Free Text Similarity Checker – Compare Two Texts Online

How Text Similarity Is Measured and What Each Score Means

A text similarity checker takes two pieces of writing and tells you how much they overlap, expressed as a number rather than a vague impression. Paste text into the left box and text into the right box, and you get three separate scores plus a visual map of the phrases they share. The three numbers exist because there is no single correct definition of similarity: two documents can share almost every word yet be arranged completely differently, or share a rare exact phrase while differing everywhere else. Editors comparing two drafts, teachers checking whether two submissions look suspiciously alike, and writers making sure a rewrite is genuinely different all need a way to see overlap that a quick read misses. The first score is word-set Jaccard similarity. Both texts are lowercased and split into a set of unique words, then the tool counts the words that appear in both sets and divides by the count of words that appear in either. A result of 1 means identical vocabulary and 0 means no shared words at all. The second score is cosine similarity on term-frequency vectors. Here word counts matter, not just presence, so a text that repeats a word ten times is treated differently from one that uses it once. The two texts become vectors of word frequencies, and the cosine of the angle between those vectors becomes the score. The third score is 3-gram shingle overlap. The text is broken into overlapping runs of three consecutive words, and the tool measures how many of those short sequences are shared. Because shingles preserve word order, this catches reordering that fools the word-set approach. Work through a concrete pair. Text A is the cat sat on the mat and Text B is the dog sat on the rug. The unique words in A are the, cat, sat, on, mat and in B are the, dog, sat, on, rug. The shared words are the, sat and on, which is three. The combined vocabulary is the, cat, sat, on, mat, dog, rug, which is seven. Jaccard is therefore 3 divided by 7, roughly 0.43. Cosine will read higher because the repeated word the appears in both and pulls the frequency vectors closer together. The 3-gram overlap will be lower still, because the sat on and sat on the are shared but the surrounding shingles differ. Three scores, one input pair, three honest angles on the same question. Specific situations make the three numbers useful together. A journal editor checks a resubmitted manuscript against the original and sees Jaccard near 0.95, confirming the author changed almost nothing despite claiming a rewrite. A content team compares a draft against a competitor page and reads a low 3-gram overlap as reassurance that no sentence-level copying happened. A student uses it to confirm that a paraphrase of a source shares meaning but not exact phrasing before submitting. The matched-phrase highlighter paints shared runs in colour on both sides, and the longest common substring is reported separately, so a single lifted sentence stands out even when the overall percentages look modest. The scores are directional signals, not verdicts, and reading all three at once is the point. One critical honesty note belongs here: this is not a plagiarism checker. A plagiarism checker compares your text against a live index of billions of web pages and published works, which requires a server and a crawled database. This tool only compares the two texts you paste against each other, so it can never tell you whether either one was copied from somewhere on the internet. It is built for comparing documents you already have, not for detecting copying from unknown sources. Everything runs locally in your browser; neither text is uploaded, stored, or logged, which makes it safe for unpublished manuscripts and confidential documents.

🔥 Trending Tools