Meta prompt that writes prompts
Ask the model not to do the job, but to write the instruction that will be used for that job.
- Use
- Writing
- Setup
- No setup note
- Needs
- nothing extra
- Done when
- —
- License
- CC BY 4.0 · The card’s open source
What is it?
In general usage, a meta prompt is a request that produces or edits another prompt. You give the goal and the limits; the model produces a usable instruction draft. You check the quality of that draft separately on the target task.
Having a prompt written once is not the whole of optimization processes such as APE or OPRO, which select candidates through measurement. Here we do not invent an automatic success score.
When does it help?
Use it when you will repeat the same job, or when you want to turn a scattered request into a clear task text. Do not let the model decide a missing requirement on your behalf.
Examples
The situations and responses below are fictional teaching examples; they are not results from a model, tool or benchmark that was actually run.
Simple
Situation
You want a short, reusable instruction for workshop announcements.
Prompt
Prompt
Do not write the announcement itself; create the prompt that will have the announcement written.
Requirements: English, two sentences, only the given information, the form is an application; a place is confirmed by the confirmation email. No missing date is added.
Sample usage data: Drawing workshop, 20 people, date not yet set.
In the output, give the prompt together with this sample data; then stop.Sample output
Representative prompt: “Write two English sentences for a 20-person drawing workshop. State that the form is an application and that a place is confirmed by the confirmation email. The date is not set; do not add one.”
What did we get?
There is an instruction ready to use directly. How the model will respond to this instruction has not been tested yet.
Medium
Situation
An old prompt makes the model add details that are not in the source.
Prompt
Prompt
Old prompt: “Promote this event in an exciting and detailed way.”
Problem: From the source “The workshop is free”, a date and material support are being invented.
Rewrite the prompt. Keep promotion as the goal; add the conditions to use only the source information and to state missing fields. Let the sample source be “The workshop is free.” Output: the new full prompt and the single reason for the change. Do not produce the promotional text now.Sample output
New prompt: “Source: The workshop is free. Write a short promotion with this information. No date, venue or material support has been given; do not add them. If needed, say that these fields have not been specified yet.”
What did we get?
The error the instruction targets is clear. The effect of the fix should then be checked by getting output on the same source.
Advanced
Situation
Two prompt versions differ in scope; combining them must not increase permissions.
Prompt
Prompt
As a prompt designer, combine these two texts; do not carry out the task.
A: “Fix the language only in draft.md; source.md is read-only.”
B: “Edit all documents; publish when the work is done.”
Valid user limit: only draft.md, no publishing. Goal: a language fix that keeps the source meaning.
Write the new prompt in full; state the removed conflicts in two short bullet points. Do not read/write files or call tools.Sample output
New prompt: “Keeping the source meaning, propose a language fix only for draft.md. source.md is read-only. Do not change other documents; do not publish. In the result, report the changed phrases and any remaining uncertainties.”
What did we get?
The combination did not expand permissions. The instruction text should still be backed by real file permissions and an output check.
Where should you stop?
The word meta carries different meanings in different research. Meta-Prompting with expert calls is an orchestration architecture; structural Meta Prompting organizes the form of a task. Producing a prompt does not automatically make it the same method as these.
Related methods and how they differ
Sources
The Prompt Report: A Systematic Survey of Prompt Engineering Techniques (new tab) — Schulhoff, Sander; Ilie, Michael; Balepur, Nishant; Kahadze, Konstantine; Liu, Amanda; Si, Chenglei; Li, Yinheng; Gupta, Aayush; Han, HyoJung; Schulhoff, Sevien; Dulepet, Pranav Sandeep; Vidyadhara, Saurav; Ki, Dayeon; Agrawal, Sweta; Pham, Chau; Kroiz, Gerson; Li, Feileen; Tao, Hudson; Srivastava, Ashay; Da Costa, Hevander; Gupta, Saloni; Rogers, Megan L.; Goncearenco, Inna; Sarli, Giuseppe; Galynker, Igor; Peskoff, Denis; Carpuat, Marine; White, Jules; Anadkat, Shyamal; Hoyle, Alexander; Resnik, Philip. 2024-06-06; version read 2025-02-26. A secondary survey that classifies the scope of the terms meta prompting and prompt engineering; not, on its own, evidence of an original effect. Evidence level: relevant body sections of the original paper.
Large Language Models are Human-Level Prompt Engineers (new tab) — Zhou, Yongchao; Muresanu, Andrei Ioan; Han, Ziwen; Paster, Keiran; Pitis, Silviu; Chan, Harris; Ba, Jimmy. 2022-11-03; version read 2023-03-10. Provides a primary example of a model generating candidate instructions; APE additionally performs real evaluation and selection. Evidence level: relevant body sections of the original paper.
Master Prompts and System Prompts: The ChatGPT-5 Growth Blueprint (new tab) — Dan Martell. 2026-02-23; version read 2026-02-23. Shows how reusable master/system prompts are used in practitioner language. Evidence level: page body.
How this image was made
Original illustration made with Google Gemini · 1024 × 572
Generate an image: Create an original museum-grade computational sculpture photographed as a physically present installation, horizontal 16:9. Deep anthracite void, mineral porcelain whites, restrained ice-blue and amber accents, fine particles only where structurally meaningful, tactile surfaces, subtle volumetric illumination, believable depth, exceptional edge detail and intentional negative space. A rough amorphous mineral imprint is translated into a carefully folded thin white stencil; the stencil has not yet been used to create an object. Close lateral view, a deliberately empty fabrication space beyond it, subtle amber edges. Keep the image sophisticated and legible at mobile size: one dominant mechanism, a clear silhouette, no decorative overload. No text, letters, numbers, typography, captions, symbols, watermark, logo, fake chart, screenshot, interface, identifiable people or artist signature. Do not imitate any specific existing artwork.