Text Similarity Checker

How the two scores work

Cosine similarity turns each text into a vector of word frequencies and measures the angle between them. Identical texts score 100%; texts sharing no vocabulary score 0%. Because it weighs frequency, a document that mentions a term twenty times is treated as meaningfully different from one mentioning it once.

Jaccard similarity is simpler: the number of unique words shared by both texts divided by the total number of unique words across both. It ignores frequency entirely, so it answers “how much vocabulary do these have in common” rather than “are these written similarly”.

Both are word-level measures. Neither understands meaning, which is the important limitation — two texts saying exactly the same thing in completely different words will score low, and that is a correct result for what these metrics measure, not a bug.

Reading the numbers

• Above 80% cosine — heavy overlap. If these are meant to be separate pages, they probably are not.
• 50–80% — significant overlap. Worth checking whether they target the same search intent.
• 20–50% — normal for two pages on related topics within the same site.
• Below 20% — substantially different content.

What this can and cannot tell you about duplicate content

Google does not publish a similarity threshold, and it does not evaluate duplication the way this tool does. Its systems assess meaning and context, not word-frequency vectors, and it considers duplication across the whole web rather than between two texts you happen to paste.

That means a high score here is a useful warning, not a verdict. It tells you two pieces of text are built from largely the same words, which is worth investigating — but plenty of legitimately distinct pages score high because they cover the same narrow topic with the same necessary vocabulary.

Equally, a low score is not proof of originality. Text rewritten by a spinner scores low on word overlap while remaining functionally the same content, which is exactly the case these metrics are worst at detecting.

Where it is genuinely useful

• Checking whether a rewrite actually changed anything, rather than shuffling synonyms.
• Comparing two of your own pages that seem to target the same query, before deciding to merge them.
• Checking a supplier's product description against the manufacturer's original, which many competitors will also be using verbatim.
• Comparing a draft against a competitor's page to see whether you have written something genuinely different.
• Auditing template-generated pages, such as location pages, where only a place name changes between them.

When high similarity is a real problem

The clearest case is two pages on your own site competing for the same query. Search engines pick one, your internal links and authority split across both, and neither performs as well as a single consolidated page would. This is keyword cannibalisation, and the fix is usually to merge and redirect rather than to reword.

The second case is duplicate URLs — the same content reachable at several addresses. That is not a writing problem and rewording will not help; it needs a canonical tag or a redirect.

Syndicated or supplier-provided content is the third. If your product descriptions are the manufacturer's text, so are your competitors', and none of you has a differentiating page. Rewriting genuinely helps here.

Related checks

For repetition inside a single document rather than between two, the Duplicate Sentence Finder and Duplicate Paragraph Finder catch lines and blocks repeated word-for-word — common in content assembled from several sources.

If you want to see precisely which words changed between two versions rather than a single score, the Text Diff Checker highlights every addition and removal.

Frequently asked questions

Related tools