Meta-prompting is the use of a model to create, critique, transform, or optimize prompts. Instead of asking the model to solve the end task directly, a meta-prompt asks it to improve the instructions, examples, structure, or evaluation criteria used by another prompt.
Common uses include:
- generating candidate prompts for an application workflow;
- rewriting prompts for clarity or lower token usage;
- creating prompt variants for A/B tests;
- converting informal requirements into a structured prompt template;
- critiquing a prompt for ambiguity or missing constraints; and
- producing rubrics for LLM-as-a-Judge evaluations.
Meta-prompting is related to prompt optimization, but it is not the whole optimization process. A generated prompt still needs offline tests, human review, and production evaluation. The model may produce prompts that sound rigorous while embedding false assumptions or fitting too narrowly to the examples it saw.
In professional systems, meta-prompting is most useful when paired with version control, task-specific evaluations, and clear acceptance criteria.
Comet describes meta prompting as using LLMs to optimize prompt templates in Meta Prompting: Use LLMs to Optimize Prompts for AI Apps and Agents.
The LLM Knowledge Base is a collection of bite-sized explanations for commonly used terms and abbreviations related to Large Language Models and Generative AI.
It's an educational resource that helps you stay up-to-date with the latest developments in AI research and its applications.