Plagiarism Similarity Checker

Compares two pieces of text you supply and reports how much of one appears in the other, with the matching runs highlighted. It has no web index and no document database, which is the first thing to understand about it — the honest description is a similarity measure between two texts, not a search of the internet.

What a Two-Text Comparison Can and Cannot Tell You

The word plagiarism suggests a search: paste an essay and find out whether any of it came from somewhere else. That requires an index of the somewhere else, which is a large piece of infrastructure and a legal minefield, and no browser-based tool has it. What can be done entirely locally is the other half of the job — deciding whether two texts you already have in hand share material, and if so, exactly which passages.

• Both boxes are tokenised and compared using overlapping word sequences. A sequence of the chosen length that appears in both texts is a match, and the tool looks for runs of consecutive matching sequences rather than isolated ones, so a shared phrase is reported as a phrase rather than as a scatter of individual hits.
• Three sensitivity settings control the length of the sequence being matched, and the naming is worth reading carefully because it inverts what people expect. High means four-word sequences, Mid means six, and Low means eight; the default is High. A shorter sequence matches more easily, so High sensitivity finds more overlap and produces more false positives on ordinary shared phrasing.
• The output is an overlap percentage plus the specific matching passages shown in context, which is the part that matters. A percentage tells you there is a problem; the highlighted runs tell you whether it is a copied paragraph or the phrase "in conclusion, the evidence suggests".
• Longer matching runs are grouped and ranked so the most substantial overlap surfaces first, which is what you want when one text has lifted a section and otherwise shares only common idioms.
• Each text needs at least as many words as the sequence length, and the tool says so plainly rather than returning a meaningless zero.
• Flagged passages can be copied out individually, and an optional paraphrase suggestion exists for a phrase — note that this final feature is the only part of the page that can call an AI provider, and everything else runs regardless.

Setting the Sensitivity So the Result Means Something

1. Put the text you suspect is derived in the first box and the text you believe it derives from in the second. The comparison is directional in its presentation — it walks the first text looking for runs that exist in the second — so getting the order right makes the highlights easier to read.
2. Start at High, the four-word setting, if you are looking for any shared material at all. Expect noise: four words is short enough that common constructions match.
3. Raise to Mid or Low when High returns a scattered mess of short highlights. Eight-word matches are almost impossible to produce by coincidence in normal prose, so what survives at Low is a genuine finding.
4. Read the longest runs first. A single eight-word match is a stronger signal than twenty four-word matches, because it cannot be explained by shared vocabulary.
5. Treat every match as a question rather than a conclusion. The tool cannot see who wrote what first, it can only see that the same words appear in both.
6. If a flagged phrase needs rewriting, copy it out and revise it yourself. Letting a paraphrase tool reword it usually produces something that still reads as a rewrite of the original, which is a worse position than a properly reworked sentence.

Three Ways to Check Overlap, and What Each Costs

Alternative	Notes
Two-text n-gram comparison (this tool)	Answers whether these two specific documents share material, and shows you where. Runs instantly, works with the network off, and the text never moves. Blind to a third document you did not supply — if the passage came from somewhere else, this cannot know.
A hosted checker with a web index	Answers whether the text appears on the public web. Requires uploading the document in full, submitting to a stored copy, and trusting someone else's retention policy. The index is also incomplete by design: paywalled articles, books and private repositories are largely absent, so a clean result is not a clean result.
Institutional systems with submission archives	The strongest option for academic work, because comparing against every previously submitted paper is the comparison that actually matters there. Only available through the institution, and the submission becomes part of the archive.
Reading both documents yourself	The only method with no false positives. Slow and unworkable past a few thousand words, which is why the n-gram pass exists — not to replace the reading but to point at the paragraphs worth reading twice.

The Same Two Texts at Three Settings

Consider a draft that has lifted one paragraph from a source article and otherwise reworded everything.

• At High, the four-word setting, the lifted paragraph shows as one long highlight — good — but the surrounding reworded text also lights up wherever four-word constructions coincide, including stock phrases and any technical term of four words or more. The output is accurate and hard to read.
• At Mid, the six-word setting, most of the coincidental matches disappear and the lifted paragraph remains, now broken into the runs that genuinely survive. This is usually the setting that gives a usable answer on prose.
• At Low, the eight-word setting, only the substantial portion of the lifted paragraph survives. For a document where you want to know whether anything was copied at all, this setting answers the question directly and ignores everything else.
• A shared technical term of eight words is the main false positive at Low — a long standardised definition will match across two documents that are otherwise independent. Seeing a single long match in an otherwise clean result is worth checking against the possibility of a standard formulation rather than assuming copying.
• Running all three settings and comparing is a legitimate workflow: the answer at Low is what you act on, and the difference between High and Low tells you how much of the overlap is ordinary shared language.

Why There Is No Search Button

A checker that searched the web would have to send your document to a server, keep it somewhere while it compared, and rely on a crawl that misses most of the material people actually plagiarise. This tool takes the opposite position: it compares the two texts you give it, in the page, and reports what it finds. The consequences are that it is instant, it works offline, the text is never transmitted, and it can only answer questions about documents you already have.

• Comparison is on word sequences after tokenisation, so punctuation and capitalisation differences do not defeat a match — which is what lets it survive light editing.
• Longer runs are reported as single passages, not as overlapping fragments, so the highlight you see corresponds to a passage someone could have copied.
• There is no corpus, no index and no persistent storage: closing the tab discards everything.
• Because a paraphrase changes word order rather than only wording at the token level, a thoroughly rewritten passage will not match. That is a limitation of sequence matching and not something a threshold setting can fix.

Questions About the Comparison

Does this check my text against the internet?

No, and no browser-based tool can. It compares the two texts you provide and reports the overlap between them. If your question is whether a passage exists somewhere else on the web, you need a service with an index and you need to accept uploading the document — this tool answers a different question, which is whether two documents in front of you share material, and answers it without anything leaving your machine.

Which sensitivity setting should I use?

Start at High, which is the default and matches four-word sequences, if you want to see every possible overlap. Move to Mid or Low when the result is too noisy to read. Low matches eight-word sequences and is the setting to quote from, because a run of eight identical words in normal prose is very unlikely to be coincidence. If Low returns nothing, the documents may still share reworded material, which sequence matching cannot see.

Why does it highlight a phrase that is just a common expression?

Because it is matching sequences of words without understanding them, and a long enough common expression — a standard definition, a fixed technical term, a stock transition — is a sequence like any other. This is why the sensitivity matters and why the flag is a question rather than a verdict. Reading the highlighted run takes a second and settles it immediately in most cases.

Can it detect text that has been paraphrased?

No. The comparison is on word sequences, so reordering and substituting wording defeats it — which mirrors a real limitation of every n-gram method. A passage that has been genuinely rewritten is, in the useful sense, no longer the same passage, and deciding whether a rewrite is legitimate is a judgement about the work rather than a string comparison. What this tool reliably catches is direct reuse.

Last verified: October 3, 2026

Launch Plagiarism Similarity Checker

All Tools · AI Tools · About · How We Test · Contact · Privacy

Maintained by Huang Bin · Last updated October 3, 2026