You write a good AI prompt by describing four plain things in ordinary language: who the AI should act as, what you want it to produce, who will read it, and what format you want back — no technical vocabulary required.
Here's the core insight: a prompt is just a brief. If you can explain a task to a new freelancer over coffee, you can write a prompt. You don't need special syntax, code, or secret phrases. You need to be specific about the outcome you want, the way you'd be specific with a contractor about a kitchen renovation.
The four-part prompt formula
Every solid prompt has four parts, and you can write them in any order:
- Role — who the AI should pretend to be. "Act as a patient middle-school science teacher." This sets vocabulary and tone.
- Goal — the actual deliverable. "Write a 400-word explainer on why the sky is blue." Vague goals produce vague output, so name the length, the topic, and the purpose.
- Audience — who reads the result. "The reader is a curious 12-year-old with no science background." This single line changes word choice more than any other instruction.
- Format — the shape of the answer. "Give me three short paragraphs, no bullet points, no jargon."
A non-technical user can write all four in under a minute. The trick is to resist the urge to type a bare question like "explain the sky" and instead spend fifteen extra seconds on the brief.
A concrete before-and-after example
Suppose you run a small bakery and want a social media post about your new sourdough.
Weak prompt: "Write a post about sourdough."
What you get back is generic — it could describe any bakery anywhere, and it'll probably use words like "artisanal" that don't sound like you.
Strong prompt: "Act as the owner of a family bakery in a small town. Write a 60-word Instagram caption announcing our new sourdough starter, made from a recipe my grandmother used. The audience is local customers who already follow us. Keep it warm and casual, use one emoji, and end with a question that invites comments."
Same tool, same person typing, wildly different result. The second prompt tells the AI the role, the goal, the audience, and the format. It also smuggles in a detail — the grandmother's recipe — that gives the AI something specific to work with. Specific details are the fuel; without them the AI fills the gap with clichés.
Why this works, and where it breaks
Language models predict likely next words based on the context you give them. A detailed brief narrows the range of likely words, which is why the output lands closer to what you wanted. This is also why the formula fails in predictable ways.
If your audience description contradicts your goal — say, "write for experts" but "avoid all technical terms" — the AI has to guess which instruction wins, and it usually picks the wrong one. If you ask for something it has no way to know, like your company's internal pricing, it will invent a plausible number rather than admit ignorance.
And if you ask for five different things in one prompt (a blog post, a tweet, and an email), quality drops on all three because the model spreads its attention.
The fix for most of these is iteration, not perfection. Treat the first response as a draft, then reply with a correction: "Too formal. Rewrite it as if you're texting a friend." You don't need to start over — the AI keeps the earlier context, so a short follow-up instruction is often enough.
Honest limits
This approach costs you time upfront. A four-part prompt takes longer to write than a bare question, and for trivial tasks — "what's the capital of Peru?" — the extra effort is wasted.
The formula also won't rescue a task you can't describe clearly. If you don't know what you want, the AI can't guess it, and you'll get fluent-sounding text that misses the point. Finally, prompt quality doesn't guarantee factual accuracy.
A well-written prompt can still produce a confident, wrong answer, so anything with numbers, names, or dates still needs a check against a real source before you publish it. If you want a deeper walkthrough on catching those errors, the guide on how to stop ChatGPT from making up fake facts in my content covers that side of the workflow.
One last tip that surprises most beginners: save your best prompts. When a prompt produces output you actually like, paste it into a notes file with a one-line label. After a month you'll have a personal library of reusable briefs, which beats trying to remember what you typed six weeks ago. Zero-prompt tools like AI-Mind take the opposite approach — they generate content without you writing a brief at all — which is useful for speed but gives you less control over voice and angle than a well-crafted prompt does.