Advanced Prompt Engineering / Output Control & Format

Structured Output

Intermediate [3/5]

JSON generation Schema-constrained output

Definition

Structured output constrains models to generate data in specific formats like JSON or XML. Many API providers offer "JSON mode" or schema enforcement that guarantees syntactically valid structured data.

This is essential for applications where LLM outputs feed into downstream systems that require parseable, typed data.

Key Concepts

JSON mode: API setting that forces valid JSON output
Schema enforcement: Output must match a defined structure
Type safety: Fields have expected types (string, number, array)
Parsing reliability: Output can always be parsed without errors

Examples

JSON Mode

API Configuration

response = client.chat.completions.create(
    model="gpt-4",
    response_format={"type": "json_object"},
    messages=[{
        "role": "user",
        "content": "Extract entities from: 'John lives in NYC'"
    }]
)
# Guaranteed valid JSON:
# {"name": "John", "location": "NYC"}

Schema Definition

Type Enforcement

Expected schema:
{
  "product_name": string,
  "price": number,
  "in_stock": boolean,
  "categories": string[]
}

Model output will match this structure exactly.

Interactive Exercise

✎

Design a Schema

Design a JSON schema for extracting information from job postings. Include fields for:

- Job title, company, location, salary range, required skills

Pro Tips

Always use JSON mode when you need to parse the output
Define optional vs required fields in your schema
Include type information in prompt even when using JSON mode
Have fallback handling for edge cases

Definition

Key Concepts

Examples

Interactive Exercise

Related Terms