The 3-Part Prompt Formula That Makes AI Output Actually Usable
I want to tell you about the worst prompt ever written.
It was 2022. ChatGPT had just dropped for public use and I, a mentally exhausted consultant who had been asked to review a performance report for an employee I privately believed did not deserve a mediocre review let alone a good one, saw an opportunity.
I copied the entire document into ChatGPT — no data privacy awareness whatsoever, future me cringes — and typed four words: "make this sound better."
What came back was so aggressively generic, so perfectly bland, so utterly devoid of anything useful that it almost looped back around to impressive. ChatGPT had taken a mediocre performance review and returned a slightly shinier mediocre performance review. Same vague language. Same lack of specificity. Same fundamental uselessness — just with better sentence flow.
That was my introduction to AI. And it was entirely my fault.
Why "Make This Sound Better" Is a Trash Prompt
Here's the thing about AI that took me a while to fully internalize: the model is not psychic. It doesn't know who you are, what you're trying to accomplish, who your audience is, what format you need, or what "better" actually means to you. When you give it nothing, it gives you its best guess — and its best guess is always the most statistically average version of what you asked for. Competent. Generic. Forgettable.
The good news is that fixing this doesn't require a computer science degree or a course on prompt engineering. It requires a communications brain — which, if you're a writer, a marketer, a consultant, or anyone who has ever had to think carefully about audience and message, you already have.
A mentor of mine put it plainly when I was still in my "why isn't this working" phase: people with communications backgrounds are exactly who AI needs at the wheel, because the skill of prompting well is fundamentally the skill of writing clearly and thinking about your audience. I already knew how to do that. I just hadn't connected the dots yet.
Once I did, everything changed.
The Framework: Role + Task + Format
(AI Prompt)
The single biggest upgrade you can make to your AI output is learning to give your prompt three things every time: a Role, a Task, and a Format. That's it. Three components. They take thirty extra seconds to write and they eliminate roughly 80% of the generic, unusable output that makes people give up on these tools.
Here's what each one does and why it matters.
Part 1: Role — Tell It Who To Be
AI models are generalists by default. When you don't assign a role, you get a generalist response — something written for everyone, which means it's optimized for no one. Assigning a role is how you narrow the model's frame of reference and get output that reflects actual expertise rather than a Wikipedia summary of a topic.
Think of it less like giving a command and more like briefing a contractor. You wouldn't hire a graphic designer and say "make something." You'd say "you're designing a logo for a law firm that wants to convey trust and authority." The role is the brief.
Bad prompt: "Write a summary of this report."
What you get: a generic, surface-level recap that reads like a middle school book report. Technically accurate. Completely flat.
Good prompt: "You are a senior management consultant summarizing a quarterly performance report for a C-suite executive who has three minutes and no patience for filler. Write a summary that leads with the most critical finding and uses direct, confident language."
What you get: something that actually sounds like it was written by someone who understands the audience, the stakes, and the context. Because now, as far as the model is concerned, it is.
Pro tip: The more specific your role, the better. "You are a financial analyst" is good. "You are a financial analyst specializing in SaaS metrics writing for a non-technical founder" is better. Specificity is not overthinking — it's briefing.
Part 2: Task — Tell It Exactly What To Do
This is where most people stop — they write a task and call it a prompt. And a task alone is better than nothing. But a task without a role is like giving directions without a starting point, and a task without a format is like ordering food without specifying how you want it cooked.
The task is your what. Be specific about what you actually want. Not "write something about X" — what kind of something? For what purpose? With what goal? What should it do for the reader?
The more constraints you give, the better the output. This feels counterintuitive — shouldn't more freedom mean more creativity? In theory, yes. In practice, unconstrained AI output is usually the written equivalent of beige. Constraints are not limitations. They are instructions.
Bad prompt: "Write me a recipe for a healthy dinner."
What you get: grilled chicken breast with steamed broccoli and brown rice. Every time. The culinary equivalent of a beige wall. Technically food. Aggressively uninspiring.
Good prompt: "You are a home cook who specializes in weeknight meals for busy people who are bored of the same five dinners but don't want to spend more than 30 minutes in the kitchen. Write a recipe for a flavorful, vegetable-forward dinner that isn't a salad, doesn't require any specialty ingredients, and uses pantry staples most people already have. The person cooking this is a competent but not adventurous home cook who wants something that feels a little special without being precious about it."
