How to tell AI what you need
A task, context, and an example: three building blocks for a useful first AI prompt.
Sample draftPrepared ≈ 4 min read
When you first try AI, it is tempting to hunt for the “right prompt”. I would start with something simpler: explain what you want to achieve. Think of briefing a colleague who does not yet know the circumstances.
What is a prompt?
A prompt is a request or instruction you give a model. Here we focus on text: a question, a task description, or a follow-up instruction.
1. Start with the result
“Help with an email” leaves a lot to guess. Who is it for? What is the purpose? Do you need to apologise, make an arrangement, or share an update?
Start with an action: draft an email, compare options, explain a concept, or review a draft. If there are several tasks, list them separately.
Write an email about a meeting.
Draft a short email to a partner: move our meeting to next week and offer two alternative times.
2. Add useful context
Include the circumstances that change the answer. For an email, these might be your relationship with the recipient, the reason for the change, and your available times. For a tool comparison, include the task, constraints, and selection criteria.
Ask yourself: “What would a person need to ask before starting this task?” Those answers become your context.
What does context mean here?
Context is the information that helps make sense of a request: the audience, circumstances, constraints, source text, or an example of the desired result.
3. Describe the output
Specify length, tone, and structure: for example, under 100 words, friendly, with a subject line and one question at the end. Make the requirements easy to check.
You can copy and adapt this prompt. The schedule below is a fictional example.
Draft an email to a partner to move our meeting to next week. Context: we have worked together for a long time. My schedule has changed. Offer Tuesday at 11:00 or Thursday at 15:00. Format: a subject line and a body under 100 words. Use a friendly, professional tone without corporate jargon. End by asking which time works. Do not invent the recipient’s name, calendar dates, or additional reasons for the change.
4. Refine one step at a time
Read the answer and name a specific change. “Make it better” invites another guess. “Remove the apologies and shorten the first paragraph” gives a clear direction.
Keep the parts that work and revise the weaker part. You do not have to start a new conversation for every edit.
Shorten the email to 60 words. Keep both alternative times and the friendly tone. Remove the repeated explanation for rescheduling. Do not add new facts.
5. Check before using
A useful prompt does not replace a review. Before sending the email, I would check:
- Are the times, dates, and recipient’s name correct?
- Did any reasons or commitments appear that you never provided?
- Is it clear what the recipient should do next?
- Does the tone fit your relationship?
Try this sequence on a small task: result → context → format → refinement → review. Your first attempt will give you a better sense of the information that matters for your own work.
Sources and scope
The guidance on clear requests, context, tone, and iterative refinement draws on OpenAI’s official ChatGPT prompting guide ↗. Source checked on 4 November 2026.
The meeting scenario, templates, and explanations were prepared for this sample. Comparing the quality of different models is outside its scope.
Curiosity is a good place to start.
← All notes