CSV · TSV · PIPE · SEMICOLON → JSON

CSV to JSON: one paste.

Paste a spreadsheet export, an API mock, or a test fixture and Rowson hands back clean JSON in four shapes, including JSONL for LLM pipelines. Delimiters are detected for you, and every byte is converted right here in your browser.

Start converting

Input · CSV

Output · JSON

Output will appear here

Paste or drop CSV to begin Processed locally, data never leaves the browser

Reference

CSV dialect and JSON output reference

“CSV” is not one format but a cluster of incompatible dialects, and the quiet failure point is escape rules: a field holding a literal quote survives the parser that wrote it, then breaks under another dialect’s parser. This reference lays out which dialects exist, their quoting conventions, what each of the four JSON output shapes is good for, and which libraries implement which behavior.

CSV dialects, separators, and quoting

Dialect	Field separator	Quote char	Escape rule	Defined by
RFC 4180	,	"	double the quote: ""	IETF RFC 4180 (2005)
Excel CSV	, (locale-dependent)	"	double the quote	de facto, Microsoft Office
TSV (IANA)	\t	optional "	backslash or doubled quote	IANA text/tab-separated-values
TSV (Unix)	\t	none	tabs and newlines forbidden in fields	de facto, Unix tools
MySQL OUTFILE	,	"	backslash: \"	MySQL SELECT INTO OUTFILE
PostgreSQL COPY	,	"	double the quote	PostgreSQL COPY ... CSV
European semicolon	;	"	double the quote	Excel, decimal-comma locales
Pipe-delimited	\|	usually none	often forbid pipes in fields	de facto, mainframe and ETL

The most common silent failure sits at the boundary between RFC 4180 (quote-doubled) and MySQL (backslash-escaped): a field holding a literal " reads fine in the dialect that wrote it and corrupts in the other.

Separator	Hex	Decimal	Used by	Detected by Rowson
Comma	0x2C	44	RFC 4180, Excel (en-US), MySQL, PostgreSQL	yes (priority 1)
Tab	0x09	9	IANA TSV, Unix TSV	yes (priority 2)
Semicolon	0x3B	59	Excel (European locales)	yes (priority 3)
Pipe	0x7C	124	mainframe, ETL, Apache Hive	yes (priority 4)

Rowson detects the top four. Anything else, such as the ASCII unit separator (0x1F), a caret, or a colon, needs a pre-processing step that swaps the separator for a tab or comma before paste.

For quoting, RFC 4180, Excel, and PostgreSQL COPY all quote a field when it contains a delimiter, a quote, or a CR/LF, and write a literal quote as two adjacent quotes:

id,name,note
1,"Smith, John","She said ""hello""."

MySQL OUTFILE instead backslash-escapes (\"), which is not RFC 4180. Round-tripping through it and back into a default parser mis-reads any field with a literal backslash. IANA TSV backslash-escapes tab and newline inside fields; the looser Unix convention forbids them entirely and uses no quoting, which is what awk -F'\t', cut, and sort assume.

JSON output formats and where each is used

Format	Specification	MIME type	Extension	Typical consumer
Array of objects	RFC 8259 / ECMA-404	application/json	.json	JS apps, REST request bodies
JSONL	jsonlines.org	application/x-ndjson	.jsonl, .ndjson	BigQuery bq load, DuckDB, OpenAI fine-tuning
Array of arrays	RFC 8259	application/json	.json	Chart libraries (Plotly, Highcharts)
Keyed object	RFC 8259	application/json	.json	Lookup tables, ID-indexed cache loads

Each format trades a property the others keep.

Property	Array of objects	JSONL	Array of arrays	Keyed object
Preserves all rows when keys collide	yes	yes	yes	no
Preserves header names	yes	yes	first row only	yes
Parseable line by line	no	yes	no	no
Parseable as a single JSON value	yes	no	yes	yes
Round-trips back to CSV cleanly	yes	yes	yes	no (lossy)
Supports streaming insert	no	yes	no	no

JSONL is the only line-orientable format; the other three require the consumer to buffer the whole document before parsing. For a multi-hundred-megabyte file destined for a data warehouse, JSONL is the only viable choice.

Encoding, headers, and parser libraries

Encoding	BOM bytes	What modern parsers do
UTF-8 (no BOM)	none	parse as-is
UTF-8 with BOM	0xEF 0xBB 0xBF	PapaParse 5.4.1+, Python csv 3.9+, Go 1.20+ strip it
UTF-16 LE/BE	0xFF 0xFE / 0xFE 0xFF	most parsers fail; pre-decode to UTF-8
Windows-1252	none	decoded as Latin-1 unless the reader sniffs

When an Excel-saved CSV reads back as mojibake, the cause is almost always Windows-1252 vs UTF-8; save as “CSV UTF-8 (Comma delimited)” instead.

Header convention	First row contains	Common in
Headers present	column names	Excel exports, web downloads, API fixtures
No headers	data from row 1	sensor logs, older database exports
Commented headers	#-prefixed rows above data	Unix tools, R read.csv(comment.char="#")

Rowson assumes headers present. For headerless input, prepend a synthetic header row or pick array-of-arrays and let the consumer assign keys.

Language	CSV library	JSON output
JavaScript / TS	PapaParse (Rowson uses this)	array of objects; auto-delimiter via delimiter: ""
Python	pandas.read_csv	to_json(orient="records"); add lines=True for JSONL
Go	encoding/csv (stdlib)	manual; pair with encoding/json
Java	OpenCSV, Apache Commons CSV	manual; pair with Jackson
Shell	csvkit (csvjson)	AoO, plus --stream for JSONL

Common conversion failures

Symptom	Cause	Fix
Numeric columns become quoted strings	CSV has no numeric type	Cast after parsing: Number(row.amount)
Leading zeros stripped on IDs	Excel coerces 00123 to 123 on edit	Import the column as text in Excel
Output has fewer rows than the CSV	Keyed-object mode collapsed duplicate keys	Switch to array of objects or JSONL
null values come out as "null"	CSV has no null literal	Normalize empty, "null", "\N" (MySQL) to JSON null
Characters render as Ã© instead of é	UTF-8 file decoded as Latin-1	Re-save source as UTF-8 with BOM
First field name appears as id	UTF-8 BOM not stripped	input.replace(/^/, "") in preprocessing
JSONL rejected by BigQuery	Trailing blank line or non-UTF-8 byte	Strip trailing newline; verify with iconv

Related concepts

RFC 4180: the closest thing to a CSV specification. Two pages. Describes the comma-quote-doubled-CRLF dialect most libraries default to; silent on encoding, headers, and escape variants.
JSON Lines (jsonlines.org): the informal spec for newline-delimited JSON. Predates RFC 7464 and is more widely used.
ND-JSON (application/x-ndjson): identical content to JSONL, different MIME type and trailing-newline rules. Often interchangeable.
text/csv MIME type: registered in RFC 7111, defines optional charset and header=present|absent parameters almost no producer sets.
CSVW: a W3C effort to add schema and semantics to CSV via a sidecar JSON manifest. Seen in government open-data publishing; rare elsewhere.