Common JSON Errors and How to Fix Them

October 30, 2026 · 5 min read

Most invalid JSON comes from a handful of mistakes: a trailing comma, single quotes instead of double quotes, unquoted keys, comments, or a value JSON does not support such as NaN or undefined. Paste the text into a JSON validator to get the exact line and column, then apply the matching fix below.

JSON's grammar (RFC 8259) is deliberately small. That is why it is easy to break: anything you are used to from JavaScript object literals, Python dicts or YAML that is not in the spec is an error.

Quick reference

What the error messages mean

Parsers report the position where they gave up, which is often just after the real mistake. Messages differ by engine. In current Chrome and Node.js, JSON.parse errors look like Unexpected token '}', ..."a": 1,}" is not valid JSON; Firefox says JSON.parse: expected double-quoted property name at line 1 column 10; Python raises json.decoder.JSONDecodeError: Expecting property name enclosed in double quotes: line 1 column 10 (char 9).

The fixes, one by one

Trailing commas

JavaScript, Python and most config formats allow a comma after the last element; JSON does not. They usually appear after deleting the last item of a list by hand. Remove the comma before every } and ].

Single quotes and unquoted keys

Strings and keys must use double quotes. This often happens when someone prints a Python dict with print(d) or str(d) instead of json.dumps(d), or copies a JavaScript object literal. Fix it at the source by serializing properly:

import json

data = {"active": True, "owner": None}
print(str(data))         # {'active': True, 'owner': None}  <- not JSON
print(json.dumps(data))  # {"active": true, "owner": null}

Comments

JSON has no comments. If your file is a config that supports them (VS Code's settings.json, tsconfig.json), it is JSONC, not JSON, and a strict parser will reject it. Strip the comments or parse with a JSONC-aware library. See JSON vs JSON5 vs JSONC for which tools accept what.

NaN, Infinity and undefined

JSON numbers must be finite. In JavaScript, JSON.stringify already converts NaN and Infinity to null and drops undefined object properties, so if you see them in output, something built the string by hand. Python's json.dumps writes NaN by default; pass allow_nan=False to raise an error instead, and replace the values with null first.

Escapes inside strings

Only these escapes are valid: \", \\, \/, \b, \f, \n, \r, \t and \uXXXX. Windows paths are the classic trap: "C:\new" is valid but means a newline, while "C:\data" is invalid because \d is not an escape. Write "C:\\data", or use forward slashes.

Numbers

No leading zeros (007), no leading +, no hex, no trailing decimal point (1.). Values with meaningful leading zeros, such as ZIP codes and IDs, should be strings anyway. Very large integers are valid JSON but JavaScript's JSON.parse silently loses precision above 2^53 - 1 (9007199254740991), so send big IDs as strings.

Unexpected end of input and HTML responses

If parsing an API response fails at position 0 with <, log the raw body: you are getting HTML. If it fails at the end, the response was truncated or empty. Check the status code and Content-Type header before calling response.json():

const res = await fetch(url);
const text = await res.text();
if (!res.ok || !res.headers.get("content-type")?.includes("json")) {
  throw new Error("Expected JSON, got " + res.status + ": " + text.slice(0, 200));
}
const data = JSON.parse(text);

Find the error fast

1. Paste the text into the validator. It highlights the line and column and explains the problem in plain words.
2. Look a few characters before the reported position, especially for trailing commas and missing commas.
3. For many small mistakes at once, use the JSON fixer. It repairs common mistakes such as trailing commas, quotes and comments and shows a diff, so you can review each change.
4. Once it parses, run the formatter so the structure is readable and mistakes are easier to spot next time.

If the broken JSON came from ChatGPT or another model, there are extra failure modes such as Markdown fences and truncated output; see how to fix broken JSON from LLMs.