Structured Output
Structured output means constraining a language model to return a fixed format, such as JSON that matches a schema, so that software can use the result reliably instead of parsing free text.
Also known as: structured outputs
Language models write free text by default, which is fine for a person and fragile for a program. If an application needs to pull a customer name, a date and an amount out of an email, it needs those fields in the same shape every time. Structured output asks the model for that shape, typically JSON with named fields, and many model APIs can constrain the response to a schema you supply. A looser “JSON mode” only guarantees valid JSON, not your fields, and even enforced output needs handling for refusals and responses cut off at a length limit.
It matters most when AI output feeds other software: extracting data from documents, classifying tickets, or deciding which tool an agent should call and with what arguments. A valid format doesn’t make the content correct, so teams still check the values. How AI agents use tools shows the same idea at work in tool calls.
