Structured output
Asking an AI to answer in a fixed, machine-readable shape, like named fields, instead of free text.
What it is
Structured output means the AI returns data in an exact format, usually JSON with named fields: { "category": "Billing", "urgency": "high" }. The app can then save it, sort by it, or show it in specific places.
Free text is fine for a human to read but fragile for software to use. If an AI feature feeds into the app (tagging, extracting, scoring), ask for structured output and list the fields and allowed values.
It also reduces rambling, because the model has a clear shape to fill.
How to ask for it
“Have AI read the receipt and get the info.”
“Use structured output to extract from each receipt: merchant (text), date (YYYY-MM-DD), total (number), currency (3-letter code), and category (one of Travel, Meals, Software, Other). Leave a field empty if it can't be read.”
"Get the info" returns unpredictable text; structured output with named, typed fields can be saved directly.
You've seen this in
- RReceipt scanners pulling out merchant, date, and total
- EEmail triage tools tagging category and priority
- RResume parsers