At a glance
| Aspect | JSON | XML |
|---|---|---|
| Syntax | Braces / brackets / colons | Nested tags |
| Size | Compact | Verbose |
| Data types | String, number, bool, null, array, object | All values are strings |
| Comments | Not allowed (JSON5 allows) | Yes |
| Schema | JSON Schema | XSD, DTD |
| Namespaces | No | Yes (xmlns) |
| Attributes | No (everything is a key) | Yes |
| Mixed content | Awkward | Natural |
| Transformations | Ad-hoc | XSLT, XPath |
| Tooling | Ubiquitous — every language | Heavy but mature |
| Readability | Higher (usually) | Lower at scale |
When to pick which
- JSON: web APIs, config, mobile, most new systems.
- XML: document-centric data (Office OOXML, DocBook, SVG), SAML, SOAP, finance (FIXML, XBRL), healthcare (HL7), government mandates.
- YAML (3rd option): human-editable config (K8s, CI). More readable than JSON, pickier parser.
- Protocol Buffers / Avro: binary formats — smaller and faster than either, at the cost of requiring schemas for decode.
Worked example: the same record
One person record in JSON: {"name":"Ada","age":36}, which is 23 bytes. The same record in XML: <person><name>Ada</name><age>36</age></person>, which is 46 bytes, twice as large, because every value carries an opening and closing tag. Compression such as gzip narrows the gap a great deal, since the repeated tags compress well, so measure the compressed size before deciding that size matters.
The types also differ. In the JSON above, 36 is a number and a parser returns it as one. In XML, 36 is text until a schema or your code says otherwise. On the other hand, JSON numbers have no defined precision: many parsers read them as 64-bit floating point, which represents whole numbers exactly only up to 253 (9,007,199,254,740,992). Larger identifiers, such as some database IDs, should be sent as strings.
Common mistakes
- Adding comments to JSON. The standard (RFC 8259) has none; a parser will reject them. Use a field, or JSON5 or YAML for configuration files.
- Assuming key order. JSON objects are unordered by definition; do not rely on the order of keys.
- Wrong encoding. RFC 8259 requires JSON exchanged between systems to be UTF-8. Legacy encodings cause mojibake.
- Parsing untrusted XML with defaults. Some XML parsers resolve external entities, which can leak files or cause denial of service; disable DTD and entity expansion.
- Converting blindly. XML attributes, repeated elements and mixed text have no single JSON form, so a round trip can lose information.
Notes
- XML is better at mixed content (e.g., "
Hello world
"). JSON needs convention (e.g., objects with "_text" and "children"). - JSON parsing is typically 2–10× faster than XML at equivalent data size.