Prompting
Structured Output
Forcing an LLM to return data in a specific format — usually JSON matching a schema — so downstream code can parse it reliably.
01 ——
In plain English
Structured output is the capability for an LLM to return data in a strict, parseable format (most commonly JSON, sometimes XML or a custom format). Instead of free-form text you have to regex out, you get a predictable object you can hand straight to the next step in a pipeline.
Why it matters:
- Reliability — production systems need predictable shapes
- Tool use — every function call is a structured output under the hood
- Less prompt engineering — the schema does most of the work
- Easier testing — structured outputs are easy to validate
How providers implement it:
- OpenAI — Structured Outputs (guaranteed schema compliance via constrained decoding)
- Anthropic — tool-use schema + JSON mode
- Google Gemini —
responseSchemaparameter - Open-source — Outlines, Instructor, BAML, JSONFormer enforce schemas on any model
Common patterns:
- Extracting structured fields from unstructured text (resumes, invoices, emails)
- Returning multiple candidate options as an array of objects
- Building agent action schemas (the model picks an action + fills in arguments)
Most production LLM pipelines lean heavily on structured output.
02 ——
Related terms
Function Calling
A feature that lets an AI model trigger specific functions or APIs in your app instead of just returning text.
Tool Use
The ability of an AI model to call external tools — like a calculator, search engine, or API — to help answer a question.
Prompt Engineering
The practice of crafting inputs to an AI model carefully to get better, more reliable outputs.
System Prompt
The high-level instructions given to an AI model at the start of a conversation that define its role, behaviour, and constraints.
API
An interface that lets developers send requests to an AI model and get responses programmatically — the way most AI tools talk to LLMs.
AI Agent
Software that uses an LLM to plan and act — picking tools, taking actions, and adapting based on results to complete a user’s goal.
Back to glossaryLast reviewed May 2026