EN
Sign inGet started
Glossary

Metaprompt

What is Metaprompt?

A metaprompt is an instruction that asks a language model to draft or improve prompts or system instructions itself. You describe the goal, audience and desired tone and supply examples; the model then produces structured instructions with a role, rules and limits. This yields good first versions faster, which are then tested and refined by people. Example: “customer service for a bicycle shop” becomes a complete draft set of instructions.

How KI.ordo helps

KI.ordo’s template “Write instructions for an agent (metaprompt)” turns a goal, audience and examples into new instructions, including rules on what the agent should not do.

Learn more →: Metaprompt

Matching templates

Topics

  • prompting
  • agents
  • instructions

Try KI.ordo

Agents, flows and templates in one workspace – free 24-hour demo.

Start demo
← All terms

Related terms

Prompt

A prompt is the input that tells an AI model what to do – a question, a task or an instruction, often supplemented with context, examples and the desired format. The quality of the answer depends heavily on how clear and complete the prompt is. For example, rather than “Write something about our product”, use “Write three short benefits of product X for trade businesses, in a factual tone”.

  • prompting
  • input
  • +1
Template

Negative prompt

A negative prompt specifies what an AI model should explicitly not generate or do. In image generation, for example with Stable Diffusion, it is a separate input field for unwanted elements such as “blurry” or “text in image”. With language models, it takes the form of rules in the instructions, such as “Do not make up prices” or “Do not give legal advice”. Such limits make results more predictable but do not replace review.

  • prompting
  • instructions
  • +1

Recursive prompting

Recursive prompting refers to techniques in which a language model's output is fed back in as new input – for instance to have a draft reviewed and revised, or to break a large task down step by step into subtasks. For example, a draft proposal is produced first, the model then assesses it against a checklist, and finally it is improved. This often raises quality but costs extra processing time and needs a clear stopping criterion.

  • prompting
  • revision
  • +1

Few-shot learning

Few-shot learning is a model’s ability to solve a task from only a handful of examples. With language models this usually happens directly in the input: you show two or three pairs of question and desired answer, and the model applies the same format and style to new cases – without retraining. Example: three correctly categorised customer enquiries are often enough for further enquiries to be sorted the same way.

  • Prompting
  • Examples
  • +1