The quality of an AI assistant’s answer depends heavily on the quality of your request. Vague prompts get generic replies. A little structure goes a long way.
The five-part prompt framework
- Role or context: who is this for and what is the situation?
- Task: exactly what you want produced.
- Constraints: length, tone, format, things to avoid.
- Examples: a sample of the style or output you like.
- Success criteria: how you will judge a good answer.
Before and after
| Weak prompt | Stronger prompt |
|---|---|
| Write an email about the meeting. | Write a friendly, 120-word email to my team confirming Thursday’s 10am planning meeting. List the three agenda items below as bullets and ask them to add topics by Wednesday. |
| Explain inflation. | Explain inflation to a 15-year-old in under 200 words, using one everyday example about grocery prices. |
Iterate instead of starting over
Treat the first answer as a draft. Ask for changes: “shorter”, “more formal”, “turn this into a table”, or “what did you assume?”. Follow-up instructions are often faster than rewriting the whole prompt.
Ask it to ask you
End complex prompts with “Ask me any questions you need before you start.” It surfaces missing details early.
Always review the output
Even a perfect prompt can produce errors. Check facts, names, numbers and links before you publish or send anything.
