Structured Output

You might say

AI gives me a big block of text I can't use directly. Can it return the answer in a fixed format?

Model output returned in an agreed, machine-checkable shapeTo fill a calendar form from an email, a product can require fields such as title, date, and location with defined types. JSON is a common encoding, not the same thing as structured output, and business rules still need validation.

Know first

JSON API

Turn an event email into usable calendar fields

Invitation email

Product review next Wednesday at 3:00 PM in A301. Alex is the owner.

→

Model returns JSON

{
 "date": "next Wednesday",
 "time": "15:00",
 "room": "A301"
}

→

Program validationWaiting for validation

How text becomes data a program can check

Define the shape: Specify fields such as title, date, and location before asking the model to extract them.

Return the agreed fields: The model returns data in the required structure rather than a free-form summary.

Validate before use: The product checks the schema and real business rules before filling a form or creating an event.

You can say this to an AI Agent

Make task extraction return a fixed shape: title is required text; dueDate is YYYY-MM-DD or null; owner is text or null; status is one of todo, doing, done. Reject a missing title, invalid date, or unknown status before use; keep an absent due date as null rather than guessing. Test complete, no-date, and invalid-status inputs.

Learn next

1Tool Calling→2Token→3Validation→

Recommended tool

Web fine-tuning tool · Selector Pro

Select page elements for precise AI edits

Click a button, text, or card, note the change, then copy its exact location to AI.

Open Selector Pro website

A page must render model results as task cards. Which return format is most reliable?

A. Ask for prose and guess title and due date with string rulesB. Ask the model to return full task-card HTMLC. Request structured output matching a task-field schema and validate required fields