Structured Output

What it means

The product constrains generation to fields, types, and enums (often via schema validation) so downstream code and design systems can render reliably.

Why designers should care

Design explicit field layouts, validation errors, and partial-fill states. Users should edit structured AI output like a form, not a chat bubble only.

Example

A user research importer returns JSON with quote, theme, and severity; the UI maps each to editable table rows before syncing to Notion.

Common mistakes

• • Parsing fragile markdown tables from chat when schema output was available.
• • No UI for invalid or missing required fields after generation.
• • Letting users choose free-form when the backend requires strict JSON.

Related terms

• JSON: JSON is the most common interchange format for structured model output: field names, typed values, and arrays the product can validate and render predictably.
• Schema: A schema is the contract between prompt, model output, and interface: which fields exist, what they mean, and what “valid” looks like.
• Prompt: Prompts set role, task, constraints, format, and context; the model’s reply is a direct response to that combined instruction stack.
• Tool Use: The model emits structured calls (name + arguments) that your app executes, then feeds results back into the conversation for the next step.
• Workflow: Defined stages (inputs → generate → review → export) often spanning people, systems, and multiple model calls with clear deliverables.