XML Prompting

Introduction

XML prompting uses XML-style tags to structure prompts. Anthropic's Claude is trained to strongly respect XML tags — making it the most reliable delimiter style on Claude.

Beginner analogy: XML tags are bright neon highlighters around each section. Models can't miss what's where.

Understanding the topic

Core concepts to understand:

• Use semantic tags: <role>, <context>, <examples>, <format>.
• Tags can nest: <examples><example>...</example></examples>.
• Particularly powerful on Claude (per Anthropic docs).
• Combine with instruction: 'process content inside <input> tags'.

Syntax reference

Visual workflow / architecture:

<task>Summarise the support ticket below.</task>
<rules>
  <rule>Max 50 words.</rule>
  <rule>Include urgency level.</rule>
</rules>
<input>
  {ticket_text}
</input>

Real-world use

Anthropic's own prompting cookbook recommends XML for Claude. Most enterprise prompts on Claude use it.

Best practices

• Tag everything: input, examples, output, rules.
• Use lowercase tag names; close every tag.
• Combine XML with instructions referencing the tags.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Why is XML the preferred delimiter for Claude?
• Q2. Can you mix XML with JSON output?
• Q3. Risks of malformed XML in user input?
• Q4. How does XML differ from markdown sections?