOU:
"Before answering, ask me 3 clarifying questions about [TOPIC] to ensure your response is accurate."

# Technique 4: Cognitive Scaffolds
Break complex tasks into explicit thinking steps:
"First analyze [X], then consider [Y], finally synthesize [Z]. Show your work for each step."

# Technique 5: Emotional Priming
Set emotional context for tone:
"Write this with [EMOTION: excitement/caution/urgency]. The reader should feel [DESIRED EMOTION]."

# Technique 6: Uncertainty Chain-of-Thought
Force acknowledgment of unknowns:
"If you're uncertain about any part, explicitly state 'UNCERTAIN:' and explain why before proceeding."

# Technique 7: Steelman-First
Require strongest version of opposing view:
"Before critiquing [POSITION], first present the strongest possible argument FOR it."

# Your Task
[DESCRIBE YOUR TASK]

Apply 2-3 of these techniques that best fit the task. State which techniques you're using and why.