Teams get uneven results from AI assistants mostly because everyone prompts differently. A shared structure fixes most of that. Use these four parts every time.
The four parts
- Role: who the assistant should act as, for example a senior sales coach or an HR business partner.
- Context: the background it needs: audience, goal, constraints and any material to work from.
- Task: the specific job, stated as one clear action.
- Format: how you want the answer: length, tone, table or bullets, language.
Before and after
Weak: "Write a proposal email."
Stronger: "You are a senior account manager. Our client is a Bangkok logistics company that asked about a two-day team workshop. Write a 150-word proposal email in a warm, professional tone, ending with two proposed meeting times. Reply in English."
Habits that improve results
- Give an example of a good output when you have one.
- Ask the assistant to ask you questions if something is unclear.
- Break big jobs into steps and check each one.
- Always verify facts, numbers and names before you use the output.
Build a shared prompt library
When someone finds a prompt that works, save it with a name, the tool used and an example. Give the library an owner and review it monthly. This turns individual tricks into a team asset.
Frequently asked questions
Do I need technical skills for prompt engineering?
No. Business prompting is clear writing: say who, why, what and in which format. Only developer tracks such as API and agent building need programming.
Is one prompt framework enough?
A simple framework like role, context, task and format covers most daily work. Add examples and step-by-step checking for complex tasks.
Can the same prompts work in ChatGPT, Claude and Copilot?
Mostly yes. The structure transfers across tools, but test each prompt in the tool you actually use, since responses differ.