실무에서 검증된 디자인, 개발, AI workflow 강의를 제공합니다.

# The Structure of a Good Prompt
When you delegate work to an LLM, output quality often depends more on prompt design than on the model. Building a reviewable structure first is the key.

## What should you make explicit first?
Lock down three things first: role, goal, and output format. {tip:Role assignment}(a sentence giving the model a perspective or expertise, e.g. “You are a technical editor.”) sets tone and judgment, and stating the output format makes post-processing easy.

### Do examples really help?
Yes. One or two good examples ({tip:few-shot}(showing a few input→output pairs so the model learns the pattern)) convey format and tone more precisely than long explanations.

## What if answers keep getting verbose?
Constrain length and format with numbers. Stating “within 3 sentences” or “5 bullets” lets the model self-regulate its length.

### What is a more reliable way to enforce format?
Provide a JSON schema or template and insist “output only this format.” It is especially effective when you will parse the result downstream.

From role, goal, and output format to few-shot examples and length constraints — the prompt structure behind consistent LLM results.