Compare two texts

How to use it

1. Paste the old version on the left and the new version on the right — or drag files into the fields. It reads txt, md, csv, json, xml, html, as well as docx, odt, and pdf: the tool unpacks them and extracts the text itself, in the browser.

2. Read the listing: a minus means a line was removed, a plus means one was added, and yellow within the line highlights what changed. Blue marks a section that was not removed but simply moved.

3. Switch to “two columns side by side” if you prefer comparing visually from left to right: empty spaces opposite removed and added lines keep the two sides aligned.

4. Move through changes with the “previous — next” buttons or the n and p keys. The bar to the right of the listing is a map: its markers show where there are many changes, and you can click them.

5. If there are more differences than expected, relax the strictness settings: case, indentation, and blank lines. Often it is not the text that differs but trailing spaces left after copying from an editor.

6. For JSON, XML, and YAML, turn on “Structure”: the data will be parsed and normalized, and differences will be listed as paths such as “items[3].price: 100 → 120”. If both fields are clearly JSON or XML, the tool switches automatically.

7. You can take the result with you: “.patch” is a standard unified diff for patch and git apply, “highlighted page” is a standalone html file that opens without a network connection, and “as a table” is Markdown markup for a task or email.

The formula and what it includes

• Matching lines — The length of the longest common subsequence: the longest set of lines that appears in both texts in the same order. The lines do not have to be consecutive.
• Changed pair — A deleted line and an added line that are at least 60% similar by words. This pair is shown as one line with highlighting, and it contributes to similarity not in full but exactly in proportion to how similar the lines are.
• Pair similarity — The same measure as for the entire text, but calculated by the words in one line: matching words × 2 ÷ (words on the left + words on the right).
• Moved section — Consecutive lines that disappear in one place and appear word for word in another. The text did not change; its order did, so these lines count as matching in the similarity score.
• Difference block — Consecutive changed, deleted, and added lines not separated by matching lines. Replacing one line with another is one block, not two.

Example breakdown: a four-line shopping list

The input data is the same as the default field values: you can calculate it mentally and confirm that the tool is telling the truth.

• Left: Milk, Bread, Eggs 10 pcs, Butter 180 g — 4 lines
• Right: Milk, Bread, Eggs 20 pcs, Cheese 200 g — 4 lines

1. The first two lines match and appear in both texts in the same order — this is the common subsequence, containing 2 lines.

2. “Eggs 10 pcs” and “Eggs 20 pcs” are split into words: on the left, “Eggs”, a space, “10”, a space, “pcs”; on the right, the same with “20”. Four out of five words on each side match, so the pair similarity is 4 × 2 ÷ (5 + 5) = 0.8. That is more than 60%, so the line is considered changed, and “10” is highlighted as the only edit.

3. “Butter 180 g” and “Cheese 200 g” have almost nothing in common: their similarity is 0.5, below the threshold. So this is not one changed line but a removed line and an added line.

4. All four lines are consecutive, with no matches between them, so there is one difference block.

5. Similarity: (2 matching + 0 moved + 0.8 for the changed pair) × 2 = 5.6, with 4 + 4 = 8 lines total.

Similarity = 5.6 ÷ 8 × 100% = 70%. One line changed, one was removed, one was added, and there was one difference block.

Frequently asked questions

• How large a text can it handle?

  Two hundred thousand lines on each side. Comparison takes roughly as long as the number of differences, not the product of the input lengths, so one hundred thousand nearly identical lines are processed in a fraction of a second. The old five-thousand-line limit and the fallback to “everything old deleted, everything new added” are gone. The listing shows the first four thousand lines of the diff — no one will read more anyway — while the complete difference goes into the .patch file.

• Why isn't changing one word shown as two lines?

  Because a deleted line and an added line are first compared word by word. If they are at least 60% similar, they count as one changed line, and only the exact changes are highlighted. If they are not similar, they remain a minus and a plus: presenting a completely rewritten line as an edit would be misleading.

• How does “by words” differ from “by lines”?

  It depends on the comparison unit. Line-by-line analysis is useful when the text was edited in specific places: you can see which line of the file to look at. Word-by-word analysis is useful when a paragraph was rewritten completely: line-by-line comparison shows the entire paragraph being replaced, while word-by-word comparison leaves the words that survived untouched. Character-by-character analysis is needed least often, mainly to catch a typo or a look-alike character substitution.

• What does “Structure” mode do?

  JSON, XML, and YAML are parsed and normalized: keys are sorted alphabetically and indentation is set to two spaces. After that, only substantive differences remain — not key order or reformatting. A list of paths is also shown: “items[3].price: 100 → 120”. Lists are compared by index, so a reordered array element is still shown as an edit — structure cannot help with that.

• Are the texts sent to the server?

  No. Everything is calculated in the browser, and the page makes no requests after loading. Docx and odt are unpacked right here, and pdf is parsed locally as well. “Link to result” embeds the texts in the URL itself after the hash — the browser does not send that part of the URL to the server, so it does not leave your device either.

• Why can't pdf sometimes be read?

  Because a pdf may contain no text at all: a scanned document contains an image, while files with complex embedded fonts store letters as codes that mean nothing without the font itself. In that case, the tool honestly says there is nothing to read instead of showing gibberish. The only workaround is to open the file in a viewer, copy the text, and paste it into the field.

• What does “moved” mean, and why is it there?

  A paragraph moved lower down is shown by ordinary line-by-line comparison as a deletion in one place and an addition in another — making it look as if half the document changed. Here, such sections are found by exact text match, marked in blue, and counted as matching in the similarity percentage: the text did not change, only its position did. If that gets in the way, clear “Move a section to another location”.

• How can I take the result with me?

  In four ways. “.patch” is a standard unified diff understood by patch, git apply, and any diff viewer. “Highlighted page” is a standalone html file with all styling included: it opens without a network connection or this site. “As a table” is Markdown markup for inserting into a task or email. “Link to result” is a URL that opens exactly the same comparison with the same settings for a colleague.