Text compare that marks the words, not the lines

Paste the earlier version into the left box and the later one into the right, and the differences appear immediately. There is no upload step, no account and no button — the two columns fill in as you type, with the changed words marked inside the lines they belong to rather than the whole line flagged as different.

Because the paragraph is stored as one very long line, and inserting a word near the beginning re-wraps everything after it. Editors that hard-wrap at 72 or 80 characters make this worse: change “the” to “a” in the first sentence and every subsequent line shifts by two characters, so a line-based comparison reports the entire paragraph. This page marks the changed words inside the line, so a re-wrapped paragraph shows one green word rather than forty red lines. If you write in a plain-text editor and expect to compare often, keeping one sentence per line removes the problem at the source.

Only when you ask it to, and the two toggles do different jobs. “Ignore spacing” treats every run of spaces, tabs and line breaks as a single space, which is what you want after a paragraph has been re-flowed or after someone converted double spaces after full stops to single ones. “Ignore capitalisation” lowercases both sides before comparing, so a heading restyled from Title Case to sentence case stops registering. Both change the counts in the summary, not just the colours, so the numbers always describe what you chose to care about.

Anything made of letters or digits, including hyphenated compounds and words with apostrophes — “well-known” and “didn’t” are each one word. Spacing and standalone punctuation are compared and shown but not counted, so moving a comma does not inflate the total, and a sentence rewritten in place is counted twice over: once for the words that left and once for the words that arrived.

Yes, and the invisible characters that come with it are exactly what this comparison is good at finding. Word processors autocorrect straight quotes to curly ones, hyphens to en dashes and ordinary spaces to non-breaking spaces, so a passage that looks identical to a passage typed in a code editor is not. Those substitutions show up here as marked characters — switch to “Character by character” to see precisely which one changed, since at word level a curly-quoted word simply reads as a different word.

Word by word for prose, character by character when the change is inside a word. Word level is how an editor thinks — a replaced phrase reads as one edit — but it cannot show you that a reference number went from 4417 to 4471, because it just reports one word swapped for another. Character level splits on code points rather than UTF-16 units, so an accented letter or an emoji is one unit and never breaks in half. Lines longer than 6,000 words or characters are shown as wholly replaced instead, because marking inside them costs more time than the answer is worth.

No. Both drafts stay in the browser tab: the comparison is JavaScript running on your own machine, nothing is posted to a server and nothing is stored, which is what makes it safe to paste an unpublished chapter or an internal memo in here.