How to compare two texts for similarity
- Paste the first text on the left and the second on the right: two essays, product descriptions, an AI answer and a source, or two versions of a document.
- The overall similarity score combines three classic measures, shown separately: word overlap, character sequence overlap and edit similarity.
- Shared three-word phrases are listed, which helps spot copied passages and near-duplicate content.
How the scores work
Cosine similarity treats each text as a "bag of words" and measures how closely their word counts point in the same direction; 100% means the same words in the same proportions, regardless of order. Ignoring common words and matching word forms makes it focus on content. Jaccard similarity compares the sets of three-character sequences, which catches small spelling changes and works for any language. Edit similarity is based on Levenshtein distance, the number of single-character insertions, deletions and substitutions needed to turn one text into the other, so it is sensitive to word order.
These are lexical measures: they compare the words used, not the meaning. "The film was great" and "I loved the movie" score low, even though they say the same thing. Semantic similarity needs an embedding model, which would mean sending your text to an AI service; this page keeps everything in your browser. Lexical scores are excellent for detecting duplicates, reused content, small edits and close paraphrases, and for checking whether a model's answer quotes its source. For line-by-line changes, use our Diff Checker.
Is this a plagiarism checker?
It compares two texts you provide. It does not search the web, but shared phrases make copied passages easy to spot.
What score means duplicate content?
Above about 85% the texts are nearly identical. 60–85% usually means light rewording of the same text.
Does it understand meaning?
No. It compares words and characters, so different wording with the same meaning scores lower.