Structured output

Structured output is asking an AI to return its answer in a fixed shape, such as a table, a list, or labeled fields, so the result is ready to use or pass straight into the next step.

Structured output is asking an AI to return its answer in a fixed shape, such as a table, a list, or labeled fields, so the result is ready to use or pass straight into the next step. Instead of accepting a paragraph and rereading it, you tell the agent the exact format you want and it fills that format in. Say you ask an assistant to read five meeting notes and pull out follow-ups. A free-form reply gives you sentences you still have to sort. Structured output gives you back rows with an owner, a task, and a due date in each one, so a calendar or a to-do app can read them without a person retyping anything.

Builder example

When one step's output feeds another step automatically, free-form prose breaks the chain because the next step cannot reliably find the piece it needs. If your agent extracts an order and the reply says "the customer wants two of item 5512 shipped Friday," downstream code has to guess where the quantity ends and the date begins. Tell the model to return labeled fields for quantity, item, and ship date, give it one filled-in example, and the next step reads each value by name instead of parsing a sentence.

Common confusion: Asking for a format in plain words helps, but it does not guarantee the model returns valid, parseable data every time. What makes output dependable is constraining the response to a declared schema and validating it, so a stray sentence or a missing field is caught before the next step runs on bad input.