When you ask ChatGPT, Claude or Gemini to help with work, your prompt is the request and supporting information you provide. A useful prompt makes the job clear: what to produce, who will use it, and what a good result must include. You can start with ordinary language.
1. Describe the finished result
Name one deliverable before adding a role or a long list of rules. A request for a decision memo needs different instructions from a request for ideas.
- Task: what should the AI produce?
- Reader and purpose: who will read it, and what should they do next?
- Output: what length, structure or file format do you need?
- Checks: which facts and requirements must the result satisfy?
2. Give the missing context
Include the notes or source material needed for this task. Separate supplied facts from your instructions. If a style matters, show a short example and say which feature to follow.
OpenAI and Anthropic both recommend explicit instructions, relevant context and examples. The practical test is whether another person could understand the assignment without guessing.
Write an email about the delay.
Draft an email to the project lead. Example facts: our test run needs two more days; the draft is ready; the final check is still pending. Explain the impact, propose sending the draft today, and ask whether the revised deadline works. Use a calm tone and no more than 120 words.
Illustrative requests: the second makes the audience, facts and next action explicit.
3. Copy this starting prompt
Replace the brackets with your own material. Remove fields that do not matter; more instructions are useful only when they clarify the job.
Task: Create [one deliverable].
Reader and purpose: [who will use it and why].
Source material: [paste the relevant facts or notes].
Output: [structure, length and format].
Must include: [up to three requirements].
Do not invent missing facts. Mark anything that needs confirmation.
If an essential detail is missing, ask one focused question. Otherwise produce the draft and list any unresolved checks.4. Choose a method, then test it
Start from a prompt for the same kind of work. An email prompt can help organize a message; a presentation prompt should help choose a story and slide structure. An expert-inspired method is a starting point you adapt to your own sources and audience.
Read the result against your checks. Correct one recurring problem and try another representative task before keeping the instruction. Clear prompting can help, but it cannot guarantee factual accuracy or the same result across AI models.