Skip to content

Word Document Compare

Diffs two .docx files paragraph by paragraph with word-level marks inside changed paragraphs, side by side or unified, and exports an HTML or text report.

Two versions of a contract and the question of what actually changed. Word’s own compare needs both files open in Word; this reads them in the browser, matches paragraphs by longest common subsequence, and marks the changed ones word by word.

🔒 Runs entirely in your browserworddiffcomparelegal

The earlier document (.docx)

Read inside this tab. Nothing is uploaded.

The later document (.docx)

Drop the revised version of the same document.

Paragraphs are matched by longest common subsequence, and a deletion and an addition that resemble each other are paired as a changed paragraph and compared word by word. A table row counts as one paragraph. Formatting — font, colour, indentation — is not compared: if the text matches, the documents match.

How to use it

  1. Drop the earlier and the later document

    Both are read in this tab. Neither is uploaded, which matters for a contract.

  2. Read the four counts

    Added, removed, changed and unchanged. A large “changed” count with small word-level differences usually means a reformat rather than a rewrite.

  3. Switch to “changes only” for a long document

    It hides the paragraphs that match, which on a fifty-page contract is most of them.

  4. Export a report

    The HTML report keeps the colours and prints; the text one has +/- markers and pastes into an email.

Common questions

Does it compare formatting?

No. Font, colour and indentation are ignored: if the text matches, the documents match. That is deliberate — a reformat should not read as a hundred changes.

How are tables compared?

A row counts as one paragraph. That finds a changed cell and does not try to align columns between two tables whose structure may differ, which is where table diffs usually go wrong.

A paragraph shows as both removed and added.

It was rewritten enough that the two no longer resemble each other. The pairing threshold is a similarity measure, and below it the honest reading is that one paragraph went and another arrived.

Feed it from

Next

These accept what this tool produces, so the result can go straight in.