How similarity is calculated on this page
This tool compares two passages using two complementary measurements: a phrase-level check (are there identical three-word sequences in both passages?) and a word-level check (do the two passages draw from the same overall vocabulary?). Combining both gives a more reliable picture than either alone — phrase matching catches close paraphrasing that reorders sentences but keeps chunks intact, while word-level overlap catches passages that share vocabulary and topic even when phrasing has been substantially reworded.
Both calculations run instantly in your browser using a technique called n-gram shingling, borrowed from classic information-retrieval and plagiarism-detection research, applied here at a scale and simplicity appropriate for a quick, free, client-side check.
The Jaccard index, explained simply
The core math behind both scores is the Jaccard index, a similarity measurement that's been used in information science since the early 1900s (originally for comparing plant distributions across regions, of all things). It's calculated as: the number of items two sets have in common, divided by the total number of unique items across both sets combined.
Applied to text, "items" can be individual words or, as this tool does for its main score, overlapping three-word phrases (trigrams). If Text A contains the trigrams "the quick brown," "quick brown fox," and "brown fox jumps," and Text B contains two of those same three trigrams plus two others, the Jaccard index for that pair is: 2 shared ÷ (3 + 4 - 2 shared) = 2/5 = 40%.
This tool blends a trigram-based Jaccard score (weighted 70%) with a simple word-level Jaccard score (weighted 30%) into the single percentage shown on the gauge, because phrase-level matching is generally the stronger signal for "did this get copied or closely paraphrased," while word-level overlap adds useful context about shared topic and vocabulary even when phrasing diverges completely.
What this tool is useful for
- Checking your own paraphrase. If you've rewritten a passage in your own words, run the original and your rewrite through this tool — a very high score suggests you're still tracking the original too closely.
- Comparing two AI outputs. If you generated the same prompt through two different models, or the same model twice, this tool shows exactly how much phrasing they share.
- Spotting duplicated content across your own site. If you maintain multiple tool pages or articles and want to check they're not accidentally too similar to each other, paste them in directly.
- Teaching what "close paraphrasing" looks like. The highlighted shared phrases make the abstract idea of "too similar" concrete and visible.
Why this isn't a plagiarism checker
It's worth being explicit about a common point of confusion: this tool has no access to the internet, no index of published books or articles, and no memory of text from other sessions. It can only ever tell you how similar the two specific passages you paste in are to each other. If you need to check whether a piece of writing matches something already published somewhere on the web, you need a dedicated plagiarism-detection service that crawls and indexes external content — this tool simply isn't built for that job, and we'd rather say so plainly than let you assume otherwise.
Why AI paraphrase tools often score higher than expected
If you've ever run text through an AI "paraphrasing" or "rewriting" tool and then compared the output to the original here, you may notice the similarity score is higher than you'd expect from something marketed as a full rewrite. This happens because many paraphrasing tools work by swapping individual words for synonyms while leaving sentence structure almost entirely intact — which registers as high similarity on the trigram comparison, since most three-word sequences around the swapped words are unchanged. Genuine rewriting — reordering ideas, changing sentence structure, combining or splitting sentences — reduces phrase-level similarity far more effectively than synonym substitution alone. This is the same principle behind why our AI Humanizer tool includes sentence-restructuring, not just word swaps.
How to read a similarity score responsibly
A similarity percentage describes overlap, not intent, and it can't tell you why two passages are similar — coincidence, shared source material, a common topic with limited ways to phrase it, or direct copying are all possible explanations for the same number. Use this tool to flag passages worth a closer manual look, not as a final judgment on its own, especially in any context where the result might inform an academic, editorial, or employment decision.
Frequently asked questions
Is this the same as a plagiarism checker?
No. A plagiarism checker searches billions of indexed web pages and academic papers to find where specific text may have originated. This tool only compares the two passages you paste in directly against each other — it has no external database and can't tell you whether either passage matches something published elsewhere.
What counts as a "high" similarity score?
As a rough guide: above 60% usually means the two passages share substantial wording and structure — a close paraphrase or near-duplicate. Between 25-60% usually reflects genuinely shared ideas with real rewording. Below 25% usually means the passages are largely independent, even if they cover the same topic.
Why do two independently written AI outputs sometimes score high?
Language models tend to converge on similar phrasing for similar prompts, especially for common topics, because they're drawing from overlapping patterns in training data. Two AI-generated answers to the same question often share more structure and phrasing than two human writers would, purely because of how the underlying models generate text.
Does word order matter in the comparison?
Yes, partially. The trigram (3-word phrase) comparison is order-sensitive — "the cat sat" and "sat the cat" are treated as different phrases. The word-overlap comparison ignores order entirely and just checks which individual words appear in both passages.