XML Tagging & Extraction

Expert

Format inputs and outputs using strict XML tagging structures for stable parsing.

01 The Concept

While JSON is standard, XML tags are easier for models (especially Claude) to segment and parse without escaping errors.

XML Schema Tagging surrounds parameters and text with matching tags (e.g. `<transaction>`...`</transaction>`), preventing parsing degradation on messy inputs.

02 Weak vs. Strong

EX 01XML Tagged Prompt

Summarize the customer request enclosed in <email> tags. <email> Hello, I paid $20.00 but got no confirmation. </email> Format output strictly as: <summary> [Summary text] </summary>

→ Why it works

Enclosing both input and output in tags makes it trivial for regex parsers to extract the summary without getting confused by text formatting.

03 Key Points

01Boundary Isolation: Using tags to isolate untrusted data from system rules.

02Nesting Structures: Creating hierarchies (e.g. <invoice><item>...</item></invoice>).

03Parsing Stability: XML tags prevent JSON escaping issues when text contains quotes.

04Semantic Labeling: Giving clear semantic names to tags to guide model attention.

05Tool Orchestration: Structuring model calls to trigger specific functions using tags.

04 Model-Specific Notes

Claude is natively optimized for XML. Using tags is the single most effective way to structure Claude prompts.

05 For Your Role

It's like putting different papers into labeled folders (e.g. 'Invoices', 'Letters') so they don't get mixed up on your desk.

Next: Jailbreak Forensics →