Structured JSON output
Get valid, schema-shaped JSON from free text every time, with nulls instead of guesses, ready to feed into code or a spreadsheet.
Fill in your details
Values are inserted into every prompt below. Nothing you type leaves your browser.
3 field(s) left — unfilled variables stay as {{placeholders}}.
1Write the schema
A short JSON Schema or an example object with types works. Keep field names in snake_case and add enums wherever there's a fixed set of values; enums are the biggest single improvement to consistency.
2Run the prompt
In API use, prefer the provider's native structured-output or JSON mode with this schema and keep the prompt for the extraction rules. In chat, this prompt alone works for most cases.
Task: {{task}} Return a single JSON object that conforms to this schema: {{json_schema}} Rules: - Output only the JSON object. No prose, no markdown code fences. - Use exactly the keys in the schema. Do not add keys. - If a value is not stated in the input, use null. Do not guess or use typical values. - Strings must be copied from the input where possible; dates as YYYY-MM-DD; numbers without units or thousands separators. - For enum fields, use only the allowed values; if none fits, use null. Input: {{input_text}}3Validate
Paste the output into a JSON validator or schema checker before using it. If it fails, send the validation error back to the model and ask for a corrected object only.
Example input
A job posting: 'Senior Data Engineer at Northwind, London (hybrid, 2 days in office). £75,000-£90,000 plus bonus. Apply by 30 Nov.'
Example output
{"title":"Senior Data Engineer","company":"Northwind","location":"London","remote":"hybrid","salary_min":75000,"salary_max":90000,"currency":"GBP"}
Troubleshooting
- Output is wrapped in ```json fences: strip them in code, or repeat 'no code fences' at the end of the prompt.
- The model invents values for missing fields: make the null rule the first rule and add an example with nulls.
- Nested arrays come back inconsistent: provide one complete example object alongside the schema.
Known limitations
- Prompt-only JSON can still be invalid occasionally; validate before use, or use native structured outputs in the API.
- Very large schemas (dozens of fields) reduce accuracy; split into two passes.