Does Turnitin detect Grammarly? Not the checker, yes the rewriter
No for the spelling and grammar fixes, and probably yes for the rewrites. Turnitin answers this one itself, in a single entry on its FAQ page. Its detector is not tuned to Grammarly's corrections, it says, and output from Grammarly's generative features will likely be flagged. Which button you press is the whole question. Write from sources you can open, in your own sentences, and the question never comes up.
Turnitin answers this itself, in two halves
Most pages on this question are guessing on your behalf. Turnitin has published an answer, under a question written in almost the words you typed: if students use Grammarly for grammar checks, does Turnitin detect it and flag it as AI?
The answer opens with one word, and then spends the rest of the paragraph narrowing it. Our detector, Turnitin writes in its AI writing detection capabilities FAQ, read on 20 September 2026, "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". It adds that on human-written documents with no AI 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".
Then the clause that sends half the people reading this to the wrong conclusion. That answer, Turnitin says, "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."
Two things in that paragraph are worth holding onto. The first is "in most cases", which is not "never". The second is that the line Turnitin draws is not between Grammarly and other tools, it is between two things inside Grammarly. Nothing about which app you opened decides this. What you asked it to do decides it.
Which Grammarly you are using
Grammarly stopped being one product some time ago, and the naming does not help. Underlines that appear as you type and offer a correction are the checker. A pencil or a lightbulb that offers to write something for you is the other half, and Grammarly files it separately in its own documentation, as generative AI assistance that helps you "quickly compose, rewrite, ideate, and reply".
| What you asked Grammarly to do | Which half of Turnitin's answer it lands in |
|---|---|
| Fix a spelling, a comma, a tense | Not what the detector targets |
| Rewrite or rephrase a sentence | Generative, likely flagged |
| Shorten, expand or summarise a passage | Generative, likely flagged |
| Draft a paragraph from a prompt | Generative, likely flagged |
| Run text through the Humanizer | Generative, likely flagged |
The Humanizer deserves its own line, because Grammarly sells one. It is free on the web page, with a fuller version behind Pro, and it rewrites AI text to sound less like AI. Grammarly also sells an AI detector and a plagiarism check, which puts it on both sides of the same table.
Grammarly's own answer, on its own humanizer page, is blunter than most of what ranks for this query. Asked whether the tool will help you bypass AI detectors, it says "This tool is not intended to bypass AI detectors." Asked whether using one is cheating, it says that in academic settings or anywhere AI use is restricted or must be disclosed, "using an AI humanizer to disguise the use of AI may be considered unethical and could constitute cheating or plagiarism". The company selling it is not the one promising you it works.
Why a rewrite is a different object from a comma
Turnitin's two answers look arbitrary until you know what the detector reads. It is not matching your text against a library of Grammarly output. Sentences are extracted from your submission and segmented into overlapping sections, each segment is scored between 0 and 1 for how likely it is to be human or machine written, each sentence inherits its segment's score, and the pooled scores make the percentage an instructor sees. We take that number apart in what your Turnitin AI score means.
So the question the model is asking, sentence by sentence, is whose word choices these are. Change a comma and the sentence is still the one you built. Accept a rewrite and the sentence that lands is one a language model produced, from your idea, in its own wording. That is the same object the detector was built for, arriving by a different door, and Turnitin's coverage list makes the point explicitly: it names the models it detects and then adds "and tools based on these LLMs as well".
There is a second effect, in the opposite direction, that catches honest writers. Among the things Turnitin lists as producing false positives is text with little structural variation, text that repeats itself, and text "that has been paraphrased without developing new ideas". Flat, evenly weighted, reworded prose is what the model associates with machine writing, whoever produced it. Polishing every sentence of your own draft into the same register can push the number up without a single generated word in the document, which is the mechanism behind getting flagged for writing you did yourself.
The one published test that ran Grammarly over AI text
Almost everything written about Grammarly and Turnitin is assertion. One peer-reviewed study actually did the experiment, and it is worth reading for its method as much as its result.
Muhammad Abid Malik and Amjad Islam Amjad, writing in the Journal of Applied Learning and Teaching in early 2025, generated fifteen essays of up to 500 words, five each from the free versions of ChatGPT 3.5, Gemini and Perplexity. They then applied three adversarial techniques to fresh copies: editing through Grammarly, paraphrasing through QuillBot, and 10% to 20% editing by a human language expert. For the Grammarly set they used the paid version and, in their words, decided to accept all the changes Grammarly suggested. Sixty documents went through four detectors.
Turnitin returned a 100% AI score on every file, before and after. The Grammarly pass moved nothing. The other detectors barely moved either: ZeroGPT's average on the ChatGPT essays went from 99.4% to 99%, on the Gemini essays from 95.4% to 95.2%, and Writer AI, poor throughout, went from 26.8% to 25.8% on the same set. QuillBot, applied to those texts instead, roughly halved ZeroGPT's averages and cut Writer AI's by more than half. Grammarly did not behave like a paraphraser, because on that setting it is not one.
Now the limits, which the authors are open about. Five essays per model, 500 words each, one run through each detector, and 2024-era free chatbots. Turnitin sat at 100% on the raw text, so it had no room to show a difference in either direction. And the study does not test the case most people asking this question are actually in, which is human writing put through Grammarly. Nobody has published that experiment. The authors' own conclusion is that a detector score "should not be taken as a final verdict, and further checks and investigations should be carried out before labelling a text as AI-generated or otherwise". Their paper also notes an earlier test that reported Turnitin at 0% on GPT-4 text, which tells you how wide the spread in this literature is.
