2 URLs to compare
Similarity
—
—
0
Words URL 1
0
Words URL 2
0
Shingles URL 1
0
Shingles URL 2
Paste 2 URLs: the tool fetches both pages, extracts editorial text, compares via 4-word shingles. To check whether content was copied.
The tool tokenizes the text of both pages (chrome stripped, lowercase, accents removed), then generates shingles (sequences of 4 consecutive words). The Jaccard score is the ratio of common shingles / total shingles. ≥ 85% = identical (pure copy-paste). 50-85% = near-duplicate (recycled paragraphs). 20-50% = derived (paraphrase, shared topic). < 20% = original.
The 4-word shingle is the academic standard for duplicate detection (Broder 1997 paper). Too short (2-3 words) and you catch noise (« the », « and the »). Too long (8-10 words) and a light paraphrase breaks all shingles. 4-grams capture full or near-full sentences — exactly what copy-paste produces.
The tool only detects word-for-word duplication (or near). A FR→EN translation won't be flagged — by design, otherwise we'd false-positive on every multilingual site. To detect translation-copy, you need a multilingual embedding approach, not feasible in pure JS without a backend.
Often no. On a niche topic, two articles will cite the same facts, sources — 30% shingle similarity is coherent. The red flag is above 60%: it means the sentence structures themselves are recycled, not just the facts. If you see a score > 70% on an article you know is original, check the shared shingles — often boilerplate (signature, footer, CTA) polluting the score.
When we write an original article on the Nautilinks network, we write it once and publish it exclusively on one network media. No syndication, no duplication.