A good prompt gives AI enough direction to be useful. It does not need to include every detail, use special vocabulary, or get everything right on the first attempt.
Think conversation, not command.
Prompting is less like programming a machine and more like briefing a capable collaborator. You set a direction, review what comes back, and help the AI understand what a better result looks like.
Use the response to discover what is missing. Add context, correct assumptions, request alternatives, and keep steering until the result fits your needs.
Build a useful brief
Start with four ingredients.
The task
Say clearly what you want the AI to do.
Draft an email announcing our new flexible-working policy.
The context
Share the background the AI needs to make good decisions.
The email is for 60 employees across France and the UK. Some employees are concerned that the policy will reduce flexibility.
The desired result
Describe what a successful response should achieve.
Employees should understand what is changing, feel reassured, and know where to ask questions.
The format and boundaries
Explain how the response should be presented and what it should avoid.
Keep it under 300 words. Use a warm, direct tone. Include a subject line and three bullet points. Avoid corporate jargon.
What counts as useful context?
Context is often the most important part of a prompt. It can include who the work is for, why it is needed, what has already happened, what the audience knows or cares about, and any relevant facts, examples, documents, or source material.
Put it together
Here is the complete prompt.
Draft an email announcing our new flexible-working policy. It is for 60 employees across France and the UK, and some people are concerned that the change will reduce flexibility. The email should explain what is changing, reassure employees, and tell them where to ask questions. Keep it under 300 words, use a warm and direct tone, include a subject line and three key bullet points, and avoid corporate jargon.
This gives the AI a task, relevant context, a definition of success, and clear boundaries. That is enough to produce a useful first draft.
Keep steering the conversation.
You rarely need to fit every instruction into one enormous prompt. Each response gives you something concrete to react to. Tell the AI what is working, what is not, and what should change next.
Make the opening warmer and less formal.
What important information might be missing?
Give me three alternatives with different tones.
This feels too generic. Make it specific to a small consulting company.
Keep the structure, but cut the length by 30%.
Before rewriting, ask me three questions that would improve the result.
Explain which assumptions you made.
Check this against the source material I provided.
This is called multi-turn prompting. Each response becomes material for the next turn. You are not starting over. You are guiding the work.
A framework to remember
Task + Context + Outcome + Format
Help me [task].
Here is the relevant context: [context].
A good result should [outcome].
Please present it as [format], and follow these boundaries: [constraints].
The short version
Clarity beats cleverness.
A good prompt helps the AI understand your situation. Treat its response as a draft, be specific about what is and is not working, and keep steering. You do not need the perfect first prompt. You only need to begin the right conversation.
Bring practical AI skills to your team →