How to Intermediate · 4 min read

How to extract structured data from unstructured text with AI

Q: How to extract structured data from unstructured text with AI

Use a large language model like gpt-4o to parse unstructured text by prompting it to identify and format key data fields into structured JSON or CSV. This involves crafting clear instructions and examples in the prompt to guide the model's output format.

Quick answer

Use a large language model like gpt-4o to parse unstructured text by prompting it to identify and format key data fields into structured JSON or CSV. This involves crafting clear instructions and examples in the prompt to guide the model's output format.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the OpenAI Python SDK and set your API key as an environment variable to authenticate requests.

bash

pip install openai>=1.0

Step by step

This example shows how to extract structured data such as names, dates, and amounts from unstructured text by prompting gpt-4o to output JSON.

python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

unstructured_text = '''\nInvoice from Acme Corp\nDate: March 15, 2026\nTotal: $1,234.56\nCustomer: John Doe\n'''

prompt = f"Extract the invoice data as JSON with keys: company, date, total, customer.\nText:\n{unstructured_text}\nJSON:" 

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": prompt}]
)

structured_data = response.choices[0].message.content
print(structured_data)

output

{
  "company": "Acme Corp",
  "date": "March 15, 2026",
  "total": "$1,234.56",
  "customer": "John Doe"
}

Common variations

You can use other models like claude-3-5-haiku-20241022 or gemini-1.5-pro for extraction. Async calls and streaming responses are also supported for large texts or real-time processing.

python

import os
import anthropic

client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

unstructured_text = '''\nInvoice from Acme Corp\nDate: March 15, 2026\nTotal: $1,234.56\nCustomer: John Doe\n'''

system_prompt = "You extract invoice data as JSON with keys: company, date, total, customer."
user_prompt = f"Text:\n{unstructured_text}\nJSON:" 

message = client.messages.create(
    model="claude-3-5-haiku-20241022",
    max_tokens=256,
    system=system_prompt,
    messages=[{"role": "user", "content": user_prompt}]
)

print(message.content[0].text)

output

{
  "company": "Acme Corp",
  "date": "March 15, 2026",
  "total": "$1,234.56",
  "customer": "John Doe"
}

Troubleshooting

If the output is not valid JSON, add explicit instructions in the prompt to "only output JSON" and consider using a JSON schema or examples.
If the model misses fields, provide more detailed examples or increase the prompt context.
For very long texts, chunk the input and aggregate results.

✅

Key Takeaways

Use clear, explicit prompts instructing the model to output structured JSON for reliable extraction.
Choose models like gpt-4o or claude-3-5-haiku-20241022 for best accuracy in data extraction tasks.
Chunk large unstructured texts to maintain context and improve extraction quality.
Validate and parse the model output programmatically to handle any formatting inconsistencies.
Provide examples in prompts to guide the model toward the desired structured format.

Verified 2026-04 · gpt-4o, claude-3-5-haiku-20241022, gemini-1.5-pro

Verify ↗