CSV cleaner

Methods last reviewed 30 September 2026.

How to use it

1. Paste your data into the box or open a .csv, .tsv or .txt file. Cells copied from Excel, Google Sheets or Numbers paste as tab-separated text and are detected automatically. Large files are cleaned in full; the box then shows only their first lines.
2. Look at the preview. Changed cells are highlighted, and hovering over one shows its old value. Each step shows how many cells, rows or columns it changed. Tick "Changed rows only" to review just those.
3. Switch steps on or off, open a step to adjust it, and move steps earlier or later: they run from the top. Add a second find-and-replace, or any other step, with the control under the list.
4. If a yellow question appears, answer it: the tool will not guess whether a date is day-first or whether 1,234 is one thousand.
5. Download the result, copy it, or open it in one of the chart tools.

What each step does, and where it stops being reliable

Step	Rule	Limits
Trim whitespace	Removes leading and trailing spaces, tabs, line breaks and zero-width characters, in cells and column names.	Leaves spaces that are meant to be there, inside the text.
Collapse internal spaces	Any run of spaces, tabs or non-breaking spaces becomes one space.	Line breaks inside a cell are kept; fixed-width text loses its alignment.
Empty rows and columns	A row or column is empty when every cell is blank or whitespace.	A cell holding "N/A" is not blank; use Fill blanks with the null option first.
Duplicate rows	Exact match on all columns, or on the ones you tick; the first copy is kept.	No fuzzy matching: "Jon Smith" and "John Smith" are different rows.
Dates to ISO 8601	Day and month order is decided per column from dates that can only be read one way.	Asks when a column never settles it. Two-digit years 69–99 are read as 19xx, 00–68 as 20xx.
Normalise numbers	Currency signs and codes, spaces and grouping separators removed; the decimal mark becomes a point.	Percentages keep their % sign. Leading zeros are kept, so IDs and postcodes survive.
Change case	Title Case capitalises the first letter after a space, hyphen, slash or bracket.	Names such as McDonald or van der Berg need a find-and-replace afterwards.
Find and replace	Plain text, or JavaScript regular expressions with $1 for groups.	A regular expression that never ends on very long cells will make the step slow.
Split and merge	Split makes up to 20 columns; extra parts stay in the last one. Merge skips blank parts.	Split does not respect quotes inside a cell.

How dates are read

The step recognises ISO dates (2024-03-04, with or without a time), numeric dates with slashes, dashes or dots (04/03/2024, 4-3-24, 21.03.2024), and dates with a month name (19 Feb 2024, Feb 19, 2024). Every date is checked against the real calendar, so 2024-13-45 and 29/02/2023 are reported as unreadable instead of rolling over into another month. Times are kept, as T09:15.

The order of day and month cannot be read from one cell like 03/04/2024. It can often be read from the column: one 15/04/2024 proves the column is day-first. The rule is applied to each column separately, because a file joined from two systems can have an American order date next to a European invoice date. When a column has evidence both ways, the step converts each date that can only be read one way, leaves the rest, and tells you the column needs checking by hand. When there is no evidence at all, it asks.

How numbers are read

A cell is treated as a number when, after removing a currency sign or code ($, €, £, USD and others) and a minus sign or accounting brackets, only digits and separators are left, grouped in threes. (1,200.00) becomes -1200.00, 1.234,50 € becomes 1234.50, and a Swiss 1'234 becomes 1234. Digits are rewritten as text, never converted through floating point, so no precision is lost and 0.10 stays 0.10. Cells that are not numbers, such as phone numbers and dates, are left alone.

Opening the cleaned file in a spreadsheet

Spreadsheets reinterpret a CSV as they open it. Excel reads dates using your computer's regional setting and drops leading zeros from anything that looks like a number. Import the file with Data › From Text/CSV instead of double-clicking it, and set ID, postcode and phone columns to Text. Choose semicolons as the delimiter if your Excel uses a comma as its decimal mark, and add a byte-order mark if names with accents come out garbled.

Where this tool stops being the right one

It cleans one table at a time. It does not join two files, match names that are spelt differently, check that email addresses or phone numbers are real, or remember a set of steps for next month's export. For repeated jobs, or files of hundreds of megabytes, a script in Python with pandas, or OpenRefine, which has clustering for near-duplicates, will serve you better. To see what is in the cleaned data, try the summary statistics or pivot table tools; to present it, the bar and line chart maker or the Markdown and HTML table converter.