Start free

AI detectors vs plagiarism checkers

Plagiarism checkers and AI detectors answer different questions. One matches your text against known sources, the other scores writing signals, and confusing the two leads to bad decisions.

What a plagiarism checker does

A plagiarism checker compares your text against a corpus of known sources: academic papers, web content, student submission archives, books. It returns matches: here is a 14-word string from your paper that appears verbatim on this URL. Turnitin, iThenticate, Grammarly's plagiarism tool, and Copyscape are all corpus-matching systems under the hood.Plagiarism checkers are deterministic. If the corpus contains your exact string, it is matched. If the string has been reworded, the match drops. They cannot flag ideas that came from somewhere, only text that appears somewhere.

What an AI detector does

An AI detector does not compare your text to anything specific. It computes statistical features such as sentence-length variation (burstiness), stock AI phrases, hedging and transition words, em-dash density, and repetition, then scores how likely those features are to have come from a language model. The reference is a probabilistic baseline, not a corpus. We break this down in our guide to how AI detectors work.An AI detector can flag text that has never been seen before. It can also flag human text that coincidentally matches the baseline. That is its core tradeoff, and why false positives are a real problem. A detector score is a signal, not proof, and it should never be the only basis for accusing anyone.

When to use which

Use a plagiarism checker when you want to know:
  • Did this student copy from another student's paper?
  • Does this blog post lift passages from competing blogs?
  • Is this research paper recycling material from an earlier publication?
Use an AI detector when you want to know:
  • Was this text drafted by an LLM like ChatGPT or Claude?
  • Does this submission match patterns of machine-generated writing?
  • Should I have a conversation about how this was produced?

Where they overlap (but should not)

Turnitin now offers both a classic plagiarism score and an AI writing score in the same report, which is useful but also confusing. The two scores measure different things and are calibrated differently. A 5% similarity score and a 70% AI indicator answer different questions.

A common failure mode: the "it passed plagiarism check" defense

Students and writers sometimes argue that because a plagiarism checker returned a low score, the piece could not be AI-written. That is a category error. Plagiarism checkers would not catch AI writing because the text was generated fresh, so it does not match any source. A low plagiarism score is perfectly consistent with 100% AI authorship.The reverse also happens: a detector flags a passage as AI, the writer says "but my Turnitin similarity was 0%" as if that settled it. Different tools, different questions.

What about Turnitin's combined workflow?

Turnitin's Originality product combines both checks because educators want a unified view. The workflow is reasonable: look at similarity and the AI indicator together, read the piece, then decide whether to open a conversation. Treating either score as proof is where things go wrong. We cover how Turnitin's AI indicator behaves in our breakdown of Turnitin's accuracy.

What this means for policies

If your school or company has a policy that references "plagiarism detection software" and treats AI output as plagiarism, the policy needs updating. AI-generated text is not plagiarism in the legal sense: it does not copy a specific copyrighted source (most of the time). It is its own category: undisclosed machine authorship. The policy should distinguish the two, and the tooling should too.For individual writers: if you are publishing AI-assisted work, disclosing AI use where policy requires it is the honest move. Our guide to citing AI use covers current standards.

How Leap fits

Leap offers a free AI detector and a humanizer, both running in your browser with no signup and nothing sent anywhere. We do not do plagiarism checking; that is a different corpus-matching problem we are not solving. If you need plagiarism detection, use Turnitin, iThenticate, or Copyscape. If you want to know whether text shows common AI writing signals, try the free Leap detector.

Frequently asked questions