Treat the confident percentages elsewhere accordingly. The pages promising you that Grammarly raises a score by some exact amount are selling a humanizer or an editing service, and none of them describes a method you can repeat, which is the same pattern we documented on the QuillBot version of this question.
What a marker actually sees
The AI writing indicator is visible to instructors and administrators, and to you only if an instructor downloads the report and shares it, and it is thinner than its reputation. Scores from 1% to 19% show as an asterisk with no percentage at all, because false positives are likeliest there. Only prose sentences in long-form writing are scored, so lists, bullet points and code are skipped and the percentage need not match the highlighted text. For the English check a submission needs at least 300 words of prose and no more than 30,000, and the detector runs only in English, Spanish, Japanese and Modern Arabic, with the paraphrase and bypasser detection English only. Turnitin also says it undercounts on purpose to hold its false positive rate under 1% on documents over 20% AI, validated before each release against over 700,000 papers written before ChatGPT existed.
Alongside it sits a product people confuse with it. Turnitin Authorship, part of Originality, is a different technology: it "uses metadata as well as forensic language analysis to detect if the submitted assignment was written by someone other than the student", and Turnitin is careful to add that "it will not be able to indicate if it was AI written; only that the content is not the student's own work". Neither tool prints the words Grammarly or ChatGPT anywhere on a report.
Grammarly's answer to all of this is a record of its own. Grammarly Authorship tracks the writing as it happens, collecting the text you write, including text you write and then delete, text you paste with the name of the source it came from, and your prompts to generative AI. It labels passages as typed by a human, copied from a source or copied from an unknown source, and the report a reader opens carries the colour-coded document and a replay of how it was written. It runs in Google Docs through the browser extension, in the Word desktop app up to 20 pages, and Grammarly says it may not capture 100% of cases accurately.
That record is genuinely useful, and there is one published test of how well it reads. In November 2025 Jonathan Bailey of Plagiarism Today turned it on for an article he wrote normally and got a fair account back: 95% written by a human, 92% typed directly, the one unattributed passage explained by a paste he made before switching the tracking on. Then he generated an essay with Gemini inside a Google Doc, which came back as 100% unknown, and ran that text through Grammarly's Humanize feature. The report's headline then read 75% typed by a human, while the fine print underneath said 0% typed by a human and 75% rephrased with Grammarly's AI. Six days later he added an update: Grammarly's owner had been in touch to say it was changing how humanized text is displayed. We could not confirm from outside whether that shipped.
Read that carefully and you have the honest shape of the whole question. A process record helps you, and only if the person reading it opens the detail. Whatever the header says, the writing stats showing a 500-word essay finished in under two minutes say more.
Your university has already answered the harder question
Detectors change every quarter. Conduct rules do not, and they were written with this exact split in mind.
Arden University's guidance, updated 21 April 2026, says using Grammarly for its basic spelling and grammar checking features is fine, but it is your responsibility to ensure the software does not rewrite parts of your work for you while checking it. On the other half it is unambiguous: "Using these generative AI features (whether on the free or paid-for versions) is not allowed, and will likely result in an allegation of academic misconduct." It adds that the same rule applies to any writing software, not only Grammarly.
The University of Birmingham's Academic Skills Centre draws the same line and then does something more useful: it lists both sides. In its safe zone are fixing spelling, punctuation and basic grammar, correcting words that sound alike, and removing filler without changing meaning. Outside it, the guide says, you must not use any tool to "Generate or rephrase sentences, expand ideas, or improve your arguments", restructure an assignment, reword content to sound more academic, or shorten your work to meet a word count. That last one catches people who thought they were doing housekeeping.
Both universities put the boundary where Turnitin puts it, and neither needs a detector to enforce it. Read your own course's policy, and our guide to whether using AI to write essays counts as cheating covers how these rules are usually drawn.
It matters because the cost of landing on the wrong side of a number is real. In April 2024 EdSurge reported the case of Marley Stevens, a junior at the University of North Georgia, whose paper was flagged through Turnitin's AI detection. She says she wrote it herself using standard grammar checking. She took a zero, dropped below the grade threshold for her HOPE scholarship, was placed on academic probation and paid $105 for a required seminar. Turnitin's chief product officer told the reporter that the tool starts conversations rather than settles them, and that "We very much had to train the teachers that this is not proof that the student cheated." Grammarly's head of education did the arithmetic in the same piece, warning that even at 98% accuracy a typical university would falsely call 1,000 papers a year cases of cheating. If you are living through one, we wrote what to do when you are accused of using AI.
The draft that needs no explaining
Strip the tools away and this question has a simple core. You want to know whether a machine will read something into your sentences that you did not put there, or whether it will find something you did.
The second half you control completely. Use the checker for what it is good at and let the arguing, the ordering and the phrasing stay yours, and there is nothing in the document for the generative clause of Turnitin's answer to attach to. The first half you cannot control, so build the evidence instead. Keep your sources, your notes and your drafts. A version history is worth more than any score, and we set out what actually persuades a marker in how to prove you did not use AI.
There is a check worth more than either. Rewriting never fixes a reference to a paper that does not exist, and a marker who clicks one dead DOI has something far more concrete than a percentage. You can test every reference in a draft in about a minute with our free citation checker, which takes no login.
That is the standard CiteOwl is built around, and it is deliberately not a humanizer. Its agent researches real papers, attaches every factual claim to a source you can open with the quote that supports it, and proposes each change as a diff you accept, reject or edit, so the sentences that land are the ones you chose. If your question is really about a chatbot draft rather than a grammar pass, the Gemini version of this question covers what Turnitin says it can and cannot see there.