Paraphrase Detector

Jaccard similarity = shared items / total distinct items. For paraphrase detection, items are bigrams (word pairs) or trigrams (word triples). Paraphrase keeps meaning but changes words — bigrams/trigrams catch shared phrases even when surrounding words differ. A trigram Jaccard score above 0.25 is a strong signal of paraphrase.

Plagiarism is presenting someone else's work as your own — including verbatim copying, near-exact copying, and uncredited paraphrase. Paraphrase itself is legitimate — the problem is uncredited paraphrase, where you restate someone's idea without citing the source. This tool finds where two texts say the same thing differently, signalling that citation may be required.

Jaccard bigram/trigram thresholds: 0–0.1 (different topics), 0.1–0.2 (same topic, different treatment), 0.2–0.4 (likely paraphrase), 0.4–0.6 (heavy paraphrase or near-copy), 0.6+ (very close to verbatim). Short texts and domain-specific vocabulary will score higher even without paraphrase.

Yes — paste your original left, AI output right. A high trigram score means shared phrase structure without significant rewriting. Note: LLMs naturally vary sentence structure, so a low score doesn't guarantee no paraphrase — it means different phrases, not necessarily different meaning.