Prompt with Purpose
The difference between a useful AI response and a frustrating one is usually context. Not better AI — better input. This experiment shows you the four layers of a strong prompt and gives you a ready-to-use example you can take to Claude.ai, Gemini, or any AI tool you use.
Every strong prompt is built from the same four pieces. You do not always need all four, but the more you include, the better the output.
Role
Who is the AI? Setting a role gives the AI a lens — it shapes tone, vocabulary, and depth before you have asked for anything.
Task
What do you need? Be specific about format, subject, and purpose. Vague tasks produce vague answers.
Constraints
What are the rules? Word count, tone, audience, things to avoid. Constraints feel limiting but they sharpen the output.
Examples
Show, don’t just describe. Paste a sentence with the voice or style you want matched. This works better than trying to describe tone in words.
Weak prompt vs strong prompt
Here is the same request written two ways. The subject is identical. The results will not be.
“Help me write an email to my team about the new policy.”
“I’m the operations manager for a 20-person remote team. Write an email explaining our new flexible working hours policy. The team asked for more schedule autonomy and this responds to that feedback. Warm and collaborative tone. Under 300 words. Include a short FAQ covering core hours and client meetings.”
Why the strong version works
- It has a role — operations manager
- It has a clear task — explain the policy
- It sets constraints — warm tone, under 300 words, include FAQ
- It provides context about the audience and why the policy exists
A ready-to-use example prompt
Here is a fully built prompt you can copy, adapt, and paste into any AI tool. The colour coding shows which layer each part belongs to.
You are a technology journalist writing for non-technical readers. Write a short LinkedIn post announcing a new AI tool I have been testing. The tool helps small business owners automate their customer follow-up emails. Keep it under 150 words. Warm and conversational tone. No jargon. End with a question that invites comments. Match this style: “I’ve been testing something quietly for a few weeks. Here’s what surprised me.”
Now build your own
Use the four layers to write your own prompt below. Work through them one at a time:
Role — who should the AI be?
Start with “You are a…” and describe the expertise or perspective that fits your task. A technology journalist. A friendly teacher. A senior product manager. A copywriter with a punchy style.
Task — what exactly do you need?
Name the format, the subject, and the purpose. Not “write a post” but “write a 150-word LinkedIn post announcing X to an audience of Y, so they do Z.”
Constraints — what are the guardrails?
Word count, tone, audience level, things to avoid, format requirements. Pick two or three that matter most for your task. Do not leave everything open.
Examples — what should it sound like?
Paste one or two sentences that have the voice or style you want matched. This is the most underused layer, and often the most powerful. Instead of saying “make it sound like me,” show it what you mean.
One more thing
- If the first response is not quite right, do not start over with a new prompt
- Tell the AI what to change — “This is good but too formal” or “shorter sentences please”
- Iteration is part of the process and gets you there faster than rewriting from scratch
Take it further
Open Claude.ai and try writing your prompt from scratch using the four layers. Start with just Role and Task, then add Constraints and Examples one at a time. See how each layer changes the output. That comparison is the experiment.