How to write a good AI prompt: a step-by-step guide

How do you write a good AI prompt?

A good prompt gives the AI four things: context, a role, a clear task, and the output format you want. Most weak answers come from skipping these — the model has to guess what you mean. Add them and you'll get sharp, useful results from ChatGPT, Claude, Gemini, or any other tool. Then treat the first answer as a draft and refine it.

The four parts of a great prompt

1. Context — the background the AI needs: who it's for, what you've tried, any constraints. "I'm emailing a client who missed a deadline; tone should stay warm."
2. Role — who the AI should act as: "You are an experienced customer-success manager."
3. Task — the specific thing to do: "Write a 3-sentence reply that asks for a new date without blaming them."
4. Format — how you want it: length, structure, tone. "Under 80 words, friendly, no jargon."

You won't need all four every time, but the more ambiguous the task, the more they help. This is the heart of prompt engineering.

Before and after: a weak prompt vs a good one

Weak: write a post about our new feature

You'll get something generic, because the AI is guessing the audience, tone, length, and the feature itself.

Good: You are a product marketer. Write a LinkedIn post (under 120 words, confident but not hypey) announcing our new AI feedback feature for our app that helps people learn AI. Audience: busy professionals. End with a question to drive comments.

Same model, dramatically better output — because it now has context, a role, a task, and a format. For more reusable structures, see the 7 prompt patterns that work everywhere.

How to refine the first answer

The first answer is a draft, not the deliverable. Improve it with quick follow-ups:

• Adjust: "Make it shorter / more formal / more specific."
• Add an example: show one sample of "good" and ask it to match.
• Ask for options: "Give me three versions, ranked."
• Push back: "That intro is generic — try a sharper hook."

Two or three rounds of this beats hunting for a perfect one-shot prompt.

Common prompting mistakes

Being vague. "Make it better" gives the model nothing to aim at — say what "better" means. Skipping context. The AI can't read your mind about audience or constraints. Accepting the first draft. Iterating is where the quality comes from. Chasing magic words. There aren't any — clarity beats tricks. If your prompts keep underperforming, see why your AI prompts aren't working, and practice hands-on in the Prompt Lab.