结构化输出

A structured extraction prompt for clean JSON output from unstructured text, notes, or documents.

審校時間: 2026-03-18

Use this when you need reliable machine-readable output from free-form text and you want the model to stay inside a strict schema.

Prompt 模板

Extract the information below and return valid JSON only.

Schema:
{schema}

Rules:
- If a field is missing, use null.
- Keep original wording where possible.
- Do not invent facts.
- Return one JSON object and nothing else.

Input:
{text}

正式使用前，請先替換模板變數並補充你的限制條件。

適用場景

Extracting entities from articles or notes
Turning meeting notes into action items
Normalizing messy text before downstream automation

期望輸出

Valid JSON only
Null for missing fields
No hallucinated facts
Stable schema matching

可替換變數

schema

Target JSON schema or shape

Example: { customer, issue_type, due_date, action_items[] }

text

Source content to parse

Example: Meeting notes, email thread, article

優化建議

Put the schema directly in the prompt, not just in your head.
Tell the model how to handle missing values.
If output validity matters, validate it after generation and retry with the error message.

示例用法

Extract customer name, blocker, owner, due date, and actions from a long support escalation email.

預期效果

The model returns a clean JSON object that can be stored or processed by code without manual cleanup.