I've watched thousands of students write their first prompt. Almost every one of them does the same thing: they type a question, hit enter, and then stare at the output wondering why it's generic.
The problem isn't the AI. The problem is the input.
Why most prompts fail
A prompt is a briefing. When you brief a freelancer, you tell them who you are, what you need, how you want it formatted, and what tone fits your brand. You don't just say "write something about marketing."
AI is no different. It will give you generic output if you give it nothing to work with. And most people give it nothing.
In 2022 I needed a way to teach prompting that my students could memorise and apply in 60 seconds. I tried teaching frameworks with 7 steps, 10 components, nested templates. None of it stuck.
CRAFT is what I landed on after cutting everything that didn't survive the classroom.
The 5 letters
C · Context
Tell the AI who you are and what situation you're in. "I run a digital marketing institute in Chennai. I'm writing a LinkedIn post for a professional audience of 30-45 year old marketers." That context changes every output that follows.
R · Role
Tell the AI what role to play. "You are a senior copywriter who has written for B2B SaaS brands in India." The role primes the model's output register. A practitioner sounds different from a generalist. Choose the practitioner.
A · Action
This is where most prompts are vague. "Write something" is not an action. "Write 3 LinkedIn hook options, each under 9 words, about the CRAFT prompt framework, leading with a specific number or contrarian claim" is an action. The more precise the action, the better the output.
F · Format
Tell the AI how to deliver the output. JSON. Numbered list. A table. Three paragraphs. Without this field, you spend time reformatting. Format it correctly the first time.
T · Tone
Tone is the most underused field. "Direct, slightly contrarian, no jargon" gives you a completely different output from "warm, encouraging, beginner-friendly." Name the tone. Don't make the AI guess.
What a full CRAFT prompt looks like
Here's an example I use for LinkedIn posts about AI tools:
C: I run Digital Scholar, an AI and digital marketing institute in Chennai. My LinkedIn audience is mid-level marketers and business owners, 28 to 45 years old, India-based.
R: You are a practitioner sharing field notes, not a guru. You've used these tools in real campaigns, not just demos.
A: Write 3 LinkedIn hook options for a post about the CRAFT prompt framework. Each hook under 9 words. One leads with a specific number, one leads with a contrarian claim, one leads with a relatable failure.
F: Return as a numbered list. No explanations. Just the 3 hooks.
T: Direct. Slightly contrarian. No filler. No corporate vocabulary.
That prompt takes 45 seconds to write. The output is usable. Every time.
Where CRAFT works
I've run this framework across ChatGPT, Claude, and Gemini. It works on all three. It also works for content that has nothing to do with marketing: legal document summaries, Excel formula generation, code debugging, performance review writing.
The structure is model-agnostic because it's really about clarity of instruction, not any specific AI quirk.
The one mistake I still see
People master C and A and skip R, F, and T. They get output that's directionally right but tonally wrong, formatted badly, and sounds like it was written by nobody in particular.
Fill all 5 fields. Every time. For 100 prompts. Then tell me whether it made a difference.
If you want to run it live, the CRAFT Prompt Builder is free and takes 60 seconds.
