---
title: Structured / JSON Output — Prompt Engineering Technique
description: Structured output constrains the model to emit machine-readable data — usually JSON matching a schema — so responses integrate reliably with code. Modern APIs can enforce the schema so every response parses.
canonical: https://prompeteer.ai/techniques/structured-output
category: Output control & formatting
---

# Structured / JSON Output

Structured output constrains the model to emit machine-readable data — usually JSON matching a schema — so responses integrate reliably with code. Modern APIs can enforce the schema so every response parses.

## What it is

Structured output makes the model return data in a strict, parseable format rather than free prose, so downstream code can consume it directly. The most common form is JSON that conforms to a specified schema. Beyond just asking for JSON in the prompt, current provider APIs offer schema-enforced modes (structured outputs / JSON mode / tool schemas) that constrain decoding so the result is guaranteed to be valid and to match the requested shape.

This turns the model into a reliable component of a software pipeline: field names, types, and enums are predictable, eliminating brittle parsing of natural-language responses.

## When to use

- The output feeds code, a database, or another API and must parse deterministically.
- You need specific fields, types, or an enumerated set of values.
- Free-text responses are too variable to consume programmatically.
- You are building extraction, classification, or function-calling workflows.

## How it works

1. Define the target schema: field names, types, required-ness, and allowed values.
2. Use the provider's structured-output / JSON mode (or a tool schema) to constrain decoding when available.
3. Instruct the model to return only the structured object, with no prose around it.
4. Validate the parsed result against the schema and handle any violations.

## Illustrative structure

The structure specifies a shape and asks for it exactly: "Return a JSON object with keys 'category' (one of A|B|C) and 'confidence' (0–1). Respond with JSON only." Ideally the same schema is passed to the API's structured-output mode so decoding is constrained.

## Pitfalls

- Prompt-only "please return JSON" without schema enforcement can still yield invalid or wrapped JSON.
- Over-nested or huge schemas raise error rates and token cost.
- Extra prose, markdown fences, or trailing commentary break naive parsers — demand raw output.
- Always validate; even schema-constrained output can contain semantically wrong values.

## Sources

- [OpenAI — Structured Outputs](https://platform.openai.com/docs/guides/structured-outputs)
- [Google — Gemini API: generate structured output (JSON)](https://ai.google.dev/gemini-api/docs/structured-output)

## Related techniques

- [constraint-prompting](https://prompeteer.ai/techniques/constraint-prompting)
- [system-prompts](https://prompeteer.ai/techniques/system-prompts)
- [prompt-chaining](https://prompeteer.ai/techniques/prompt-chaining)