A model's answer is only useful if the next step can read it. When the output feeds a database, an API, or another program, a friendly paragraph of text is not enough. Structured outputs make the model reply in a predictable shape, such as JSON with the exact fields your code expects.
You'll start with formatting in the prompt, where you describe the structure you want and use JSON to keep data clear and compact. Then you'll learn how schemas and Pydantic models turn the reply into typed objects, and how validation and retries catch malformed or incorrect answers before they reach the rest of your system.
Once your outputs are reliable, you can start tuning how the model generates them. That's the focus of Model Parameters & Selection, the next module in this course.
How to use JSON in your prompts so the data you send to a language model stays precise and cost-efficient.
How Pydantic models and function calling turn model replies into typed, validated objects instead of fragile JSON strings.
A follow-up on why response schemas and validation still beat parsing raw JSON, including retrying when the output fails a check.