Draft
Two passages can share words for many reasons. They may quote the same source, use a required definition, describe the same method, or copy one another without acknowledgment. A similarity calculation finds a relationship between strings of words. It does not decide which explanation is correct.
What our comparison actually does
DraftLens splits both passages into words, normalizes letter case and apostrophes, and forms sequences of five consecutive words. These sequences are called shingles. Repeated instances of the same sequence count once in each passage.
The percentage is the number of unique shingles shared by both passages divided by the number of unique shingles present in either passage. This is Jaccard similarity. Commonplace sequences made entirely of a small set of stopwords are excluded.
For example, Text A is “Small gardens bring neighbors together every spring.” Text B is identical. Both contain the same set of five-word sequences, so the overlap is 100%. If Text B has no shared five-word sequence, the overlap is 0%, even if both passages discuss gardening.
Why paraphrases can disappear from the score
Change one word in the middle of a short passage and several five-word sequences change at once. A paraphrase can express nearly the same idea while sharing no complete shingle. Our method does not use embeddings, infer semantic equivalence, or identify unacknowledged ideas.
Conversely, required wording can produce a high overlap without wrongdoing. A laboratory procedure might need exact technical phrasing. Quotations, properly marked and cited, can legitimately match a source. The score does not know whether the writer had permission or fulfilled an assignment’s requirements.
A zero does not mean original
A pairwise tool sees only the two texts provided. It cannot compare your draft with a book you did not paste, a private database, an unpublished essay, or the rest of the web. It cannot return a reliable “plagiarism-free” certificate.
Even database-backed products are limited by their corpus and matching methods. Search coverage, publication access, and licensing differ. A displayed source match still needs a human to review quotation, attribution, and context.
Read the University of Oxford guidance on plagiarism for an example of how an institution frames attribution responsibilities. Follow your own institution’s rules, since policies and disciplinary processes vary.
Use overlap constructively
A useful comparison starts with a question. Did a revision remove accidental duplication? Does a summary reuse too much wording from a source? Are two versions still mostly identical? Paste the relevant source and your draft, inspect the highlighted phrases, and read the surrounding sentences.
If you are summarizing a source, first make sure you understand the claim. Write a summary from your notes, check it for fidelity, and add the appropriate citation. Merely swapping synonyms does not solve an attribution problem.
Protect sensitive material
The default comparison runs on your device and does not upload either passage. That helps when reviewing unpublished material, but you should still follow any confidentiality rules that govern the text. Downloaded reports contain the passages. Avoid sharing them more widely than the original documents allow.
Try the text similarity checker with identical passages, then change one sentence. Watching the shared phrases change is a useful way to understand the method. The result is a comparison between your two inputs, never a verdict on the writer.
Put it into practice.
Take a fresh look at your draft with the related local tool.
Open text similarity