Updated · Built and checked by the PlinzoTools team, formulas tested against published sources
Calculate the similarity percentage between two texts using Jaccard similarity.
60.0%
This tool measures how similar two texts are using the Jaccard similarity coefficient (Jaccard, 1912) -- the ratio of shared words to total unique words across both texts.
Paste two texts. Their similarity percentage appears instantly.
Jaccard Similarity = |A ∩ B| / |A ∪ B|
Where A and B are the sets of unique words in each text.
Before comparing, each text is lowercased and split into words. A word here is any run of letters, digits, and apostrophes, so punctuation, line breaks, and capitalization never affect the result. Each text is then reduced to a set of unique words, which means repeating a word ten times counts the same as using it once. The tool does not remove common words such as "the" or "and", does not stem words, and does not understand synonyms. "Run" and "running" are two different words, and "car" and "automobile" are treated as completely unrelated.
Take Text A, "The quick brown fox jumps over the lazy dog", and Text B, "A quick brown fox leaps over a lazy dog". Text A has eight unique words: the, quick, brown, fox, jumps, over, lazy, dog. Text B also has eight: a, quick, brown, fox, leaps, over, lazy, dog. Six words appear in both (quick, brown, fox, over, lazy, dog). The union of the two sets is 8 + 8 - 6 = 10 words.
Similarity = 6 / 10 = 0.60, shown as 60%
Notice that the two sentences say almost the same thing, yet the score is only 60 percent. That is because the swapped verb and the different articles each add words that only one side has. Short texts swing wildly on a single word, so treat scores from very short inputs as rough.
Because common words like "the", "of", and "to" appear in nearly every English text, two unrelated paragraphs will rarely land at exactly zero. If you want to judge topical overlap, mentally discount the shared small words and look at whether the meaningful nouns and verbs match.
Keep the limits in mind. The method ignores word order, so a paragraph with its sentences shuffled would still score 100 percent against the original. It also ignores word frequency, so a long text and a short excerpt of it can score low simply because the long text contains many more unique words. For best results compare two texts of similar length. Finally, this is not a substitute for a plagiarism or academic-integrity check, which compares against large external sources. A high score here only tells you the two pieces you pasted overlap heavily.
Paste only the body text you care about. Navigation menus, bylines, and boilerplate footers add shared words that inflate similarity. If you are comparing a rewrite, paste the same section of both versions rather than the full documents, so the score reflects the part you actually edited. Numbers count as words too, so tables of figures that repeat between two documents will raise the score.
No -- this compares word sets, not sequences, so it measures shared vocabulary rather than exact phrasing or structure.
It's a similarity metric, not a plagiarism scanner (it doesn't check against any external database) -- useful for comparing two texts you already have, like checking if a rewrite has drifted too far from the original.
Any single word that appears in only one text lowers the score, including an extra "a", a typo, or a number. Even a one-word difference in a short text shifts the percentage noticeably.
No. Both texts are lowercased and punctuation is ignored, so "Hello, World!" and "hello world" contain the same word set.
If only one text has words, the shared count is zero and the score is 0 percent. If both boxes are empty, the tool shows a prompt asking for text instead of a score.
The word splitter only recognizes the letters a to z, digits, and apostrophes. Accented or non-Latin characters break words apart, so results for other languages can be unreliable.
Reference: Google Search Central