How do I get structured JSON output from an AI API?

To get structured JSON output from an AI API, the usual goal is to make the model return data that matches a schema you define, instead of free-form text.

Best options

1) Use the API’s native JSON / structured output mode

If your AI provider supports it, this is the most reliable approach.

Typical patterns:

• JSON mode: forces valid JSON only
• Structured outputs / schema mode: forces JSON that matches a schema
• Function/tool calling: model returns arguments in a structured format

Example idea:

{
  "name": "Alice",
  "age": 32,
  "skills": ["python", "api", "ml"]
}

If the API supports a schema, define something like:

{
  "type": "object",
  "properties": {
    "name": { "type": "string" },
    "age": { "type": "integer" },
    "skills": {
      "type": "array",
      "items": { "type": "string" }
    }
  },
  "required": ["name", "age", "skills"],
  "additionalProperties": false
}

2) Prompt the model very explicitly

If structured mode is not available, use a strong prompt:

Example:

Return only valid JSON. Do not include markdown, explanations, or extra text.
Use this schema:
{ "name": string, "age": number, "skills": string[] }

This helps, but it is less reliable than native schema enforcement.

3) Validate and retry

Always:

• parse the output as JSON
• validate against your schema
• if invalid, retry with a correction prompt

This makes the system much more robust.

---

Practical example

Say you want extracted info from text:

Input:

“John is 29 and knows JavaScript and Go.”

Desired output:

{
  "name": "John",
  "age": 29,
  "skills": ["JavaScript", "Go"]
}

Prompt:

Extract the person's name, age, and skills.
Return only valid JSON matching:
{
  "name": "string",
  "age": "number",
  "skills": ["string"]
}

---

Tips to improve reliability

• Keep schemas simple
• Set temperature = 0 for consistency
• Specify additionalProperties: false if supported
• Ask for arrays/objects explicitly
• Don’t ask for prose and JSON in the same response
• Post-process with a JSON parser and schema validator

---

If you tell me your API/provider

I can give you a copy-paste example for:

• OpenAI
• Anthropic
• Gemini
• Azure OpenAI
• local models / Ollama
• generic REST API