How Text Similarity Detection Works
Detecting how similar two pieces of text are sounds simple but has surprising depth. This tool uses a SimHash-style approach: it normalizes text, extracts word and word-pair features, converts those features into 64-bit fingerprints, and compares the fingerprints with Hamming distance. Similar texts tend to produce similar fingerprints, so the result is useful for near-duplicate and revision checks.
What This Tool Calculates
| Metric | What it measures | Best for |
|---|
| Similarity score | Percentage derived from 64-bit fingerprint overlap | Quick near-duplicate assessment |
| Hamming distance | Number of different bit positions between fingerprints | Technical comparison where lower means more similar |
| Fingerprint | A compact hexadecimal representation of the text features | Comparing and recording duplicate checks |
| Text stats | Words, characters, sentences, and extracted features | Checking whether two inputs are comparable in length |
| Sample and swap controls | Load examples, swap inputs, or clear the comparison | Fast editorial QA workflow |
Similarity Score Interpretation
- 0-20%: Different topics. Low overlap is normal for unrelated content.
- 20-40%: Same domain, different angles. Common between articles on the same topic.
- 40-60%: Substantial overlap. Investigate — could be intentional (revisions, translations) or accidental (cannibalization).
- 60-80%: High duplication. If this isn't intentional, consolidate into one piece.
- 80-95%: Near-duplicate. Likely a revision or republication. Set canonical tags appropriately.
- 95-100%: Essentially identical. Check for copy-paste mistakes or scraped content.
SEO Content Cannibalization
Cannibalization happens when two of your own pages target the same query and compete for rankings. Both rank lower than one combined page would. Symptoms:
- Two URLs from your site appearing for the same query in Search Console
- Rankings fluctuating between the two URLs for the same query
- Neither URL ranking on page 1, when individually they should
- Click-through rates split between the two URLs
Fix: compare the two pages with this tool. If similarity is high, review the overlapping sections manually and decide whether to consolidate, rewrite one page for a different intent, or adjust internal links and canonicals. If similarity is low but both pages target the same query, the issue may be keyword targeting rather than duplicated wording.
Known-Source Plagiarism Check
- Provide both texts: paste the submitted text and the suspected source text.
- Check the similarity score: high similarity suggests a close relationship between the two inputs.
- Review manually: this page does not highlight passages, so inspect the sections that look suspicious.
- Use a full plagiarism service if needed: this tool does not search external web, academic, or private databases.
- Document the result: copy fingerprints or notes if you need an editorial QA record.
When Similarity Is OK / Expected
- Templates: product pages, location pages often share boilerplate; differentiate the unique parts
- Legal pages: Terms of Service, Privacy Policy across sites often borrow from common templates
- Translations: the same content in different languages will have high cross-language similarity
- Standardized sections: author bios, footers, "about us" copy
- Revisions: v1 → v2 of same article will have high similarity by design
- Quotation / attribution: citing other sources naturally creates overlap
High similarity isn't automatically a problem — context matters. Use the score plus a manual review to make decisions.
Privacy
All comparison runs in your browser. Neither text is sent to any server. Safe for proprietary content, confidential documents, or student work where privacy is important.
