How to Write Better AI Prompts: A 6-Part Framework

Most disappointing AI answers come from vague questions, not weak models. If you ask “write me an email about the project delay,” the AI has to guess who the email is for, how serious the delay is, what tone to use, and how long it should be. It fills those gaps with safe, generic choices, and you get a safe, generic email.

The fix is to give the model what a capable new colleague would need to do the job well. This guide gives you a six-part framework for doing that, with examples you can adapt. It works the same way in ChatGPT, Claude, Gemini and most other assistants.

The six parts of a good prompt

You won’t need all six every time. For a quick question, one or two are enough. For anything you’d spend more than ten minutes on yourself, use most of them.

1. The task: say exactly what you want

Start with a clear verb and a clear output. “Help me with my CV” is a topic. “Rewrite the summary section of my CV in three sentences” is a task.

• Weak: Tell me about marketing for my shop.
• Better: Suggest five low-cost ways to get more customers into a small bakery.

2. Context: the background the AI can’t know

The model knows nothing about your situation unless you tell it. Include who you are, who the output is for, and anything unusual about the situation.

I run a small bakery in a residential area. Most customers are families who walk past on the way to school. We have no marketing budget, but I can spend about two hours a week on it.

This one paragraph changes the answer completely. Without it, you’ll get advice about paid ads and influencer campaigns. With it, you’ll get ideas that fit a bakery near a school.

3. Audience and purpose: who reads it and what should happen

Say who the output is for and what you want them to do after reading it.

• This is for my manager, who hasn’t followed the project closely. I want her to approve a two-week extension.
• This is for beginners who have never used a spreadsheet.

Purpose matters because it tells the model what to put first. An email asking for approval leads with the request. A status update leads with the news.

4. Format: describe the shape of the answer

If you don’t specify a format, you get the model’s default, which is often too long and full of bullet points. Be specific:

• Under 150 words.
• A table with columns for option, cost and time needed.
• Plain paragraphs, no headings or bullet points.
• Three versions, each with a different tone.

5. Examples: show, don’t just tell

An example is the fastest way to get the style you want. If you have an email you liked, a past report, or a few lines in your own voice, paste them in.

Here’s an update I wrote last month that my team liked. Match its tone and length: “Quick update on the website: design is done, development starts Monday, and we’re still on track for the 15th. One risk: we’re waiting on product photos from the supplier.”

One good example usually works better than a paragraph of instructions about tone.

6. Constraints: what to avoid

Tell the model what not to do. This is where you remove the habits that make AI writing obvious.

• Don’t use phrases like “I hope this email finds you well.”
• Don’t invent statistics. If you’re unsure of a fact, say so.
• Avoid jargon; my readers aren’t technical.

Before and after

Here’s a real example of the framework in action.

Before:

Write an email about the project delay.

After:

Write an email to a client, Sara, telling her our website launch is moving
from 1 November to 15 November.

Context: the delay is because her team sent the product photos two weeks
late. We don't want to sound like we're blaming her; we value the
relationship.

Purpose: she should accept the new date and confirm she'll send the
remaining 10 photos by 4 November.

Format: under 120 words, friendly but professional, with one clear request
at the end.

Avoid: apologizing more than once, and phrases like "unforeseen circumstances".

The second prompt takes about 60 seconds longer to write. It also gets you an email you can send after a light edit, instead of one you have to rewrite.

Three habits that help even more

Ask the AI to ask you questions first. For bigger tasks, end your prompt with: “Before you start, ask me up to five questions about anything that’s unclear.” The model will often spot gaps you didn’t think of.

Iterate instead of starting over. If the first answer is 70% right, don’t rewrite your prompt from scratch. Say what to change: “Good. Make it shorter, drop the second paragraph, and make the request more direct.”

Save prompts that work. When a prompt gives you a great result, save it in a notes file with blanks for the parts that change. Over time you’ll build a personal library for the tasks you do most often, like weekly reports, meeting summaries and replies to common questions.

A copy-paste template

Prefer a form? Our free AI Prompt Builder assembles these six parts for you.

Task: [what you want, starting with a verb]
Context: [who you are, the situation, anything unusual]
Audience and purpose: [who reads it, what they should do or understand]
Format: [length, structure, tone]
Example: [paste something in the style you want, if you have one]
Avoid: [phrases, assumptions or content to leave out]

Common mistakes

• Cramming several tasks into one prompt. “Summarize this report, write an email about it and make a slide outline” works better as three separate requests, each checked before you move on.
• Trusting facts without checking. A clear prompt makes errors less likely but doesn’t remove them. Our five-minute fact-checking routine shows how to verify what matters.
• Being too polite to be specific. You don’t need “please” and “if possible” in every line. Clear instructions are easier to follow than polite, vague ones.

Key takeaways

• Give the AI what a new colleague would need: the task, context, audience, format, an example and constraints.
• Specific prompts take a minute longer to write and save much more time in editing.
• Refine answers with follow-up instructions instead of starting over, and save the prompts that work.

Next, learn how to apply these ideas to long material in our guide to summarizing long documents with AI.