What is JSON?

What JSON is

JSON (JavaScript Object Notation) is a lightweight, text-based format for representing structured data, objects, lists, and the simple values inside them. It was derived from JavaScript’s object syntax but is now language-independent: every mainstream language can read and write it. When one program sends data to another, over an API, in a config file, in a log line, the data is very often JSON. It won because it is human-readable and maps almost one-to-one onto the data structures (objects/dicts and arrays/lists) that code already uses.

The syntax, at a glance

JSON is built from key/value pairs inside objects, and ordered lists inside arrays. Keys are always double-quoted strings; values are one of the six types below.

{
  "name": "Ada Lovelace",
  "born": 1815,
  "active": true,
  "fields": ["mathematics", "computing"],
  "collaborator": { "name": "Charles Babbage" },
  "spouse": null
}

Two rules trip everyone up at least once: keys and string values use double quotes (never single), and there is no trailing comma after the last item. JSON also has no comments.

The six data types

Notably absent: no date type (dates travel as strings, usually ISO 8601 like "2026-06-19") and no undefined. Anything richer is encoded on top of these six.

Parsing & serializing

JSON is a string until you parse it into a live object, and you serialize an object back into a string to send it. The two operations are mirror images, and look nearly identical across languages:

// JavaScript: text <-> object
const obj = JSON.parse(text);      // string  -> object
const text = JSON.stringify(obj);  // object  -> string

// Python: identical idea, different names
import json
obj  = json.loads(text)            # string -> dict
text = json.dumps(obj)             # dict   -> string

Need a typed shape from a sample? The JSON → TypeScript tool turns any JSON into an interface; CSV → JSON and YAML → JSON convert the neighbors.

JSON vs XML

For a full breakdown of the document side, see what XML is.

JSON & AI

Model APIs speak JSON: your request, the model’s response, and every tool call are JSON payloads. The big reliability win is structured outputs, where you give the model a schema and it is constrained to emit JSON that matches, so your code can JSON.parse the answer instead of scraping it out of prose. Build one visually with the Structured Output Builder.

Common mistakes

• Trailing comma after the last array/object item. Valid in JavaScript, invalid in JSON.
• Single quotes or unquoted keys. JSON requires double-quoted keys and strings.
• Comments. JSON has none; strip them (or use JSON5/JSONC, which are different formats).
• Sending a date or function as-is. Encode dates as strings; functions can’t be represented.
• Confusing the string with the object. Parse before you index into it, stringify before you send it.

FAQ