A vague question makes ChatGPT fill in the blanks. Sometimes its guesses are useful. Sometimes they produce a polished answer for the wrong task.
OpenAI’s current guidance emphasizes clear, specific instructions, enough context, and iterative refinement. In practice, that means explaining the decisions that matter and improving the answer through conversation.
Why a reasonable question can still miss
Write an email about the meeting.
ChatGPT does not know:
- Who will receive the email
- Whether you are inviting, confirming, or following up
- Which facts must be included
- How formal the relationship is
- What length would be useful
The request is not “bad.” It is incomplete for the result you probably have in mind.
Five useful parts of a better question
Task
Say what you want the AI to do: explain, draft, compare, or plan.
Context
Give the background it cannot know from the task alone.
Audience
Say who will read or use the result.
Limits
Include length, tone, deadline, budget, or facts it must not add.
Format
Ask for a short email, checklist, table, or a few clear options.
A real email example
In the author’s first documented email test, the original request was written in Chinese:
给我在美国的好朋友写一封邮件，询问他的近况，以及向他请教AI在那边的社会影响和发展趋势。
It asked for an email to a close friend in the United States, checked how he was doing, and asked about AI’s local social impact and development. The resulting draft understood that purpose, but the greeting still contained “[Friend’s Name]” because the preferred name had not been provided.
Write a friendly email in natural American English to a close friend in the United States. First ask how he and his family are doing. Then ask for his personal observations of how AI is affecting daily life and work there. Keep it under 180 words and ask no more than three questions. His preferred name is Li.
The clearer version removes guesses about language, name, length, and number of questions. It is not longer for its own sake; each added detail changes the draft.
Improve a weak answer without starting over
Read the first answer and name the problem. A useful follow-up might say:
- “The tone is too formal. Make it warmer.”
- “Keep the first paragraph, but remove repeated ideas.”
- “Use words a complete beginner would understand.”
- “You added a deadline I did not provide. Remove it.”
- “Ask me three questions before you revise this.”
Specific feedback gives ChatGPT a new constraint. “Try again” gives it much less information.
A short question can still be clear
Explain this paragraph in plain English. Keep every date and number unchanged. Paragraph: [paste text]
This works because the task, reading level, important limit, and source material are all present. There is no need to add a role, elaborate background, or special phrase.
Check your question before sending
- Is the task clear?
- Did I include the background that changes the answer?
- Does ChatGPT know who the result is for?
- Did I state any important limits?
- Would a format make the answer easier to use?
- Did I remove private information?
- Can I judge whether the result is correct?
More examples
Use a starting point instead of a blank box.
The prompt library gives you 50 practical requests to customize for learning, writing, work, and everyday life.
Browse 50 beginner prompts →Common questions
Do I need to use a prompt template?
No. A template can remind you what to include, but ordinary language is enough when the goal and relevant context are clear.
Does a longer question always get a better answer?
No. Add details that change the answer. Unrelated background can make the request harder to follow.
What if I do not know what details ChatGPT needs?
Ask it to list the missing information or to ask you up to three questions before answering.
Should I start over when the answer is weak?
Usually not. Tell ChatGPT what is wrong, add the missing context, and ask it to revise the useful parts.