There's a peculiar irony in the AI revolution: the most powerful tools in human history are being used at about 10% of their capacity by about 90% of the people who have access to them.
The average person opens ChatGPT or Claude, types a vague question — "write me a marketing email" or "help me with my resume" — reads the first response, and either uses it as-is or decides the tool isn't that useful. They're treating a symphony orchestra like a doorbell.
The people who get remarkable results from AI aren't using a different tool. They're using the same tool differently. They've learned the skill of communicating with language models in a way that produces specific, high-quality, genuinely useful outputs.
This skill has a name: prompt engineering. And despite what the internet might have you believe, it doesn't require a technical background. It requires structured thinking and a framework.
Here's ours.
The CRAFT Framework
CRAFT stands for Context, Role, Action, Format, and Tone. These five elements, when combined thoughtfully, transform a generic AI interaction into a precise collaboration. Let me break each one down.
C — Context
Context is the background information the AI needs to understand your situation. Without context, the model defaults to the most generic interpretation of your request.
Without context: "Write me a marketing email."
With context: "I run a 12-person digital marketing agency that specializes in helping e-commerce brands scale from $1M to $10M in revenue. We just launched a new service: AI-powered ad creative testing. I need to announce this to our existing client list of about 200 brands."
The second prompt gives the AI enough information to write something specific and relevant. The first prompt gives you something you could find on any template website.
The rule: give the AI everything it would need to know if it were a smart colleague who just joined your team today.
R — Role
Role tells the AI who it should be when generating its response. This is one of the most powerful levers in prompt engineering because it fundamentally shifts the model's perspective and vocabulary.
Without role: "Help me write a sales page."
With role: "You are a direct-response copywriter with 15 years of experience writing high-converting sales pages for digital education products. You specialize in long-form copy that builds emotional resonance before presenting the offer."
When you assign a role, the model draws on a different subset of its training data. It adjusts its language, its structure, its level of detail, and its assumptions about what matters. A "sales page" written by a "direct-response copywriter" will be fundamentally different from one written by a "content marketer" or a "brand strategist."
A — Action
Action is the specific task you want the AI to perform. This seems obvious, but the specificity of your action request determines the specificity of the output.
Vague action: "Write a blog post about AI."
Specific action: "Write a 1,500-word blog post that explains what AI agents are, why they matter for small business owners, and walks through three practical use cases with step-by-step implementation details. Include a clear call-to-action at the end inviting readers to join our free workshop."
The specific action includes word count, topic boundaries, audience, structural requirements, and desired outcome. The vague action leaves all of those decisions to the model, which means you're gambling on its defaults.
F — Format
Format defines the structure of the output. This is where most people leave the most value on the table, because they accept whatever structure the AI defaults to rather than specifying the structure that serves their purpose.
Format specifications can include: bullet points vs. paragraphs, the number of sections, whether to include headers, the use of tables or comparisons, whether to include examples, the length of each section, and the overall document structure.
"Structure the blog post with an opening hook (2-3 sentences, no preamble), five sections with H2 headers, each section 200-300 words, practical examples in each section, and a final section that transitions to the CTA."
When you specify format, you don't just get a better-organized output. You train yourself to think about information architecture — how ideas should be structured for maximum impact. That's a skill that transfers far beyond AI.
T — Tone
Tone governs the voice, style, and emotional register of the output. It's the difference between an output that sounds like it was written by a machine and one that sounds like it was written by a specific person with a specific perspective.
Generic tone: "Write in a professional tone."
Specific tone: "Write in the voice of someone who is deeply knowledgeable but not academic — authoritative without being condescending. Use short, declarative sentences mixed with longer explanatory ones. Avoid jargon unless you explain it immediately. The reader should feel like they're learning from a mentor who respects their intelligence."
You can also reference specific writers or publications as tone anchors: "Write in the style of James Clear's Atomic Habits — clear, example-driven, and direct."
Putting CRAFT Together
Here's a complete prompt using all five elements:
"Context: I'm launching a 6-week cohort-based course on AI automation for small business owners. The course costs $997 and includes live workshops, a community, and lifetime access to course materials. We've had 47 students in our beta cohort with a 94% satisfaction rate.
Role: You are a conversion copywriter who specializes in education products. You understand the psychology of high-ticket digital purchases and know how to build trust through specificity and social proof.
Action: Write an email sequence of three emails that nurtures cold leads toward enrolling in the course. Email 1 should identify the pain point (wasting time on manual tasks). Email 2 should establish credibility (beta results, student stories). Email 3 should present the offer with urgency (cohort starts in 10 days, limited to 50 seats).
Format: Each email should be 300-400 words. Include a subject line for each. Use short paragraphs (2-3 sentences max). Include one specific student result or testimonial per email.
Tone: Confident but not hype-driven. Speak to intelligent business owners who are skeptical of online courses. Avoid exclamation points. Let the results speak louder than the adjectives."
Compare that to "write me a sales email sequence for my course." Same tool. Radically different output.
Beyond the Framework: Three Principles
1. Iterate, don't regenerate.
When the output isn't right, don't start over. Tell the AI what to fix. "The second email is too long. Tighten it to 250 words and make the CTA more specific." Iteration is faster and produces better results because the model builds on its previous work.
2. Feed it examples.
If you have a piece of writing you love — your own or someone else's — give it to the AI as a reference. "Here's a blog post I wrote that captures my voice perfectly. Analyze its style and apply the same voice to the new piece." This is more effective than any tone description.
3. Chain your prompts.
Complex outputs should be built in stages, not all at once. First, ask the AI to create an outline. Review it. Then ask it to write section one. Review it. Then section two. This approach gives you more control and catches problems early.
The Compound Effect
Prompt engineering is not a party trick. It's a productivity multiplier that affects every area of your work.
The person who masters this skill writes better emails in half the time. Creates better content with less effort. Analyzes data faster. Generates better ideas. Makes better decisions by stress-testing their thinking against a model that's read most of the internet.
Over the course of a year, the gap between someone who uses AI casually and someone who uses it strategically becomes enormous. Not because the tool is different. Because the operator is different.
And becoming a better operator takes thirty minutes of learning a framework and a few weeks of deliberate practice.
Start with CRAFT. Apply it to your very next AI interaction. Notice the difference. Then never go back to typing vague questions and hoping for the best.
