Structured data extractor

When to use

A person wrote the input, and your code needs fields that match a schema you already chose.

Prompt

You extract a JSON object from messy user text.

Return only JSON that matches this schema. Do not add keys. Do not add commentary.

Schema:
{{schema}}

Rules:

- Copy facts from the input. Do not infer ids, emails, dates, or amounts that are not stated.
- If a value is absent, use null.
- Normalize only the fields the schema describes.

Input:
{{input}}

Variables

• {{schema}}
• {{input}}

Example

Can you move Priya's demo to Thursday at 2 and tell Alex? It's about the annual plan.

{
  "attendees": ["Priya", "Alex"],
  "day": "Thursday",
  "time": "14:00",
  "topic": "annual plan",
  "timezone": null
}

Why it works

The schema is the contract. Nulls make missing facts visible instead of letting the model invent them.

Common mistakes

• Listing no schema and asking for JSON in general.
• Forgetting to reject keys that are not in the schema.
• Skipping an eval once a few real messages parse successfully.

Related concepts

• What LLMs are good atStart from a job a language model can do, not from a feature name.
• Structured dataTurn unstructured text into fields your program can use.
• Prompting fundamentalsA prompt is a job description: role, task, context, constraints, examples, and output shape.
• Structured outputsRequire the model to produce data that matches a known structure.

Build this

• Build an AI Product Brief GeneratorTurn a product idea into a validated brief and save it to Neon.

Related prompts

• Entity extractorList the people, organizations, products, and dates a document actually names.
• JSON output systemInstructions that keep a model on a schema and silent about everything else.