CSV Cleaning Studio

Clean messy CSV files — duplicates, blanks, stray spaces and mixed dates — entirely in your browser.

Runs in your browser. Nothing is uploaded — your input is processed on this device and discarded when you close the tab.

Clean a file

Drop a CSV file here or click to browse. Files up to 20 MB and 200,000 rows are accepted, and the file is read on this device only.

Or paste CSV data

Pasted data is parsed with the delimiter and encoding options below, exactly like an uploaded file.

Load pasted data

Delimiter Auto-detect Comma (,) Semicolon (;) Tab Pipe (|)

File encoding Auto-detect UTF-8 UTF-16LE UTF-16BE Windows-1252 ISO-8859-1

Changing the delimiter or encoding after a file has loaded re-reads the original file with the new setting, so a wrong first guess costs nothing.

Columns

Rename a column by editing its name, untick Keep to drop it, tick Dedupe key to mark it as part of the duplicate check, and use the arrow buttons to change the column order. The badge beside each name is the data type the studio detected from the values themselves. Nothing here takes effect until you choose Apply operations.

Cleaning operations

Operations always run together, in a fixed order: column keep, rename and reorder first, then trimming, case changes, date standardisation, missing-value handling, the row filter, and finally deduplication. The fixed order matters: duplicate detection sees the already-trimmed, already-standardised values, so two rows that differ only by stray spaces are still recognised as copies.

Trim leading and trailing whitespace in all cells

Text case (applies to text columns) Leave case unchanged lowercase UPPERCASE Title Case

Date format (applies to date columns) Leave dates unchanged ISO — YYYY-MM-DD DD/MM/YYYY MM/DD/YYYY

Missing values (empty or whitespace-only cells) Keep empty cells Drop rows containing an empty cell Fill with a fixed value Fill with the column mean Fill with the column median Fill with the column mode

Fill value (used with “Fill with a fixed value”)

Row filter

Filter column Choose a column…

Condition Equals Contains Greater than Less than Is empty Is not empty

Filter value

Only rows matching the condition are kept. Greater than and less than compare numbers numerically when both sides are numeric. Leave the column unchosen to skip filtering.

Deduplication uses the columns you marked as Dedupe key in the column list above. If no column is marked, deduplication is off: two rows count as duplicates only when every key column matches, and the first occurrence is the one kept.

Apply operations Undo Start over

Change summary

No data loaded yet. Load a file or paste data above, and each Apply will report exactly what it changed here.

Preview — before

Before — first 200 rows

Preview — after

After — first 200 rows

Export

Formula-safe export: prefix cells starting with =, +, − or @ with an apostrophe so spreadsheets can’t run them as formulas

The apostrophe is added to the downloaded file only — your on-screen data is never altered by this option. See the FAQ below for why it matters.

Download cleaned CSV Open in Chart Builder

“Open in Chart Builder” hands the cleaned table to the Chart Builder in this browser only (via local storage) — nothing is uploaded.

How to use the CSV Cleaning Studio

1. Load your data. Drop a CSV, TSV or plain-text file onto the dropzone, click it to browse, or paste raw CSV text into the box and choose Load pasted data. The status line confirms the file name, the delimiter and encoding that were detected, and the row and column counts.
2. Check the parse. If columns arrive merged into one, or accented characters look wrong, change the delimiter or encoding selectors — the original file is re-read with the new settings, so you can experiment freely.
3. Shape the columns. Rename headers so they are consistent, untick Keep on columns you do not need, reorder with the arrow buttons, and mark the Dedupe key columns — an email address, an order number, or any combination that should be unique.
4. Choose operations. Switch on trimming, pick a case style for text columns, pick one output format for dates, decide how empty cells are handled, and optionally set a single row filter.
5. Apply and inspect. Choose Apply operations, then read the change summary: rows before and after, duplicates removed, cells changed, and a note for every fill, filter and unrecognised date. Compare the Before and After previews, which show the first 200 rows of each state.
6. Refine or download. Apply again with different settings, use Undo to step back one Apply at a time, or Start over to return to the file exactly as loaded. When the After preview looks right, download the cleaned CSV.

