Structured Output
A mode where the model is constrained to return responses in a specific format — typically JSON — that matches a defined schema.
Structured Output (sometimes called JSON mode or constrained generation) forces the model to produce responses that conform to a predefined schema. Instead of returning free-form text, the model returns a JSON object (or other structured format) with exactly the fields specified.
Why it matters for production applications: - Downstream code can parse the response reliably without regex or post-processing hacks - Schema validation can be applied before the response is used - Reduces errors from models inventing field names or omitting required data
Implementation approaches: - JSON mode: the model is instructed or constrained to output valid JSON - Schema-constrained generation: the model's output is constrained token-by-token to fit a JSON Schema - Tool/function calling: the model calls a function with typed arguments, which is returned as structured data
Most production AI applications use structured output for any data extraction, classification, or decision-making task where the result needs to be consumed programmatically.
Example
A contract analysis tool asks Claude to extract key dates, parties, and obligations from a PDF. Structured Output ensures the result is always a JSON object with consistent field names, ready to insert into a database.
Related terms
Tool Use / Function Calling
Enabling an AI to call external functions or APIs to perform tasks beyond text generation.
Prompt
The text instruction you send to an AI model asking it to do something.
System Prompt
A special instruction given to an AI model that defines its behavior and personality for all subsequent user interactions.