LLM API with JSON output

If you want an LLM API to return JSON-only output, there are a few common approaches depending on the provider.

General options

1. Prompt for JSON

Ask the model explicitly:

• “Return valid JSON only.”
• “Do not include markdown, explanations, or code fences.”

Example prompt:

Extract the following fields and return valid JSON only:
- name
- email
- phone

2. Use structured output / JSON mode

Some APIs support a built-in JSON response mode or schema enforcement.

Examples:

• OpenAI: structured outputs / JSON schema
• Anthropic: tool use / structured formatting
• Google Gemini: JSON output mode
• Azure OpenAI: similar capabilities depending on model/deployment

3. Use function calling / tool calling

Instead of free-form text, define a function schema and have the model fill it. This is often the most reliable way to get structured data.

Example schema:

{
  "type": "object",
  "properties": {
    "name": { "type": "string" },
    "email": { "type": "string" }
  },
  "required": ["name", "email"],
  "additionalProperties": false
}

Practical recommendation

If your API supports it, use:

1. JSON schema / structured outputs for strict validation
2. Function calling if you need the output to map directly into an application
3. A fallback prompt-only JSON instruction if the model doesn’t support structured mode

Important

Even with JSON mode, you should still:

• validate the response
• handle malformed or incomplete JSON
• retry if needed

If you want, I can give you:

• an OpenAI API example
• a Python example
• a Node.js example
• or a provider comparison for JSON output