CSV Quote Normalizer
Read the CSV a strict parser refuses — backslash escapes, single quotes, curly quotes, stray quotes, an unclosed field — and write it back out in one consistent dialect.
About this tool
A CSV is only half a format. The delimiter is obvious; the quoting is not, and every exporter picks its own convention. MySQL writes \" inside a quoted field. Postgres COPY can be told to do either. Word and Google Docs turn " into “ and ” the moment someone opens the file to fix a typo. A hand-edited row picks up 'single quotes', or a stray " in the middle of a comment, or a space before the opening quote. Every one of those files looks fine in a text editor and then blows up in pandas, csv.reader, Excel's import wizard, or a COPY … FROM — usually not with an error, but with a row silently split in the wrong place.
This tool reads the file with a deliberately tolerant parser and writes it back out strict. Nothing about the input has to be well-formed; everything about the output is.
What it will read
| In the input | How it is read |
|---|---|
"x \"y\"" | backslash-escaped quote → a literal " in the value |
'Boston, MA' | single-quoted field (when ' actually opens a field) |
“smart quoted” | curly quotes as field quotes |
1, "padded" | padding before or after a quote, dropped |
"He said "hi" to me" | stray inner quote kept as text |
"abc"def | text after a closing quote merged back into the value |
"oops,2 (EOF) | unclosed field, closed at end of input |
at the start, blank lines | dropped |
What it will write
One dialect, chosen by you: which fields are quoted, which character quotes them, how a quote inside them is escaped, which delimiter separates them, and which line ending terminates the row — embedded newlines included, so the whole file is consistent.
Worked example
Input (the default example on this page):
id,name,note
1,"Ada \"Countess\" Lovelace",fine
2,“Grace Hopper”, "padded, quoted"
3,"He said "hi" to me",ok
Output:
id,name,note
1,"Ada ""Countess"" Lovelace",fine
2,Grace Hopper,"padded, quoted"
3,"He said ""hi"" to me",ok
Four different quoting conventions went in; one came out. Grace Hopper lost its quotes because minimal quoting does not need them, and padded, quoted kept its because it contains the delimiter.
Controls
- Input delimiter —
auto(default) sniffs the most frequent of,;tab|outside quotes on the first logical line, or name one (comma,tab,semicolon,pipe,space) or give a single character. - Output delimiter —
same(default), or any delimiter spec to change dialect on the way out. - Quote character to read —
auto(default) picks double if any straight or curly double quote appears, and single only when a'genuinely opens a field, soit'sin a comment is never mistaken for a quote.nonetreats every quote as literal content. - Which output fields to quote —
minimal(default) quotes only what must be quoted;alwaysquotes everything, which keeps a diff stable and stops a spreadsheet retyping007as7;non_numericquotes everything that is not a plain decimal number, matching Python'scsv.QUOTE_NONNUMERIC;neverquotes nothing, and then needs backslash escaping to stay readable. - Output quote character —
"(default) or'. - How to escape a quote inside a field —
doubled("", RFC 4180) orbackslash(\"). - Read
\"in the input as an escaped quote — on by default. Turn it off when backslash is ordinary content. - Read curly quotes as field quotes — on by default. Turn it off to keep
“and”as characters in the value. - Line ending —
LF(default) orCRLF. - Result — the rewritten CSV, or a report of what was repaired.
The report
Switch Result to A report of what was repaired and you get an audit instead of a file: the detected input delimiter and quote character, the output dialect, the row and field counts, a warning if the rows are ragged, how many fields were quoted before and after, and every repair grouped by kind with the line numbers it happened on. It is the fastest way to find out why a file was breaking your loader.
Limits and edge cases
The input is capped at 5,000,000 bytes. This tool re-quotes and nothing else: cell values are never trimmed, retyped, rounded or re-cased, rows are never added, dropped or reordered, and a ragged row keeps its length — it is reported, not padded. quote_style = never with escape = doubled is refused with the offending row and field named, because there would be no way to represent a value containing the delimiter; the fix is backslash escaping or a different policy. The output always ends with one line terminator. Everything runs locally in your browser — the file is never uploaded.
FAQ
My CSV loads fine in a text editor but pandas splits a row in the wrong place. Why?
Almost always backslash-escaped quotes. An exporter wrote "x \"y\""; a strict RFC 4180 reader does not know \ is an escape, so it sees the " after the backslash as the closing quote and everything after it as a new field. Paste the file here with default settings: the tolerant parser reads the backslash convention, and the output uses "" instead, which every strict reader understands.
What is the difference between doubled and backslash escaping?
They are two ways to put a quote character inside a quoted field. RFC 4180 says double it: "she said ""hi""". MySQL's LOAD DATA, older Postgres COPY variants and most JavaScript-flavoured writers use a backslash: "she said \"hi\"". Neither is wrong, but a reader that expects one and gets the other mis-parses the row. Pick the one your destination expects under How to escape a quote inside a field — and leave Read \" in the input as an escaped quote on so the source convention is understood either way.
Where do the curly quotes come from, and should I keep them?
From autocorrect. Word, Google Docs, macOS text substitution and most CMS editors rewrite " as “ … ” as you type, so any CSV that a human opened and edited can come back with typographic quotes wrapping its fields. With Read curly quotes as field quotes on (the default) they are treated as quoting and the output gets straight quotes. Turn it off when the curly quotes are real content — a quotation inside a text column, for example — and they will be preserved as characters.
Can I change the delimiter at the same time?
Yes. Set Output delimiter to tab, semicolon, pipe, space or any single character. That is safer than a plain find-and-replace, because the switch happens on the parsed fields: a comma inside a value stays part of the value, and quoting is recalculated for the new delimiter — so a field that only needed quotes because of a comma often comes out unquoted in a TSV.
What happens to a row with an unclosed quote?
It is closed at the end of the input rather than swallowing the rest of the file, and the report tells you which line opened it. That is a repair, not a guess about your intent, so check the result: if a quote was opened by accident, everything after it on that line has been folded into one field. The report's line number points straight at the row to fix upstream.
Does it trim spaces, fix headers or pad short rows?
No — deliberately. This tool changes quoting, escaping, the delimiter and line endings, and nothing else. Padding inside cells is csv-whitespace-normalizer; reporting structural problems without rewriting is csv-structure-validator. Keeping the jobs separate means you can always tell what a step changed.
Does my file get uploaded?
No. The parser and writer are compiled to WebAssembly and run inside the page, so the CSV never leaves your browser. For a file on disk you can run the same code offline from the terminal: gizza tool csv-quote-normalizer input="$(cat data.csv)".
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool csv-quote-normalizer 'id,name,note
1,"Ada \"Countess\" Lovelace",fine
2,“Grace Hopper”, "padded, quoted"
3,"He said "hi" to me",ok'New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/csv-quote-normalizer/?input=id%2Cname%2Cnote%0A1%2C%22Ada%20%5C%22Countess%5C%22%20Lovelace%22%2Cfine%0A2%2C%E2%80%9CGrace%20Hopper%E2%80%9D%2C%20%22padded%2C%20quoted%22%0A3%2C%22He%20said%20%22hi%22%20to%20me%22%2Cok&delimiter=auto&output_delimiter=same&input_quote=auto"e_style=minimal&output_quote=double&escape=doubled&backslash_escapes=true&smart_quotes=true&line_ending=lf&output=csvMachine-readable descriptor: tool.json — title + parameters JSON Schema for agents.