Guide
Meta prompting
Meta prompting is using a model to draft or revise the prompt for another task. The first answer is a prompt, not the final work product. You then run that prompt, often in a fresh chat, on the real input.
Updated 2026-09-23
Before and after
Before
Write me a good prompt for a product launch email.
After
Write a prompt I can reuse for a product launch email. The prompt must include slots for product name, audience, one proof point, and the single action. It must require a subject line, under 150 words of body, and no discount unless the slot is filled. Output only the prompt, not a sample email.
How to do it
1. Describe the job, not the wording
Say who the prompt is for, what input they will paste, and what done looks like.
2. Ask for a prompt, not the deliverable
“Write the prompt I should paste” stops the model from doing the task itself.
3. Require slots for missing facts
Tell it to leave bracketed fields for names, dates, and numbers it does not have.
4. Run the result in a new message
Paste the prompt, fill the slots, and judge the real output. Edit the prompt where the output is wrong.
When to use this
Use meta prompting when you know the outcome you want but you are unsure how to specify role, constraints, or format. Do not use it as the last step for high-stakes work. Read the generated prompt, delete anything vague, and add facts the model cannot know, such as your audience and your real limits.
Meta prompting versus a prompt enhancer
Meta prompting is the pattern. A prompt enhancer is a tool that applies the pattern for you. Prompt God is a Chrome extension that rewrites the text already in the chat box. You stay in ChatGPT, Claude, or Gemini, click enhance, and the rough line becomes a structured prompt. That is meta prompting with the draft step automated.
The risk is the same either way: a polished prompt can still ask for the wrong thing. The enhancer does not know your unstated goal. Add the facts, then enhance.
Sources
Questions
What is meta prompting?
Meta prompting asks a model to write or improve a prompt. You then run that prompt on the real task.
Is meta prompting the same as a prompt generator?
A prompt generator is a product that does meta prompting for you. The technique is the request: produce a prompt, not the finished article or email.
Should I trust a generated prompt as-is?
No. Fill in facts the model cannot know, delete vague lines, and judge it by the output of a real run.