Meta-prompting
Using a model to generate or refine prompts for itself or other models — prompting about prompting.
Think of it like
Like a writer editing their own instructions to make them clearer.
Example
You give Claude a task and ask: "Generate the best system prompt for an AI that does X." It writes a refined prompt you can reuse.
How it actually works
Meta-prompting is a form of in-context learning and reflection. The model reasons about what instructions work. It can also be used for prompt optimization: iteratively improve a prompt by having the model critique and regenerate it. This is experimental and cost-intensive, but can yield domain-specific prompts better than hand-written ones.
For product teams
Automates prompt engineering; useful for discovering patterns.
For engineers
Implement a loop: generate prompt → test → critique → refine. Measure impact with a held-out eval set.
Related
- Prompt optimization — Related to prompt optimization and automatic prompt engineering.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome