Does Turnitin detect Grammarly as AI writing?

Spelling, grammar and punctuation fixes: Turnitin says that in tests, "in most cases", they were not flagged. Grammarly's generative features are a different matter, and Turnitin says so in the same paragraph: content produced with draft generation, paraphrasing, summarizing "and other features" will "likely be flagged as AI-generated by our detector". The line is between correcting a sentence you wrote and generating one you did not.

Here is the answer in full, from Turnitin's AI writing detection capabilities FAQs, because every clause in it gets misquoted somewhere:

"No. Our detector is not tuned to target Grammarly-generated spelling, grammar, and punctuation modifications to content but rather, other AI content written by LLMs such as GPT-3.5. Based on tests we conducted on human-written documents with no AI-generated content in them, in most cases, changes made by Grammarly (free & premium) and/or other grammar-checking tools were not flagged as AI-written by our detector. Please note that this excludes content generated by Grammarly's generative AI-powered features, including draft generation, paraphrasing, summarizing, and other features. Content produced using these features will likely be flagged as AI-generated by our detector."

Three things to take from it that the short version usually loses.

"In most cases" is a real hedge, not decoration. Turnitin is describing its own test results on documents with no AI content, and it declined to say "not flagged" without a qualifier.

The exclusion list ends with "and other features", which is open-ended. If a feature writes text rather than correcting text, assume it falls on the generative side, whatever it is called in the interface this month.

And "will likely be flagged" is about as direct as a detection vendor gets. That is not a warning about risk. It is a prediction.

Not by product, by action. Fixing a comma splice in your sentence leaves your sentence. Asking a tool to rewrite the paragraph "more clearly" returns a paragraph the tool composed. Both live inside the same subscription, which is exactly why the question keeps coming up.

The awkward middle is the rewrite suggestion that arrives as a single button labelled something friendly. If you accepted a dozen of those across a discussion section, the prose you submitted is partly composed by a model, whether or not you experienced it as editing.

Turnitin's answer above is about what the detector is tuned to target. It is not a guarantee about your document, and the same FAQ names the text properties that show up in its own false positives:

"Sometimes false positives (incorrectly flagging human-written text as AI-generated), can include content without a lot of structural variation, text that literally repeats itself, or text that has been paraphrased without developing new ideas."

Read the sentence after it, which almost nobody quotes: "If our indicator shows a higher amount of AI writing in such text, we advise you to take that into consideration when looking at the percentage indicated." That is the vendor telling instructors to weigh a high score on text of that shape differently, which is a sentence worth having to hand.

Now put those two facts next to each other. Heavy line-level editing, from any tool or any careful human, tends to smooth structural variation out of a draft. That is our reading and not a Turnitin claim, but it is the mechanism that would explain a clean-grammar paper coming back with a number on it, and it points at a defence you can actually make: bring the earlier, rougher drafts.

If your institution has switched on an authorship feature as well, do not read its output as a second opinion on AI. Turnitin's FAQ describes its own Authorship product this way: it "uses metadata as well as forensic language analysis to detect if the submitted assignment was written by someone other than the student", and "it will not be able to indicate if it was AI written; only that the content is not the student's own work".

Provenance and AI probability are two different claims about your document. Anything that records how a file came into being is answering the question a detector cannot answer, which is also why the drafting record below is worth more than it looks.

Keep the drafting record rather than changing your habits out of fear. Version history, dated files, the messy second draft with the paragraph you later cut. That evidence sits outside the argument about what any percentage means, which is why it holds up (how to keep it in Word, Google Docs and Overleaf).

If you have already been flagged, the first move is reading the report rather than rewriting the paper (what to do when you wrote it yourself).

Should I turn Grammarly off before submitting? Turnitin's stated position is that its detector is not tuned to target spelling, grammar and punctuation modifications. Turning off the generative features is the part that maps to what the FAQ actually says.

Does this apply to other grammar checkers? The quote says "Grammarly (free & premium) and/or other grammar-checking tools", so the tested claim is broader than one product. The generative exclusion follows the same logic: a feature that writes text is not a grammar checker.

My university says using Grammarly is fine. Is that the same as not being flagged? No, and they are worth keeping separate. A policy is about permission. A detector output is a statistic. You can be entirely within policy and still be holding a report with a number on it.

Does the detector know which tool touched my file? It does not report one. Turnitin describes the model as classifying segments by the probability that text is likely human or likely AI-generated (more on what that measurement is).