Structured Output
Get data, not prose.
The Pattern
Section titled “The Pattern”Constrain model output to a specific format (JSON, XML, typed schema) that can be parsed programmatically.
When to Use
Section titled “When to Use”- Output feeds into downstream code
- You need specific fields, not free-form text
- Validation and error handling matter
- Building pipelines where format consistency is critical
Implementation
Section titled “Implementation”Prompt-based:
- “Respond only with valid JSON matching this schema: {…}”
- Provide examples of the exact format
API-based:
- Use structured output / JSON mode features when available
- Define schemas that the API enforces
- Start with the simplest schema that works
- Include field descriptions in the schema
- Handle parsing failures gracefully
- Test edge cases (empty arrays, null values, long strings)
Anti-patterns
Section titled “Anti-patterns”- Over-complicated schemas (model struggles, errors increase)
- No validation on parse (trusting without checking)
- Forcing structure when prose would be clearer
- Schemas that fight the model’s natural output style