
The sample on the page is a nine line note to a team. The launch moves from Thursday at 10am to Friday at 11am. The slides are due a day later. Lunch is gone, and there's a new line about parking.
Click Use the sample on the Text Compare and Diff Checker and the line above the table reads: 1 line added, 1 removed, 2 changed. 9 words added, 7 removed. In the launch line, only Thursday and 10am are marked on the left, and only Friday and 11am on the right. The rest of the sentence stays plain.
That's the whole point of the page. A small edit should look small.
The whole paragraph turns red
Most compare tools work line by line. That's fine for code, where a line is short and one edit tends to live on one line. It's bad for prose. A contract clause or an email paragraph is often one long line. Change one word in it and a line based tool shows the whole thing twice, red and then green.
Then you read both copies, hunting for the word that moved. On a long clause that's slow, and it's easy to miss the one that matters.
I tried a 31 word clause about rent, where the late fee kicks in after the fifth day. I changed fifth to tenth. The checker reports 1 changed line, 1 word added and 1 removed, and the mark sits on fifth and tenth. Nothing else on the line is highlighted.
Change Thursday to Friday and only Thursday and Friday light up.
That line is from the page itself. The fifth and tenth clause is the same idea on a longer line.
Lines first, then words
The checker does the work in two passes.
The first pass lines up the two texts by line. It uses the Myers algorithm from 1986, which is also what git diff uses by default. Before it starts, it trims the lines both texts share at the top and the bottom, so a long file with one change near the end doesn't make it walk the whole thing.
The second pass runs only on lines that changed. When a run of removed lines sits next to a run of added lines, the checker pairs them up in order, first removed with first added, and runs the same Myers algorithm again. This time it compares words, not lines.
A word here means a run of letters, digits and underscores. Spaces stay as their own pieces, and every other character is its own piece too, so a period or a comma can be marked on its own. An emoji with a skin tone stays one piece. When the two sides are put back together, every character you pasted is still there.
The sample, row by row
Here's what the built-in sample turns into. I ran the comparison on it and copied the rows as they came out.
| Original | Changed | Row |
|---|---|---|
| The launch moves to Thursday at 10am. | The launch moves to Friday at 11am. | changed, 2 words each side |
| Please send your slides by Tuesday. | Please send your final slides by Wednesday. | changed, 1 out, 2 in |
| Lunch is on us. | removed | |
| The room is the same as last time. | The room is the same as last time. | unchanged |
| Parking is free after 9am. | added |
The greeting, the blank lines and the sign off match, so they count as unchanged.
The status line adds it up. Words added: Friday and 11am, then final and Wednesday, then the five words of the parking line. That's 9. Words removed: Thursday and 10am, then Tuesday, then the four words of the lunch line. That's 7. 11am and 10am each count as one word, because digits and letters run together.
When a line shares no word
Pairing a removed line with an added line only helps when they have something in common. If they don't share a single word, marking the words would mark every one of them, and you'd be back to a red line and a green line with extra noise on top.
So the checker checks first. If the paired lines share at least one word, the row becomes one changed line with the edited words marked. If they share none, it shows one line removed and one line added.
I ran "Lunch is on us." against "Parking costs nothing." on their own. The status line says 1 line added, 1 removed, 0 changed, with 3 words added and 4 removed. Those are two different sentences, and the page says so.
Punctuation doesn't count as sharing. Two lines that only have a period in common still show as removed and added.
Ignore spaces
Re-indent a block of code and every line in it looks different, even though nothing changed that a person would care about. Paste text from an email and you often pick up double spaces.
Ignore spaces handles both. It treats any run of spaces and tabs as one space, and it drops spaces at the start and the end of a line before comparing.
I compared a three line snippet where one line was indented with four spaces on one side, and with a tab plus two trailing spaces on the other. With the switch off, the result is 1 changed line, 0 words added, 0 removed. The line is different, but no word is. With the switch on, the page says the two texts are identical.
The switch only changes what counts as different. The text you see stays exactly as you pasted it, tabs and all. The .diff you copy also carries your original lines.
Ignore case, and line endings
Ignore case treats Hello and hello as the same word. "Hello world" against "hello world" is 1 changed line with 1 word added and 1 removed when the switch is off. With it on, the two texts are identical.
