CSV to JSON

What the parser keeps inside a cell

The first row is the header. Each header becomes a property name, and each later row becomes one object. Headers must be non-empty after you look at them, and they must be unique. A repeated header stops with Headers must be non-empty and unique. A file that is only a header, with no data row, stops with Need at least a header row and one data row. A data row whose field count differs from the header stops with a message that names the file row, counting the header as row 1. Row 2 has 1 fields; expected 2. is the usual shape. The page does not pad a short row and does not discard an extra field.

Quotes follow a small CSV rule. A quote opens a cell only at the start of that cell. Inside the quotes, a delimiter and a line break are data. Two quotes in a row are one quote character. A quote that never closes stops with Unclosed quoted CSV field. A quote that appears after the cell already closed, or a stray quote in an unquoted cell, stops with Malformed CSV quoting. Quote entire cells and escape quotes as two quotes. A leading byte-order mark is removed before the parse, so a file saved with a BOM can still use its first header. Whitespace inside a cell is kept. The page does not trim " Ada" down to Ada.

Input delimiter, the preview, and type inference

Auto-detect tries comma, semicolon, tab, and pipe against the first 20 parsed rows. A candidate scores its width when that width is greater than 1 and every tested row has the same width. The highest score wins. A tie keeps the earlier candidate, and comma is first, so a file that is only one column, such as only then line, resolves to comma. The preview prints JSON.stringify of the chosen character. The sample's preview is ";" including the quote marks. An explicit choice skips detection and the preview shows that character the same way, so a tab appears as "\t". If the text cannot be parsed for the preview, the line says invalid input. The preview does not convert. You still press Convert to JSON.

Infer numbers, booleans and null starts off. Off means every cell is a JSON string, including 30, true, and null. On, the cell must match exactly. null becomes JSON null. true and false become booleans. A numeric pattern without a leading zero, optional fraction, and optional exponent, within the safe integer range, becomes a number. 2.5 becomes 2.5. 1e2 becomes 100. 001 stays the string 001 because of the leading zero. A space before a digit, as in the two characters space and 1, stays a string. 9007199254740993 stays a string because it is past the safe integer limit. NULL in capitals stays a string, because only the exact letters null are special. The CSV delimiter converter changes the separator and does not build objects. This page builds the objects and can leave the separator decision to Auto-detect.

When JSON records are the form you want

Use this page when a spreadsheet export should become an array you can paste into a test, a fixture, or another converter. The sample is the case that usually goes wrong in a naive split: the semicolon inside Hello; world must not start a new field, and Bob's note must keep its line break. Both survive. The download is csv-to-json.json. The input cap is 1,000,000 characters, with the same too-large message the other data tools use. An empty box says Enter an input first.

Going back to a sheet is JSON to CSV, which expects an array of objects and can flatten nested objects this page will never create, because a CSV cell does not become a nested object here. An INSERT for PostgreSQL, MySQL, or SQLite is CSV to SQL, which shares the delimiter control and the inference checkbox but writes SQL instead of JSON. The guide Convert nested JSON to CSV without losing fields is the other direction's walkthrough. This page is the check you do before that, when the file you hold is still CSV and you want to see the records.

Leading zeros, duplicate headers, and a quote left open

Leave inference off for anything that only looks numeric. A postal code, a product id, and a phone number are the usual casualties. 001 with inference on is still the string 001, which is the safe case, but 10 stays a number and a later zero-padded cousin in another file might not match it. If you turn inference on, read the JSON before you download it. true must be the four letters in lowercase. True stays a string. The word null with inference on becomes a JSON null, which is not the same as an empty cell. An empty cell stays the empty string even when inference is on, unless you are on the SQL tool and have asked for NULL.

A header is a key, taken literally. Two columns both named id are an error, not a silent overwrite. A blank header is an error. A header whose text is __proto__ becomes an own property of the record rather than a change to the object's prototype. The row-width check uses the header's length, so a trailing delimiter makes an extra empty field and will fail if the header did not have that extra column. An unclosed quote often comes from a note that contains a line break and lost its closing quote in the export. The error is Unclosed quoted CSV field. Fix the quote in the box and convert again. The previous JSON is cleared when the CSV text changes, so a failed click does not leave the last good array pretending to match the broken file.

Rules and limitations

The first row is a unique, non-empty header. Every data row has the same field count. Quoted cells may contain the delimiter and line breaks. A doubled quote is one quote. A BOM is ignored. Cells are not trimmed.

Auto-detect picks comma, semicolon, tab, or pipe from the first 20 rows, preferring a consistent width above 1, with comma winning a tie. Inference is off unless checked, and it only converts exact null, true, false, and safe numbers without leading zeros. The download file is csv-to-json.json.

Common problems

If a semicolon file became one column, the preview probably shows ",". Pick Semicolon and press Convert to JSON again. Auto-detect can prefer comma when both separators look rectangular.

If an id lost its leading zeros, inference is on and the value matched the number pattern. Uncheck it. If the error names a row width, a quote broke or a delimiter was left inside an unquoted cell.

Frequently asked questions

About this tool

Maintained by Toolbench. For large or deeply nested files, split the input into smaller batches and verify representative results before using them in an import or application. Report reproducible problems through Contact.