A plagiarism checker looks for text that already exists somewhere else. An AI detector looks for writing patterns that resemble machine-generated text. They answer different questions, so a clean result from one says almost nothing about the other. An essay can be entirely original to a plagiarism checker and still be written by a model, and an essay can be fully human and still match published sources.
Key takeaways
- A plagiarism checker compares your text to a database of web pages, publications and, in some products, previously submitted work. It reports matches, not intent.
- An AI detector is a classifier. It labels passages by how much they resemble AI writing. It does not search for sources, and it does not know who wrote the text.
- Text generated by a model is usually new wording, so a plagiarism checker will often find nothing. Copied human writing has no machine patterns, so an AI detector will often find nothing.
- Both can mislead if you read the number as a verdict. Both need a person to look at the highlighted passages.
What does a plagiarism checker actually do?
A plagiarism checker is a text-matching system. It breaks your document into phrases, compares them against a large collection of material, and shows you where the wording overlaps with something it already holds.
Turnitin’s Similarity Report is the best-known example in education. A university guide describing it says the report compares an assignment against “a database of web pages, academic books and articles, as well as other students’ papers,” and that the percentage reflects how much of the writing matches earlier sources (University of Waterloo, Turnitin and iThenticate). The same page says “there is no ‘safe’ colour or percentage” and that no percentage can fully evaluate whether text has been plagiarized, because the tool cannot tell a proper citation or common phrase from copying.
Grammarly’s plagiarism checker works the same way in principle. Its support documentation says it scans a document against “databases, academic papers, websites, and published works” and highlights the parts that match external sources, suggesting citation formats when it finds a match (Grammarly support).
Two things follow from this design. A plagiarism checker is only as good as the collection it searches: a source that is not in the database cannot be matched. And a match is a match, whether it is a quoted and cited passage, a stock phrase, or lifted text. Someone has to read the highlights.
What does an AI detector actually do?
An AI detector does not search anywhere. It reads the text you give it and estimates, from patterns it learned during training, whether passages look like human writing or like model output. The result is a statistical judgment about resemblance.
On this site, the checker uses the Pangram Labs API. It labels passages as AI-Generated, AI-Assisted or Human Written, with a confidence level, and it returns labels without reasons. The headline figure is the share of text flagged as AI-like, which is not a probability that a person did or did not write the whole thing. What an AI detection percentage means explains that distinction in detail, and how AI detection works covers the general approach.
Because there is no source to point to, an AI detector cannot show you proof. It can only show you which passages looked unusual to the classifier. That is why it can be wrong in both directions, as covered in can AI detectors be wrong.
How do the two compare side by side?
| Question | Plagiarism checker | AI detector |
|---|---|---|
| Detects copied text? | Yes, when the source is in its database | No |
| Detects generated patterns? | No | Yes, as a statistical estimate |
| Checks source matches? | Yes, and usually links or names them | No, there are no sources to match |
| Produces a score, and what does it mean? | A similarity percentage: how much of the text overlaps with indexed material | A share of text labeled AI-like: how much of the text resembles AI writing |
| Can produce false positives? | Yes. Properly quoted text, references and common phrases count as matches | Yes. Formal, short, translated or non-native writing can be labeled AI-like |
Read the last row carefully. A false positive from a plagiarism checker means “this overlaps with existing text, but it is not misconduct.” A false positive from an AI detector means “this looks machine-written, but a person wrote it.” The first can usually be checked by reading the source. The second cannot be checked against anything outside the classifier, which is a reason to be more careful with it, not less.
Why can’t one replace the other?
Each tool is blind to what the other one sees.
A plagiarism checker misses newly generated text. A language model writes new sentences rather than copying a page. The result may overlap with nothing in a database, so a similarity report can come back low. An AI detection vendor’s help page makes the same point about the two categories: plagiarism detection misses original text written by AI, and AI detection does not catch human-written content copied from elsewhere (Winston AI help center).