Line endings don't get a switch, because there's no good reason to see them. Windows ends a line with a carriage return and a line feed. Unix and macOS end it with a line feed alone. A file that went through both kinds of machine can carry either one. The checker splits on both and counts them as the same, always. I compared "one" and "two" with Windows endings against the same two lines with Unix endings, and the result is identical.
| Switch | What it ignores | What you still see |
|---|---|---|
| Ignore spaces | Runs of spaces and tabs, spaces at the line ends | Your text as pasted |
| Ignore case | Upper and lower case | Your text as pasted |
| (always on) | Windows versus Unix line endings | Your text as pasted |
| Hide unchanged lines | Long runs of matching lines | Three lines around each change |
Side by side or one column
Side by side puts the original on the left and the changed text on the right. Each side has its own line numbers, because after an added or removed line the numbers stop matching. Red is what got taken out, green is what got added. Inside a changed line, a darker mark sits on the exact words that moved. A striped cell means that line only exists on one side.
One column shows the same result the way a code review does: in each run of changes, the minus lines come first, then the plus lines.
On a phone the table switches to one column by itself. That happens on any screen 639 pixels wide or narrower. Two columns on a phone squeeze every line down to a few words, and then you're scrolling sideways to read a sentence. The switch you picked is kept. Turn the phone or open the page on a laptop and side by side comes back.
Folds keep three lines of context
A long document with one change in the middle is mostly sameness. Hide unchanged lines is on by default and folds long runs of matching lines away.
Each change keeps three unchanged lines above it and three below. Everything farther out collapses into a fold that says how many lines it holds. I made a 40 line text and changed line 20. The table shows a fold of 16 lines, three lines of context, the changed line, three more lines of context, and a fold of 17. The first fold reads "Show 16 unchanged lines". Click it and it opens.
A run of one or two unchanged lines never folds. A button that hides one line takes more space than the line.
Turn the switch off and every line is drawn. When the result changes, say you edit a text or flip Ignore case, any fold you opened closes again. The rows behind it may not be the same rows anymore.
Copy as .diff
The buttons above the table are Copy as .diff and Download .diff. Both give you a unified diff, the format git and the patch command read. For the sample, it starts like this:
--- original
+++ changed
@@ -1,9 +1,9 @@
Hi team,
-The launch moves to Thursday at 10am.
-Please send your slides by Tuesday.
-Lunch is on us.
+The launch moves to Friday at 11am.
+Please send your final slides by Wednesday.
The room is the same as last time.
+Parking is free after 9am.The @@ header gives the start line and the line count on each side. Each change carries three lines of context, the same as git's default, and two changes close enough to share context merge into one block. Download .diff saves the same text as changes.diff.
A .diff is whole lines, so the word marks don't travel. Anyone with a code editor can still read it, and it's the thing to paste into a ticket or a pull request comment.
Big pastes
Each side holds up to 1,000,000 characters. Paste more and the page cuts it there and says so under the box, so the tab stays usable.
The table draws the first 5,000 lines. Past that, the page says it's showing the first 5,000 and points you at the download. The .diff file has every line.
The line matching has a budget too. If more than 2,000 lines were added and removed between the two texts, the Myers search stops early. The shared lines at the top and the bottom are still matched. The middle is shown as one block out and one block in, and a note above the table says that happened. The limit is there so the search can't eat all the memory in the tab. Two texts that far apart are usually not two drafts of the same thing anyway.
What leaves the tab
The texts don't. The comparison runs in JavaScript in this tab. Nothing gets uploaded and nothing's saved on a server.
There's a Copy settings link button. The link it gives you looks like this for the default settings:
view=split&ws=0&case=0&fold=1
That's the view and three switches. It never carries either text, so a contract or a config never ends up in a URL or a chat preview. If you want to share the actual changes, send the .diff.
This browser remembers your view and switches in local storage, so the page opens the way you left it. Clear settings puts the defaults back. The texts aren't written to storage. Close the tab and they're gone.
Mixpanel and Google Analytics load with Softery.io. They see that someone opened the tool, which switch moved, whether the input was typed, pasted or the sample, and a size bucket. A size bucket is a band like "501-2k" characters, not a count. They don't get either text.
Try it in four steps
- Open the Text Compare and Diff Checker and click Use the sample. Read the status line. It should say 9 words added, 7 removed.
- Paste your own two versions. Original on the left, changed on the right. Swap sides if you got them backwards.
- If the result is full of lines that only differ in indentation, turn on Ignore spaces. If it's case, turn on Ignore case.
- Copy as .diff or Download .diff, and send that instead of two attachments.
For JSON, run both files through the JSON Formatter and Validator first so the indentation matches, then compare. If the new version has to fit a limit, the Word and Character Counter checks it, also without sending the text anywhere.
Last updated: November 18, 2026 | Reading time: 10 minutes
Written by Softery.io, which moved a launch from Thursday to Friday and only marked two words.






