How to Intermediate · 4 min read

How to use with_structured_output in LangChain

Q: How to use with_structured_output in LangChain

Use with_structured_output in LangChain to define a structured output parser that converts AI-generated text into Python objects like dicts or dataclasses. This enables reliable extraction of structured data from LLM responses by specifying output schemas and parsing rules.

Quick answer

Use with_structured_output in LangChain to define a structured output parser that converts AI-generated text into Python objects like dicts or dataclasses. This enables reliable extraction of structured data from LLM responses by specifying output schemas and parsing rules.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install langchain_openai>=0.2.0

Setup

Install the necessary LangChain package and set your OpenAI API key as an environment variable.

Install LangChain OpenAI integration:

bash

pip install langchain_openai

Step by step

This example shows how to use with_structured_output with LangChain's ChatOpenAI to parse a JSON response into a Python dictionary.

python

import os
from langchain_openai import ChatOpenAI
from langchain_core.output_parsers import with_structured_output

# Define a simple output parser that expects a JSON object with specific keys
output_parser = with_structured_output(
    output_schema={
        "name": str,
        "age": int,
        "email": str
    }
)

# Initialize the chat model
chat = ChatOpenAI(model="gpt-4o-mini", temperature=0, api_key=os.environ["OPENAI_API_KEY"])

# Define the prompt
prompt = """
Provide the following user info as a JSON object with keys: name, age, email.
"""

# Generate the response
response = # Use chat.invoke([HumanMessage(content='...')]) for LangChain ChatOpenAI
    messages=[{"role": "user", "content": prompt}]
)

# Extract raw text
raw_text = response.choices[0].message.content

# Parse the structured output
parsed_output = output_parser.parse(raw_text)

print("Raw response:", raw_text)
print("Parsed output:", parsed_output)

output

Raw response: {"name": "Alice", "age": 30, "email": "alice@example.com"}
Parsed output: {'name': 'Alice', 'age': 30, 'email': 'alice@example.com'}

Common variations

You can use with_structured_output with different output schemas such as nested dicts or dataclasses. It also supports async calls and other LLM models supported by LangChain.

Use dataclasses for output schema.
Call chat.invoke() with async for async usage.
Switch models by changing model parameter.

python

from dataclasses import dataclass
from langchain_core.output_parsers import with_structured_output

@dataclass
class UserInfo:
    name: str
    age: int
    email: str

output_parser = with_structured_output(output_schema=UserInfo)

# Use the same chat client and prompt as before
# After getting raw_text from the model:
parsed_output = output_parser.parse(raw_text)
print(parsed_output)  # UserInfo(name='Alice', age=30, email='alice@example.com')

output

UserInfo(name='Alice', age=30, email='alice@example.com')

Troubleshooting

If parsing fails with a ValueError, ensure the model output matches the expected schema format exactly.
Use temperature=0 to reduce randomness and improve structured output consistency.
Validate the raw output before parsing to handle unexpected text or formatting.

✅

Key Takeaways

Use with_structured_output to define strict output schemas for reliable data extraction from LLM responses.
Set temperature=0 in your chat model to improve consistency of structured outputs.
You can parse outputs into Python dicts or dataclasses for easy downstream processing.

Verified 2026-04 · gpt-4o-mini

Verify ↗