C.R.I.S.P.E. Framework

The C.R.I.S.P.E. framework (Capacity/Role, Insight, Statement, Personality, Experiment) is a five-component prompt engineering methodology designed for iterative refinement. Its unique Experiment component encourages generating multiple variations, making it ideal for strategic decision-making and creative exploration.

Introduction

The C.R.I.S.P.E. Framework stands apart from other prompt engineering frameworks through its emphasis on experimentation and iteration. While most frameworks focus on producing a single optimal output, CRISPE explicitly encourages generating multiple approaches, comparing them, and refining toward the best solution.

This experimental mindset makes CRISPE particularly powerful for:

• Strategic planning where multiple viable approaches exist
• Creative work where exploring different directions leads to better outcomes
• Complex problem-solving where the best solution isn't obvious upfront
• A/B testing content where you need variations for comparison
• Research and analysis where different methodological approaches yield different insights

Origin & Background

The CRISPE framework originated as an internal methodology for structured prompt engineering and has since been adopted across technical, strategic, and creative contexts. Its design reflects a key insight: the best way to get excellent AI output isn't to write one perfect prompt, but to systematically explore the output space through controlled variation.

CRISPE's Experiment component is inspired by scientific methodology. Just as researchers run multiple trials with controlled variables, CRISPE users request multiple output variations with specified differences. This approach:

• Reveals the AI's strongest capabilities for a given task
• Identifies blind spots in your initial prompt design
• Produces genuinely different approaches rather than superficial rewording
• Gives you options to combine the best elements from multiple outputs

Most frameworks control what the AI says and how it's formatted. CRISPE's Personality component goes deeper by controlling how the AI thinks. Specifying "think like a venture capitalist" produces fundamentally different reasoning patterns than "think like an engineer," even for the same task.

The Capacity/Role component recognizes that professional identity has two dimensions: what you know (capacity/expertise) and what you do (role/function). A "senior data scientist" (capacity) who is "acting as a product advisor" (role) brings different value than the same person acting as a "team lead." This dual specification produces more nuanced outputs.

How C.R.I.S.P.E. Compares to Other Frameworks

• You want to explore multiple approaches before committing
• The task is strategic and there's no single "right" answer
• You need creative variations for A/B testing
• Personality and reasoning style significantly impact output quality
• You're willing to invest more prompt effort for better results

• For single-output tasks where one answer is sufficient (use RACE or APE)
• When communication style matters more than reasoning style (use CO-STAR)
• For step-by-step process work (use RASCEF)
• For quick, simple requests (use APE)

C.R.I.S.P.E. Framework Structure

1. Capacity/Role

This dual component sets both what the AI knows (capacity) and how it applies that knowledge (role). Be specific about expertise area, years of experience, and the functional role.

• "You are a senior data engineer (10+ years) acting as a technical advisor to a non-technical CEO evaluating data infrastructure vendors"
• "You are a behavioral economist with expertise in pricing psychology, consulting for a D2C e-commerce brand"
• "You are a former journalist turned content marketing director at a hypergrowth SaaS company"

• "You are an expert" (no domain, no role)
• "Be a marketer" (too broad)

2. Insight

Insight gives the AI the factual foundation it needs to produce relevant, grounded outputs. Include metrics, competitive landscape, constraints, and any data that should inform the response.

• Include specific numbers: "Our current CAC is $127, LTV is $890, and churn is 4.2% monthly"
• Include competitive context: "Three competitors launched similar features in Q4, pricing 20-30% below us"
• Include constraints: "We have a team of 4 with no dedicated designer, and need to launch in 6 weeks"

3. Statement

The Statement is the central ask: what you want the AI to produce. Keep it focused on a single clear deliverable. The Statement should be specific enough that you could evaluate whether the output successfully addressed it.

• "Develop a pricing strategy that increases ARPU by 15% without increasing churn beyond 5%"
• "Create a crisis communication plan for a data breach affecting 10,000 customer records"
• "Design a customer onboarding flow that achieves 80% activation within the first week"

4. Personality

Personality goes beyond tone; it shapes how the AI reasons and presents ideas. Different personalities produce fundamentally different thinking patterns and communication styles.

• "Be direct and contrarian like Peter Thiel. Challenge conventional wisdom, prioritize unique insights over safe recommendations"
• "Think like a YC partner giving advice to a founder: pragmatic, metrics-focused, with a bias toward action and speed"
• "Communicate like a patient teacher. Explain complex concepts using analogies, anticipate confusion, and build understanding incrementally"

5. Experiment

This is CRISPE's signature component. Instead of accepting the first output, systematically request variations that explore different approaches. This produces better outcomes because:

• You can cherry-pick the best elements from each variation
• You discover approaches you wouldn't have thought of
• You can validate which approach resonates most with stakeholders

• "Provide three variations: conservative, moderate, and aggressive approaches"
• "Give me two versions: one optimized for speed-to-market, one optimized for quality"
• "Present four headline options with different emotional angles: curiosity, urgency, aspiration, and social proof"

• Specify what should differ between variations (approach, tone, scope)
• Request the same core structure so variations are comparable
• Ask for pros/cons of each variation to aid decision-making
• Clear chat history between experiments for truly independent outputs

Example Prompts Using C.R.I.S.P.E.

Example 1: Product Strategy

Example 2: Content Strategy

Common Mistakes to Avoid

1. Treating Experiment as Optional

2. Shallow Personality Specifications

3. Insight Without Data

4. Overly Broad Statements

Best Use Cases for C.R.I.S.P.E.

Strategic Decision-Making

• Business strategy exploration
• Market entry analysis
• Pricing strategy development
• Partnership evaluation

Creative Exploration

• Campaign concept development
• Brand positioning options
• Content angle brainstorming
• Product naming and messaging

Problem-Solving

• Root cause analysis with multiple hypotheses
• Solution design with trade-off analysis
• Risk assessment from different perspectives
• Stakeholder communication strategies

Research and Analysis

• Competitive intelligence
• Market sizing with different assumptions
• Technology evaluation and comparison
• User research synthesis with multiple lenses

Conclusion

The C.R.I.S.P.E. Framework uniquely combines structured prompting with systematic experimentation. By building variation and comparison into the prompt itself, CRISPE produces not just better individual outputs, but better decision-making through exploration.

• Built-in experimentation produces multiple options to choose from
• Personality control shapes how the AI thinks, not just what it says
• Dual Capacity/Role specification activates both expertise and function
• Data-driven Insight component grounds outputs in reality
• Ideal for strategic tasks where the best approach isn't obvious

• Start with strategic tasks where you genuinely want to explore multiple approaches
• Experiment with different Personality specifications, as they produce surprisingly different results
• Build a library of effective Experiment templates for common decision types
• Use CRISPE for important decisions, APE or RACE for routine tasks