What you get: an actual recipe with a point of view — something that sounds like it came from a person who cooks real food for real people on real Tuesday nights, not a nutrition blogger who has never experienced a time crunch. Same task — healthy dinner recipe — completely different output because the instructions told the model who it's cooking for, what the constraints are, and what "good" actually means in this context.
Real talk: If your task is vague, your output will be vague. There is a direct relationship between the specificity of what you ask for and the usefulness of what you get back. Every time.
Part 3: Format — Tell It What To Hand You
This is the most overlooked part of the formula and, once you start using it, the one you'll wonder how you lived without. Format tells the model what the output should actually look like — not just what it should say, but how it should be structured, how long it should be, what it should and shouldn't include.
Without format instructions, AI defaults to whatever structure it thinks is appropriate. Usually that means: introduction paragraph, three to five body sections with headers, conclusion that summarizes everything you just read. Every. Single. Time. It's the AI equivalent of a five-paragraph essay — the format of someone who wasn't told what format to use.
Bad prompt: "Help me write a professional email following up on a job application."
What you get: a three-paragraph email that opens with "I hope this message finds you well," references your "passion for the industry," and closes with "I look forward to hearing from you at your earliest convenience." It is inoffensive. It is invisible. It reads exactly like every other follow-up email in that hiring manager's inbox.
Good prompt: "You are a career coach who helps women in competitive corporate fields stand out in their job search. Write a follow-up email to a hiring manager after a first-round interview for a senior consultant role. The tone should be confident and direct — not eager, not stiff. Keep it under 150 words. Open with one specific reference to something discussed in the interview to show genuine engagement. Do not use the phrase 'I hope this finds you well,' 'passion,' or 'at your earliest convenience.' End with a single clear sentence that moves the conversation forward without begging for a response."
What you get: an email that sounds like a person wrote it. A specific, confident person who remembered what was said in the interview and doesn't write like a LinkedIn template. The ask is the same — follow-up email after a job interview — but the format instructions tell the model exactly what it should look like, how long it should be, what language to avoid, and what the closing needs to accomplish.
Format instructions can include:
Length ("under 200 words" / "no more than three paragraphs")
Structure ("no bullet points" / "use headers" / "write in prose")
Tone ("formal and precise" / "conversational, like texting a smart friend")
What to exclude ("no introduction" / "don't summarize at the end" / "avoid corporate language")
What to include ("cite specific mechanisms" / "include a real example" / "end with one actionable takeaway")
You are allowed to be this specific. You are encouraged to be this specific. The model will not be offended.
Putting It All Together
Here's the full formula in action, using the performance review that started this whole journey — because we've come a long way from "make this sound better."
The 2022 version: "Make this sound better."
Result: marginally shinier mediocrity. Zero stars.
The Role + Task + Format version: "You are an experienced HR consultant who writes performance reviews that are honest, specific, and legally defensible. Review the following performance document and rewrite it to clearly articulate the employee's key contributions, areas for development, and concrete goals for the next review period. Use professional but direct language. Structure the output in three clearly labeled sections: Strengths, Development Areas, and Goals. Each section should be three to four sentences. Do not use vague filler phrases like 'team player' or 'hard worker' without specific supporting evidence."
Result: something you can actually use, hand to a client, and stand behind. The task didn't change. The prompt did.
One More Thing Before You Go
This is the foundation. Role + Task + Format will get you 80% of the way there — but there's a lot more ground to cover when it comes to getting genuinely excellent output from AI tools, and I'll be going deeper on specific use cases in future posts: corporate deliverables, blog writing, marketing copy, productivity systems, and more.
For now, start here. Pick the next thing you were going to ask an AI to do and write the prompt three times — once the way you would have before reading this, once with a Role, and once with all three components. Compare the outputs.
You'll never go back to four words again.
Note: Always be mindful of what you paste into AI tools — especially free, consumer-facing versions. Sensitive client documents, personal data, and confidential information should never go into a model without understanding the platform's data privacy policy first. Future me has strong feelings about this. Learn from past me.
- Forever Wired & Wild⚡🌿
Want to go deeper? Coming soon: prompt engineering for corporate deliverables, teachers, marketing copy, and ADHD productivity systems. If there's a specific use case you want me to tackle first, drop me a message.
Comments