Plagiarism Checker

A plagiarism checker compares a submitted document against a corpus — indexed web pages, academic databases, previously submitted work — and reports passages that match closely enough to warrant a look, usually as a similarity percentage plus a list of sources. It does not detect plagiarism. It detects textual overlap, which is a different thing: a correctly quoted and cited passage produces overlap, and a properly paraphrased uncredited argument may produce almost none. The output is a starting point for a human judgement about attribution, and treating the percentage as a verdict is the most common misuse. How matching works: the document is split into overlapping fragments, each fragment is fingerprinted, and the fingerprints are looked up against an index. This is why the corpus matters more than the algorithm — a checker can only find what it has indexed, so a passage lifted from a paywalled source, a PDF nobody crawled, or another student's unsubmitted draft is invisible regardless of how good the matcher is. It is also why two tools disagree on the same document: they are searching different libraries. In content operations the practical use is narrower than in education. An editor running a check on a freelancer's submission is looking for one specific failure — copy pasted from a competitor or from the client's own site — and the threshold that matters is not a percentage but whether any single continuous passage matches, because scattered common phrasing is normal and a matching paragraph is not. The AI-writing angle has changed the picture in a way worth being clear about. Generated text is usually novel at the string level, so it passes a plagiarism check comfortably while potentially reproducing an argument, a structure or a factual error from its training data without attribution. Plagiarism detection and AI-text detection are therefore separate products solving separate problems, and passing the first says nothing about the second. Meanwhile paraphrasing tools reliably reduce measured overlap without changing where the ideas came from — which means a low similarity score is evidence about the words and no evidence at all about the honesty of the work.