Structured Outputs

Software likes tidy data, and people write messy text. Generative A I is superb at turning one into the other. In this lesson, you will learn how to get reliable, structured results, that your programs can trust.

• Messy text in, tidy data out
• Reliable formats for programs
• Schemas and validation

Extracting data as JSON

Here we ask the model to read a customer message, and extract the details. We tell it to reply with J S O N only, with exact fields. Then we parse the answer with json loads, and use it like any Python dictionary. With a low temperature, the results are consistent. This simple pattern powers many real products.

import json
from openai import OpenAI
client = OpenAI()

msg = "Hi, I'm Sara. My order #4821 arrived broken. I want a refund."
prompt = f"""Extract JSON with keys name, order_id, problem, wants.
Reply with JSON only.\n\nMessage: {msg}"""

reply = client.chat.completions.create(
    model="gpt-4o-mini", temperature=0,
    messages=[{"role": "user", "content": prompt}])
data = json.loads(reply.choices[0].message.content)
print(data["order_id"], data["wants"])

Why validation matters

But models can still slip. A field may be missing, a number may arrive as text, or the model may add a friendly sentence before the J S O N. So never trust blindly. Define the shape you expect, with a schema, and validate every answer. If it fails the check, retry, or ask the model to fix its own output. Most providers also offer a structured output mode, that guarantees valid J S O N.

Where structure helps

Structured output unlocks many uses. Sorting support tickets into categories. Extracting names, dates, and amounts from invoices. Turning meeting notes into action items. Labelling thousands of reviews, as positive, or negative. Filling a database, from free text. Anywhere you have messy words, and need clean fields, this pattern applies.

Summary