The RACE Framework: The Complete Guide to Writing AI Prompts That Actually Work

I get asked this question at least once a week: "How do you get such good outputs from AI?" The honest answer is that I'm not smarter than anyone else. I just have a system. I call it the RACE Framework, and it's the difference between prompts that suck and prompts that actually deliver expert-level work.

The RACE Framework breaks down into four components: Role, Action, Context, and Expectation. If you nail all four, your AI outputs will be dramatically better. Miss any one of them, and you'll get mediocre results.

Let me walk you through each component, and then I'll show you how to apply it to real prompts.

R: Role

Role is who you are and who you want the AI to be. This is where most people fail. They don't tell the AI what perspective to take or what expertise it should apply, and then they wonder why the output feels generic.

Role answers: "What role should I play, and what role should the AI play?"

Here are some examples of roles you might define:

• Your professional role (founder, manager, designer, developer, etc.)
• Your industry expertise (healthcare, finance, tech, etc.)
• Your experience level (beginner, intermediate, expert)
• The role you want the AI to play (strategist, copywriter, technical expert, mentor)
• The audience perspective (executive, engineer, consumer)
• Your expertise and background
• Any relevant credentials or experience you bring

A bad role definition looks like this:

A good role definition looks like this:

That second version tells the AI exactly who you are and how it should approach the task.

A: Action

Action is the specific task you want completed. Most prompts fail here because people are vague about what they actually need. They skip explaining what the task really is and jump straight to "do the thing" without being clear.

Action answers: "What specifically do I need you to create or do?"

When you include a clear action, you're saying: "Here's the specific task. Here's what I need created. Here's the format and scope." The AI can then focus on delivering exactly what you're asking for.

Bad action (too vague):

Good action (specific and clear):

Now the AI understands exactly what you need—not just "a landing page" but a specific section with a specific word count focused on a specific problem. Everything it creates will be laser-focused on your actual need.

C: Context

Context is the background information and circumstances surrounding your request. This is where most prompts fail. People don't give the AI enough context about their situation, constraints, audience, or goals—and then wonder why the output feels irrelevant.

Context answers: "What background, constraints, and circumstances should the AI understand?"

Context includes things like:

• Your business situation (industry, company size, stage)
• Your audience (who will use this? what do they care about?)
• Constraints (budget, timeline, technical limitations)
• Success criteria (what does "good" look like?)
• Past examples (reference material, tone samples, style guides)
• Current challenges (what problem are you solving?)
• Any relevant data, market conditions, or competitive landscape

Bad context (too generic):

Good context (specific and situational):

That level of context completely changes the output. The AI understands your specific situation, not a generic request.

E: Expectation

Expectation is what you expect the AI to deliver and how you'll evaluate it. It's usually the shortest part of a great prompt. Why? Because you've already provided all the setup in Role, Action, and Context.

Expectation answers: "What should the output look like? How will I know it's good?"

This is where you set clear standards for success. But because you've set up everything else, you don't need to over-explain it here.

Bad expectation (with no setup):

Good expectation (with RACE framework setup):

That expectation is powerful because everything before it has already set the context. The AI knows who the audience is, what your voice sounds like, what problems you solve, and exactly what defines success.

Putting It All Together: Complete Example

Let me show you a complete RACE prompt in action:

That prompt is specific, contextual, and actually useful. The AI now knows exactly what you want and why you want it.

Why RACE Works

The RACE Framework works because it mirrors how expert humans work. When I hire a designer, I don't just say "make it look good." I give them background on my brand, what problem I'm solving, what specific deliverables I need, and what success looks like. Then I ask them to execute.

AI is the same way. It's not magical. It just needs clear direction.

Here's what happens when you use RACE consistently:

1. Your outputs get 10x better because the AI has actual context.
2. You spend less time editing because you were clear about what you wanted.
3. You can reuse and modify prompts for similar tasks (Resources stay the same, only Execution changes).
4. You can train other people on your prompts using a consistent framework.

I use this framework for every single thing I ask AI to do—blog posts, code, emails, presentations, even strategic analysis. The RACE Framework isn't just a prompt template. It's a way of thinking about how to communicate with AI clearly.

The Quick Checklist

Next time you're writing a prompt, run through this checklist:

• Role: Have I defined who I am and what role I want the AI to play?
• Action: Have I been specific about what I need created or done?
• Context: Have I given the AI enough background about my situation, audience, and constraints?
• Expectation: Have I clarified what success looks like and how the output will be used?

If you can check all four boxes, you'll get dramatically better results.