Paste two passages and the tool scores how alike they are, from 0 to 100%, then shows which words the two share. It is a quick way to see whether a rewrite kept the original's wording, whether two articles cover the same ground, or whether a submitted text is close to something you already published.
The headline figure compares overlapping three-character sequences. Underneath it, a second figure compares unique words. The two disagree often, and understanding why is most of the work of using this tool well.
Two scores, two measures
The headline similarity is a Dice coefficient over character trigrams. Each passage is cut into every three-character run it contains, the two sets are compared, and the score is twice the number of shared runs divided by the total number of runs in both. That is why the result is sensitive to spelling and word order: changing a letter changes three or four trigrams at once.
Word overlap is a Jaccard index over unique words. It ignores order and repetition, so two passages built from the same vocabulary but arranged into different sentences can score high on words and lower on sequences.
The quick brown fox jumps over the lazy dog.
B:
The quick brown fox leaps over the lazy dog.
Similarity:
88% Word overlap: 78%
Shared words:
7 Shared sequences: 36One word changed out of nine drops the sequence score by 12 points. The word score falls further, because "jumps" and "leaps" are both removed from the shared set and the union grows.
The four rules, and what they hide
| Rule | Effect | Affects |
|---|---|---|
| Ignore case | Lowercases both passages first | Both scores |
| Ignore punctuation | Replaces every non-letter, non-digit with a space | Both scores |
| Ignore extra spaces | Collapses runs of whitespace and trims the ends | Both scores |
| Ignore common words | Removes a list of 96 function words before comparing vocabulary | Word overlap only |
The first three rules change the text before either score is calculated, so they raise both figures together. The fourth is different: it only filters the word list, so the headline similarity does not move at all. It is useful when two passages share nothing but "the", "of" and "and", which is the usual reason word overlap looks higher than it should.
This is where a 100% result needs reading carefully:
Hello world
B:
hello, world!
No rules:
50% similarity
Case, punctuation and spaces ignored:
100%Nothing was edited between those two versions except capitalisation and a comma. The 50% is the honest character-level answer; the 100% is the answer to a different question — "are these the same words once I stop caring about formatting?" Both are correct, and the rules you leave on decide which one you get.
Reading the label
The status line names a band rather than leaving you with a bare number:
- 95% and above — Nearly identical. Expect a copy, a formatting variant or a trivially edited version.
- 70 to 94% — Very similar. The status turns amber. At this level the two texts are usually the same passage with substitutions.
- 40 to 69% — Some overlap. Shared subject or shared phrasing, but a genuinely different text.
- 15 to 39% — Mostly different. A few phrases in common, probably topic vocabulary.
- Below 15% — Little in common. Two unrelated passages, or two texts on the same subject written from scratch.
These bands are conventions chosen for this tool, not industry standards. Treat the amber warning at 70% as a prompt to look at the shared-word list rather than as a verdict.
The shared-word list
Up to ten words appear with a count, showing how many times the word occurs in both passages and the two totals side by side — "A 2 / B 2". Words are ordered by how often they appear in both, then alphabetically. That list is often more informative than the score: a run of content words in common means real overlap, while a run of "the" and "of" means nothing at all.
The count shown is the smaller of the two frequencies. If version A uses "library" four times and version B once, the list says "used 1 time in both". That is deliberate — it describes the shared ground rather than the total usage.
This is not a plagiarism check
The tool compares two texts you supply. It cannot search the web, does not know where either passage came from, and cannot tell whether a match is copying, a quotation, a shared definition or two people describing the same process in the same words. Its own warning says as much.
That limits what a high score means. Two safety manuals will score highly on procedure language without either author having seen the other. Two news reports of the same press release will share sentences because they share a source. And a genuine copy with every word swapped for a synonym can score low while being an obvious derivative to a human reader.
Use it as a first pass on a specific pair of texts — a submitted article against your published one, a rewrite against its original, two drafts of the same section — and then read both passages yourself. For a web-wide check, you need a search-based service, not a comparison tool.
Limits
- No word-level highlighting. You get scores and a word list, not a marked-up diff. For that, use the text comparison tool.
- Order matters to the headline score. Two passages using identical sentences in a different order will score lower on sequences than on words.
- Short texts are noisy. A five-word phrase has very few trigrams, so a single changed word can swing the score by 30 points or more.
- Accents and scripts are fine. Letters and digits from any script are kept; everything else is treated as a separator when punctuation is ignored.
- Nothing is uploaded. Both passages are compared in the tab and a reload clears them.
Run the comparison twice: once with every rule on, once with them all off. The gap between the two scores tells you how much of the similarity is formatting and how much is wording.