Structured Prompts

Introduction

A structured prompt uses clear sections (role, task, context, examples, format) separated by delimiters. It's how you build prompts that are testable, editable and extensible.

Beginner analogy: Like a form with labelled fields vs a blank piece of paper. Forms are predictable, scalable and easy to edit.

Understanding the topic

Core concepts to understand:

• Sections: role, task, context, examples, constraints, format, query.
• Delimiters: XML tags, ###, triple backticks.
• Each section is editable without touching the others.
• Easier to A/B test — change one section, measure impact.

Syntax reference

Visual workflow / architecture:

<role>You are a code reviewer.</role>
<task>Review this Python function.</task>
<context>{code}</context>
<format>JSON: [{line, severity, fix}]</format>
<query>Begin.</query>

Real-world use

Cursor, Claude Projects and OpenAI Assistants all internally use structured prompts to keep features stable as they evolve.

Best practices

• Use XML tags on Claude (it was trained to respect them).
• Keep section order consistent across a prompt library.
• Add a /* version */ comment to track changes.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Why use structured prompts?
• Q2. Why does Claude prefer XML delimiters?
• Q3. How does structure help A/B testing prompts?
• Q4. Where do you store structured prompt templates?