Plagiarism Checker

It does not compare your text against the internet

The most important thing here, and the tool says it too.

Real plagiarism detection works by comparison against a corpus — billions of web pages, journal archives, and in the case of academic services, a database of previously submitted student work. It reports matches: this phrase appears in that source. The evidence is the match, and you can go and look at it.

This tool has no corpus at all. It asks a language model for an impression of your text. So it can tell you that a passage reads as generic, or formulaic, or unlike the rest of your writing — but it cannot know whether your words exist anywhere else, and it cannot show you a source, because it never looked for one.

That has a specific consequence worth stating plainly: a high originality score here is not evidence that your work is original. You could paste in a famous passage word for word and get a reassuring number, because nothing checked. If you need to know whether your text matches published material, you need a service with a database — Turnitin, Copyscape or iThenticate, which the tool's own warning already names.

The AI likelihood score, and why it deserves real caution

The second number is the one that can cause actual harm, so it gets a proper section.

AI-detection scores are unreliable in a way that is well documented, and this is not a caveat invented for this page. Researchers have found that detectors produce substantial false positives; a peer-reviewed assessment concluded that AI writing detectors are ineffective, unreliable and harmful; and universities including UCLA have published guidance advising caution about using them at all. OpenAI withdrew its own AI text classifier because its accuracy was too low to be useful.

And the false positives are not random. They fall hardest on non-native English writers. Stanford researchers found detectors flagged writing by non-native speakers as AI-generated at dramatically higher rates than writing by native speakers — because detectors partly key on simpler vocabulary and more predictable sentence structure, which is exactly what careful second-language writing looks like. International students have been accused of cheating on this basis. If you write English as a second language, a frightening score here tells you about the detector, not about you.

Formulaic writing gets flagged too, for the same reason — technical documentation, a lab report, a legal summary, anything written to a template reads as predictable because it is supposed to.

So three rules:

Never treat a score as evidence. Not from this tool, not from any detector. There is no threshold that means someone cheated.

Never accuse anyone on the basis of one. If you are a teacher or an editor, a detector score is not grounds for an allegation. A conversation about the work, or asking to see drafts and version history, tells you far more.

If your own honest work is flagged, do not panic and do not rewrite it into something worse. The most useful protection is evidence of process — draft history, a document with version history, notes, an outline. That is real evidence, where a score is not.

What it is genuinely useful for

Having said all that, there is real value here, and it is in the parts that are not numbers.

The writing analysis and self-similarity checks give you feedback on your own text: passages that repeat each other, sections that read as generic or padded, a tone that shifts partway through, sentences doing no work. That is ordinary editorial feedback and it is useful regardless of any score.

Where it earns its place:

Finding your own padding. Passages flagged as generic usually are. That is the paragraph to cut or to replace with something specific.

Spotting an inconsistent voice across a long document, or across sections written weeks apart, or in a piece several people contributed to.

Checking a draft you rewrote from a source — if a passage still reads as unlike your own writing, you may have paraphrased too closely, and that is worth knowing before submission even though this cannot confirm it.

Reviewing AI-assisted text you intend to make your own. If you drafted with a model and rewrote it, a section still reading as generated is a section you did not really rewrite.

Read the two percentages as a reader's impression expressed as a number — which is all they are. The prose feedback is worth more than either of them.

If you are a student

A short, direct word, because most people on this page are.

Use this to improve your writing, not to test whether you will get away with something. No free tool can tell you what your institution's checker will report — it uses a database of submitted work that nothing else has access to, so a good score here predicts nothing about the result there.

The reliable protection against a plagiarism finding is method, not a checker: take notes in your own words rather than copying and intending to reword later, keep track of where every fact came from, quote and cite when you use someone's phrasing, and write from your notes rather than with the source open beside you. Close paraphrasing of an open source is how most accidental plagiarism happens.

And if you have used an AI tool, check what your institution actually permits — the rules vary enormously and are often more permissive about assistance than students assume, while being strict about disclosure. Ask rather than guess.

Your text is sent to a language model for analysis, so do not paste anything confidential or unpublished that you need to keep private. There is no account, and the limit is 30 checks an hour.

Related tools

Grammar Fixer fixes what is actually wrong, which is usually more useful than any score. Paraphraser rewrites a passage, though note that paraphrasing a source closely is still plagiarism however the words change. AI Humanizer targets text that reads as machine-written, and its page is honest about what that does and does not achieve.