An AI detector misses copied human writing. If someone pastes a paragraph from a journal article, the paragraph reads like an expert wrote it, because one did. There is nothing machine-like in it to find.
The two scores are separate even inside one product. Turnitin added an AI writing indicator to its Similarity Report, but the University of Queensland’s guide treats them as different capabilities: text matching produces the Similarity Report, while the AI indicator scans for text likely generated by an AI tool (University of Queensland). A low similarity percentage tells you nothing about the AI indicator, and the reverse is also true.
Neither tool checks whether the claims in a text are true. AI-written text can include invented facts or citations, and a similarity report or an AI label will not tell you which ones. Checking sources and claims by hand remains necessary.
When should you use each one?
Use a plagiarism checker when the question is “did this text come from somewhere?” That covers a student checking citations before submitting, an editor screening a freelance article, or a publisher checking a manuscript for reused passages.
Use an AI detector when the question is “does this look machine-generated?” That covers an editor reviewing a commissioned piece, a teacher who wants a second look at an essay that does not match a student’s earlier work, or a writer checking how their own edited draft reads.
Use both when the stakes are real and you need more than one signal, such as a high-stakes assignment or a publication decision. Then add what neither can supply: drafts, version history, a conversation with the writer, and a check of the sources.
A rough decision flow:
- Worried about copied or uncited material? Run a plagiarism checker and read each match against its source.
- Worried about machine-written text? Run an AI detector and read the flagged passages, not just the total. How to tell if text is AI-generated lists checks that carry more weight than any score.
- Both worries? Run both, treat them as two unrelated signals, and do not average them or let one explain the other.
- A result you are about to act on? Stop and gather other evidence first.
What do the vendors themselves say about reliability?
Turnitin, which sells both capabilities, is explicit that its AI indicator is not a verdict. Its published statement says it aims for a “less than 1% false positive rate” and adds: “Given that our false positive rate is not zero, you as the instructor will need to apply your professional judgment, knowledge of your students, and the specific context surrounding the assignment” (Turnitin). Those figures are Turnitin’s own claims about its own model, so they should not be read as a neutral measure or applied to other products.
This site publishes no accuracy figure for the same reason. The honest position, for plagiarism tools and AI detectors alike, is that a score is a place to start looking. See the accuracy page for what we do and do not claim.
How should you read the results together?
Suppose an essay returns a similarity score of 4% and an AI-like share of 70%. The low similarity score does not clear the essay. It says only that little of the text matches indexed sources. The high AI share does not convict the writer either. It says that much of the text resembles AI writing, and the highlighted passages are where to look next: are they generic, are they conventional, does the writer’s other work sound similar?
Now suppose the figures are reversed, with 40% similarity and almost nothing flagged as AI-like. The likely story is borrowed or quoted material from human writers, so the question becomes whether it is cited. The AI detector has no opinion on that.
In both cases the result opens a question. It does not answer it. For essays specifically, the AI detector for essays page explains how highlights fit into a review.
If you already have the text, run it through our AI-generated text detector and review the flagged passages rather than relying on the overall score. It will not check sources, so keep a plagiarism checker alongside it if copying is part of your concern.
Frequently asked questions
Does a plagiarism checker detect ChatGPT? Not by design. It matches text against indexed material, and model output is usually new wording, so it often returns a low similarity score. Detecting machine-written text is the job of a separate AI detector.
Does a low similarity score mean the writing is human? No. A low similarity score means little of the text overlaps with the sources the tool searched. Original wording can still come from a model, and a source that is not indexed cannot be matched.
Is an AI detector the same as Turnitin? No. Turnitin offers text matching, which produces the Similarity Report, and a separate AI writing indicator. This site’s detector only classifies writing patterns and does not compare text to any source database.
Can I use only one of them? You can, if you only have one worry. Use a plagiarism checker for copying and an AI detector for generated text. When a decision about a person depends on the result, use more evidence than either score alone.