After working with thousands of prompts, certain patterns emerge. The best prompts, the ones that consistently produce exceptional AI outputs, share a common anatomy. They’re not necessarily long or complex, but they include specific elements in a specific order. This article breaks down that anatomy into a simple framework you can use every time you sit down to write a prompt.
The CREATE Framework
To make this memorable, I’ve organized the elements of a perfect prompt into the CREATE framework: Context, Role, Exact task, Audience, Tone, and Extras. You don’t need all six elements in every prompt, but the more you include, the better your results will be.
C – Context
Context is the background information the AI needs to understand your situation. AI doesn’t know who you are, what you’re trying to accomplish, or what constraints you’re working under. The more relevant context you provide, the better it can tailor its response.
Good context includes: who you are (your role, experience level), what you’re trying to accomplish (your goal, the bigger picture), what you’ve already tried or considered, and any constraints (time, budget, resources, limitations). You don’t need to write a novel; a sentence or two of context often transforms the quality of responses.
Weak prompt: “Write me a cover letter.” Strong prompt with context: “I’m a recent marketing graduate applying for an entry-level social media coordinator role at a sustainable fashion brand. I have internship experience at a local boutique and run my own lifestyle Instagram account with 5,000 followers. Write me a cover letter.”
R – Role
Telling the AI what role to adopt shapes how it approaches your request. Without a role, the AI defaults to a generic helpful assistant voice. With a role, it draws on the patterns and perspectives associated with that role.
Roles can be specific professions (“act as a senior copywriter”), specific types of people (“act as a busy working parent”), or even specific attitudes (“act as a friendly but honest critic”). The key is to choose a role that aligns with the expertise or perspective you need.
Example: “Act as a seasoned real estate agent who has helped hundreds of first-time home buyers.” This prompts the AI to draw on the patterns and insights it has learned from real estate content, giving you more informed and practical advice.
E – Exact Task
The exact task is the specific thing you want the AI to do. This sounds obvious, but it’s where most prompts fail. “Help me with my email” is not an exact task. “Write a 150-word email to my manager requesting two weeks of vacation in August, addressing potential concerns about project coverage” is an exact task.
Good exact tasks include: a clear verb (write, summarize, analyze, brainstorm, critique), a specific deliverable (an email, a list, a table, a paragraph), and clear parameters (length, scope, focus). The more precisely you define the task, the less room there is for the AI to wander.
Break complex tasks into smaller, specific tasks rather than asking for everything at once. “Help me plan a marketing campaign” will give you generic mush. “Identify three potential target audiences for a new productivity app, with two pain points each” gives you something specific and useful.
A – Audience
Specifying your audience helps the AI calibrate its language, depth, and examples. The same topic might be explained very differently to a 10-year-old, a college student, a working professional, or a subject matter expert. Telling the AI who will read or use its output dramatically improves relevance.
Audience specifications can include age, expertise level, role, industry, or even specific attitudes. “Explain this for a non-technical executive who is skeptical of new technology” is more useful than “explain this simply.”
Don’t just specify the audience’s knowledge level; think about what they care about. “Explain this for a CFO who is primarily concerned with cost savings and risk mitigation” gives the AI much more direction than “explain this for a business audience.”
T – Tone
Tone is the emotional and stylistic quality of the response. Without specifying tone, you get the AI’s default, which is usually competent but bland. Specifying tone makes the output more appropriate for its intended use.
Common tones include: professional, casual, friendly, formal, urgent, conversational, technical, humorous, serious, encouraging, direct. You can also combine tones: “professional but warm” or “casual but informative.”
For maximum control, give the AI a reference: “Match the tone of this sample: [paste example].” This is especially useful when you need the output to match your existing brand voice or personal style.
E – Extras
Extras are anything else that improves the response. This includes: format specifications (“provide as a numbered list,” “use a table,” “include headers”), constraints (“keep under 200 words,” “don’t use jargon”), examples (“here’s a sample of the style I want”), and specific elements to include or avoid.
Extras are where you fine-tune the prompt to get exactly what you want. They’re often the difference between a response that’s “pretty good” and one that’s “exactly what I needed.”
Putting It All Together
Here’s what a complete CREATE prompt looks like:
“Act as a senior content strategist (Role). I’m launching a weekly newsletter about personal finance for millennials (Context). Write three potential welcome email subject lines and preview text for each (Exact Task). My audience is 25-35 year old professionals who are financially responsible but feel behind on long-term goals like buying a house (Audience). The tone should be encouraging and relatable, not preachy or technical (Tone). Keep each subject line under 50 characters, and avoid words that trigger spam filters (Extras).”
Compare that to “write subject lines for my finance newsletter.” The CREATE version will give you dramatically better, more usable results.
When to Use the Full Framework
You don’t need to use the full CREATE framework for every prompt. For simple questions (“what’s the capital of Brazil?”), a quick prompt is fine. The full framework pays off when:
- The output quality really matters (work emails, important documents, creative content).
- You’ve tried a simple prompt and got disappointing results.
- The task is complex or has multiple requirements.
- You need a specific format or tone.
- You’re creating content for a specific audience.
Iterating Within the Framework
The CREATE framework isn’t just for your first prompt; it’s also a diagnostic tool. If the AI gives you a disappointing response, ask yourself which element was missing or weak. Vague response? Maybe your Context was thin. Wrong tone? You may not have specified it clearly. Off-target content? Your Exact Task may have been unclear.
This diagnostic approach turns every prompt into a learning opportunity. Over time, you’ll develop an intuition for which elements matter most for which types of tasks, and your prompts will naturally become more effective.
Practice With the Framework
Try this exercise: take a prompt you’ve used recently and rewrite it using the CREATE framework. Even if your original prompt was decent, you’ll likely find ways to improve it. Then run both versions through your AI tool and compare the results. Seeing the difference firsthand is the fastest way to internalize the value of structured prompting.
The CREATE framework isn’t the only way to structure prompts, but it’s a reliable starting point that covers the elements most likely to affect output quality. Use it as a checklist, a learning tool, and a habit that helps you consistently get better results from any AI tool you encounter.