Worked example: a messy sign-up export

The sample below is a small customer sign-up export with the faults real exports usually have: names padded with spaces, one person entered twice, city names typed in three different case styles, two missing prices, and signup dates written in three different formats. Load it and clean it with the tool itself:

With the sample loaded, try this cleaning pass. Tick Dedupe key on the Email column, tick Trim, set text case to Title Case, set the date format to ISO, set missing values to Fill with the column median, then choose Apply operations. The verified result is:

• 7 rows become 6. The second Alice Smith row is removed because its Email matches the first — deduplication runs after trimming, so the padded spaces in the first Alice row do not hide the match.
• Two empty Price cells are filled with 20, the median of the prices present (10, 10, 20, 30 and 40), and the change summary names the column and the value used.
• Every date becomes ISO format: 05/01/2024 and 15 Jan 2024 both become dates in January 2024, written as 2024-01-05 and 2024-01-15, and 02/03/2024 becomes 2024-03-02.
• Cities are consistent: “london”, “PARIS” and “berlin” all become Title Case, while email addresses are deliberately left untouched because the case operation only applies to columns detected as text.

Undo the pass and try a different missing-value strategy — for example filling with the mean — and the summary will show how the two choices differ before you commit to either in the downloaded file.

What the CSV Cleaning Studio is, and why it matters

A CSV file looks simple — rows of values separated by commas — but that simplicity is exactly why messy files cause so much quiet damage. Almost every system that handles lists of people, products, orders or payments can export a CSV, and almost every other system can import one, which makes CSV the hand-off format between tools that were never designed to agree with each other. The disagreements are small and human: a name typed with a trailing space, the same customer exported twice, a date written as 05/01/2024 by one system and as January 5 by another, a price left blank because nobody filled in the field. Individually each fault looks harmless. Together they break the things people actually do with the data. A mail merge greets “ Alice” with a visible gap. A pivot table counts London, london and LONDON as three different cities. A duplicate row mails the same person twice, or pays a supplier twice. A blank price silently distorts an average. A column of mixed date formats sorts into nonsense and defeats every filter built on it.

Cleaning a file by hand in a spreadsheet is possible, but it is slow, hard to repeat and easy to get half right: you fix the duplicates you notice, trim the cells you happen to select, and have no record afterwards of what changed. The CSV Cleaning Studio takes the opposite approach. It applies a small set of well-understood operations — trimming, case normalisation, date standardisation, explicit missing-value strategies, one clear row filter and key-based deduplication — to the whole file at once, in a fixed, documented order, and then tells you precisely what it did: how many rows went in and came out, how many duplicates were removed, how many cells changed, and a plain-language note for each fill, filter and value it could not interpret. That report is the point. A cleaned file you cannot audit is just a different file you have to trust; a cleaned file with a change summary is one you can check, defend and reproduce next month when the same export arrives again.

The studio also respects the data it is given. Column types are detected from the values, so a case change never rewrites email addresses or numbers, a date operation only touches columns that genuinely hold dates, and a mean or median fill is calculated only from a column’s own numeric values — if a column has none, the tool says so instead of inventing a plausible-looking number. Everything runs in your browser: the file is read on your device, never uploaded, and disappears when you close the tab, which makes the tool safe to use on customer lists, payroll exports and other files that should never be sent to a stranger’s server.

Common errors and limitations

• Everything lands in one column. The delimiter was detected wrongly — common with European exports that use semicolons. Choose the correct delimiter from the selector and the file is re-read immediately.
• Accented characters look garbled. The file is usually Windows-1252 or UTF-16 rather than UTF-8. Change the encoding selector; auto-detection already handles byte-order marks and falls back gracefully, but it cannot read the sender’s mind.
• Ambiguous dates. A date such as 05/06/2024 cannot be told apart from context alone — it could be 5 June or May 6. This tool reads ambiguous numeric dates day-first, so 05/06/2024 is treated as 5 June. Check the summary count of unrecognised dates and the After preview if your source system writes month-first dates.
• Deduplication is exact, not fuzzy. Rows match only when every key column is identical after trimming and your other operations. “Jon Smith” and “John Smith” with different emails will both survive; near-duplicate detection needs a human decision this tool deliberately does not make.
• Fill statistics use numeric values only. A price column containing entries like “free” or “TBC” will have those entries ignored by mean, median and mode fills, and a column with no numeric values at all is reported and left empty rather than filled with a guess.
• Hard limits. Files over 20 MB and files with more than 200,000 data rows are refused with an explanation, and previews show the first 200 rows only — the operations themselves always apply to the full file. Very wide or badly quoted files may parse with warnings; when the status line reports parse warnings, inspect the previews before downloading.

Frequently asked questions

Related tools

Related articles