JSON

JSON is how AI agents organize information so computers can understand it perfectly. Think of it as a digital filing cabinet with labeled folders.

What

JSON = Structured Data Format

Like a form with labeled fields, JSON organizes information so both humans and computers can read it easily.

Real Example:

{
  "customer": "John Smith",
  "order": 12345,
  "amount": 299.99,
  "status": "shipped"
}

Same info as a messy email:

“Hi, John Smith ordered something, order number 12345 I think, cost around $300, already shipped”

Why JSON wins: Every piece of information has a clear label and exact value.

Business

JSON solves your data chaos problems:

Consistent

Before: “Customer name somewhere in email” After: "customer": "John Smith" Result: Always in the same place

Accurate

Before: “Around $300” After: "amount": 299.99 Result: Exact numbers, no guessing

Processable

Before: Human reads and types into system After: Agent fills forms automatically Result: Zero data entry errors

Examples

Customer Data

{
  "name": "Sarah Johnson",
  "email": "sarah@company.com",
  "phone": "(555) 123-4567",
  "company": "ABC Corp",
  "priority": "high"
}

Invoice Information

{
  "vendor": "Office Supplies Inc",
  "invoice_number": "INV-2024-001",
  "amount": 1250.00,
  "due_date": "2024-02-15",
  "status": "pending"
}

Lead Scoring

{
  "lead_name": "Tech Solutions LLC",
  "score": 85,
  "budget": "50000-100000",
  "timeline": "Q2 2024",
  "decision_maker": true
}

Templates

Copy these templates for your agents:

Email Processing

Extract information from this email and format as JSON:
{
  "sender": "[person's name]",
  "company": "[their company]",
  "subject": "[email subject]",
  "request_type": "[support/sales/billing]",
  "priority": "[high/medium/low]",
  "requires_response": "[yes/no]"
}

Document Analysis

Read this document and extract key information as JSON:
{
  "document_type": "[invoice/contract/report]",
  "date": "[document date]",
  "amount": "[dollar amount if applicable]",
  "parties": ["[list of companies/people involved]"],
  "key_terms": ["[important details]"],
  "action_required": "[what needs to happen next]"
}

Data Entry

Convert this information to structured JSON format:
{
  "field1": "[extracted value]",
  "field2": "[extracted value]",
  "field3": "[extracted value]",
  "confidence": "[how sure you are 0-100]",
  "notes": "[any important details]"
}

Rules

JSON Rules (Don’t Worry, They’re Simple):

1. Use quotes around text: "customer": "John Smith"
2. No quotes around numbers: "amount": 299.99
3. Separate items with commas: "name": "John", "age": 25
4. Wrap everything in curly braces: { ... }
5. Use square brackets for lists: "items": ["pen", "paper", "stapler"]

Agent Instructions:

Always return valid JSON format. If any field is unclear,
mark as "unknown" rather than guessing. Include confidence
score for each extracted field.

Benefits

Why Your Business Needs JSON:

1. Eliminates Data Entry Errors - Information goes exactly where it belongs
2. Enables Automation - Systems can process structured data automatically
3. Improves Consistency - Same format every time, no variations
4. Speeds Up Processing - No hunting through paragraphs for key details
5. Scales Effortlessly - Handle 10 or 10,000 records the same way

Common

“This looks complicated…” It’s simpler than spreadsheets. Just labeled boxes for information.

“Do I need to write JSON myself?” No. Tell your agent what information you need, and it creates the JSON.

“What if the agent makes mistakes?” That’s why templates include confidence scores and “unknown” options.

“How is this different from spreadsheets?” Spreadsheets store data. JSON organizes data so agents can understand and use it.

Start

Right now: Pick one business process that involves extracting information from emails, documents, or forms.

Use this template:

Extract the following information as JSON:
{
  "[what you need]": "[where it should go]",
  "[next thing you need]": "[where that goes]",
  "confidence": "[how sure the agent is]"
}

Next

Related resources:

• Markdown - Format readable outputs
• Prompting - Master instruction-writing
• Data - Organize data for agents