CSV Diff

updated 30 August 2026

Before you start

You need:

• Two CSVs — the "old" (left) and "new" (right) version of the same dataset. They should share a header row and, ideally, share most columns.
• A key column (or composite) that identifies a record in both files — usually an id, sku or email. You do not have to name it: the tool reads the headers of both files and picks one, and the chips under the bar let you change it. Only if nothing usable is found does it fall back to comparing whole rows.

The two files don't have to have the same columns or the same row order. The diff matches rows by key, then compares cells of the matched rows.

How to use it

1. Paste or drop your old CSV into the left pane.
2. Paste or drop your new CSV into the right pane.
3. Check the key. The chips under the bar are the columns both files have; the dashed one is what the tool picked. Click another to switch, click a second to make it a composite key (country + sku), or click whole row to compare everything.
4. The result updates as you paste. Compare re-runs it if you want.
5. Review the result below the panes. Colour legend:
   • green — row exists only in the new file (added).
   • red — row exists only in the old file (removed).
   • yellow — row exists in both, but at least one cell differs. The specific differing cells are highlighted.
6. Click Download diff for a CSV with a change_type column.

How the key is chosen

A column can be a key if it appears in both files, is never blank, and never repeats — that is exactly the property that lets a row in one file be matched with a row in the other. Chips for columns that fail the test are dimmed, and hovering one says why.

One exception, deliberately: a column named like an identifier (id, customer_id, sku, email…) is chosen even when it repeats, as long as it is never blank. A repeating id is a problem in the data, and matching on some other column that happens to be unique would hide it behind a diff that looks fine. The status line says how many duplicate keys there were and which side they were on.

Click whole row to compare entire rows as a set instead. That is the right choice when a file has no stable identifier, but it is weaker: a single changed cell reads as a delete plus an add, and there is no cell-level highlight.

What "changed" really means

• Comparison is string-exact. "1.0" and "1" are different. Normalise types in the source if that's noise.
• Only columns that exist in both files are compared for changes. A column added in the new file is reported as part of the "added" row context, not as a cell change for every existing row.
• If the same key appears more than once in one file, only the last row with that key takes part in the comparison — a set of rows keyed by the same value has no single answer to "did it change". The status line counts them and says which file they were in, so you can fix the source before trusting the diff.

Example

Old CSV:

id,name,city
1,Alice,Berlin
2,Bob,Paris
3,Carol,Rome

New CSV:

id,name,city
1,Alice,Berlin
2,Bob,Lyon
4,Dana,Madrid

Diff on id:

• id=1 — unchanged.
• id=2 — changed: city Paris → Lyon.
• id=3 — removed.
• id=4 — added.

Tips & common pitfalls

• Trim noisy columns first. Timestamps and "last updated" columns change on every row and drown out real diffs. Strip them or add them to a separate compare.
• Sort both files by key before download if you plan to eyeball the diff output — the tool itself doesn't need sorted input, but a sorted CSV is much easier to review.
• Big + wide files are slow because every matched row builds a string key and a per-column comparison. If you hit a wall, cut the file by column or by row with Split CSV first.
• Reordered columns are invisible to the diff — the tool lines up columns by name, not by position. If a column was renamed, it looks like "old column removed, new column added".
• Watch out for whitespace: trailing spaces in one file make the cell look "changed" even when it isn't. If the source has that problem, run the file through Dedupe CSV with trim first, or preprocess with a script.

Troubleshooting

Frequently asked questions