WritingFor students
Grammarly Is Three Acts, and the Course Policy Says Which Is Cheating
Using Grammarly isn't cheating by itself. It isn't automatically safe either, because one product name covers three separate acts. Correcting spelling and grammar is one. Rewriting whole sentences and paragraphs with the generative tools is the second, and Grammarly's own support pages say that meaningfully rewriting with those features raises the score in Grammarly's own AI detector. Prompting it for new text is the third. Integrity policies such as CUNY's charge unauthorized use, so the course policy is where a student finds which of the three is allowed.
Olive Staff & OliveUpdated
What Are the Three Acts Behind One Product Name?
Grammarly puts three different acts under one icon. The red and blue underlines fix spelling, grammar and punctuation. Rewrite tools and writing agents rebuild sentences somebody already wrote. And from a prompt, the assistant will produce new text, which is drafting. Grammarly's own support pages separate the first act from the other two. A paper's integrity position turns on which of them ran.
Grammarly's AI Detector user guide says traditional, non-generative corrections (the red and blue underlines) shouldn't, as a rule, impact the percentage score. Writing agents and the generative assistant are another matter. When they're used to meaningfully rewrite sentences and full paragraphs, the same page says, they will raise it 1.
| Act | What it does | Grammarly's own note |
|---|---|---|
| Correction underlines | Fixes spelling and grammar | Should not normally move Grammarly's own score 1 |
| Rewrite tools, writing agents | Rebuilds whole sentences | Will raise Grammarly's own score 1 |
| Generative assistant | Writes new text | Off until switched on 2 |
The generative half is opt-in. According to Grammarly's page on turning AI off, those features aren't on by default; they're available on demand, and users control when and how to use them 2. So a student who has never opened the rewrite panel has run a different product from one who rebuilt three paragraphs with it. At a hearing, both would give the same answer: "I used Grammarly."
Same words, different acts.
Naming the act instead of the tool also survives a product update. New features keep arriving under the same icon. A rule written against the icon goes stale the week it gains a button.
The same split decides whether a grammar check run through ChatGPT counts as cheating. Did the model hand back corrections, or new sentences? That's the question there too.
Which Act Does an Academic Integrity Policy Charge?
Many integrity policies charge unauthorized use, not tool use. Take the City University of New York. Its academic integrity policy, as amended to cover AI, defines cheating as the unauthorized use or attempted use of material, information, notes, study aids, devices, artificial intelligence systems or communication during an academic exercise 3. The word doing the work is unauthorized, and the UNC guidance below puts that authorization in the course. A vendor's marketing and a detector's reading grant nothing.
That wording does more than any list of banned apps could. A rule built on authorization already covers a tool nobody had heard of last term. It also puts the decision next to the assignment it applies to.
CUNY's policy goes on to treat unattributed AI-generated content as plagiarism, paraphrased content included, and asks that the AI be cited as the source 3.
Because the charge turns on authorization, the syllabus is the first place to look. In 2023 the writing center at the University of North Carolina at Chapel Hill told students to consult each syllabus for guidance or requirements related to the use of AI tools 4.
Wording runs from a flat prohibition to a requirement that any use be declared. Read the shapes an AI syllabus policy takes in week one, not the night before a deadline.
Style guidance reaches further than most students expect. MLA's 2023 guidance asks writers to acknowledge all functional uses of an AI tool, editing prose and translating words among them. The acknowledgment can sit in a note or in the text, or anywhere else suitable 5. That covers a sentence-level rewrite even when nothing the tool produced is quoted anywhere in the paper.
The third act has already reached a federal court. In Harris v. Adams, decided on November 20, 2024, the U.S. District Court for the District of Massachusetts denied a preliminary injunction sought for a Hingham High School student who'd been disciplined over a history project. Its opinion records a school administrator's account that the student admitted copying and pasting language from Grammarly directly into the draft he submitted 9. That was one ruling, on one record, in one state, and it isn't legal advice for any other case.
Will a Grammarly Edit Get the Paper Flagged?
Corrections rarely move a reading. Rewrites can move it a long way. Bloomberg Businessweek reported in 2024 on what happened when Grammarly was used to "improve" an essay or "make it sound academic": work that had passed as 100 percent human-written came back as 100 percent AI-written. Ordinary spelling and grammar suggestions had only a marginal impact 6. That measured gap is the practical reason to keep the three acts apart.
The same reporting has a case involving Grammarly. At the University of North Georgia, Marley Stevens said she was put on academic probation for a year after Turnitin flagged her essay and a disciplinary hearing decided she had cheated. She said she wrote the assignment herself. By her account her only help was Grammarly's standard spell-checking and grammar features 6.
Her account is hers. What the episode does establish is narrower and more useful: a flag on a corrected paper is a claim to be tested, and the testing happens outside the percentage.
How often is a detector wrong about writing that almost certainly no machine produced? Bloomberg checked. It ran GPTZero and Copyleaks over 500 college application essays submitted to Texas A&M University in mid-2022, months before ChatGPT was released. Bloomberg's point was that the date all but rules out machine writing. Still, 1 to 2 percent of them came back flagged, some of those flags near certain 6.
The worry is common enough to change behavior.
A Higher Education Policy Institute survey of 1,041 UK undergraduates, published in February 2025, asked what put students off AI. The top two answers were being accused of cheating, named by 53 percent, and false results, named by 51 percent 7.
Grammarly won't promise that its own number predicts anyone else's. Its guide tells students not to read a score from its AI detection as a clear indication that an instructor's preferred detector will report the same thing 1.
On a second page the company is blunter. No detector available today, Grammarly says there, can reliably and conclusively determine whether AI was used to produce text 2.
The AI Detector user guide adds that the company offers no dedicated feature for reducing the AI score in a document 1. That's worth knowing before a search for one ends up somewhere worse.
For what a flagged, Grammarly-edited draft actually needs, see what a student can show after a Grammarly-edited draft is flagged.
Check the Draft and Keep What Shows the Work
Three things can be ready before submission. Version history shows a paper arriving in pieces across days. Grammarly Authorship records typed text separately from pasted text. And a reading on the draft estimates which sentences read as machine-written, which is a different thing from a record of how they were written. The first two have to exist before anybody asks for them. Start them at the first paragraph, not at the first question.
Version history is the cheaper of the two. It's also easier to lose.
A document that appears in one paste has no trail behind it, whatever the truth is. The piece on the version history a student can actually show covers where it lives and how it fails.
Grammarly's Authorship page comes at the same problem from inside the editor. It tells text typed directly into a document apart from text pasted in, even text pasted from a source it can't see, such as a private browsing window 8.
That record helps a student who switched it on in week one. For one who switches it on the night a question arrives, it does nothing.
The third item is the reading. Human's check reads up to 1,200 words, one paper a day, with no account, and answers a narrower question than a hearing does: which sentences read as machine-written or machine-edited. This is an estimate from our detector. A flag should be reason to look closer and not verified proof of cheating. So a highlighted sentence is where a rereading starts, and the draft trail is what answers for it. Human's detector scores the text in memory and de-identifies any personal information for additional training.
If the rewrite tools rebuilt a paragraph, it's worth writing again in the student's own words, whatever it reads as. That's a writing decision. The loop open today is check, revise and check again, with the rebuilt paragraphs first in line. Alongside it, a writing tool is in private preview: it reads a draft with the same detector, colors each sentence by what the detector reads it as, keeps quotations, numbers, citations and anything the writer marks as their own untouched, and helps rewrite the rest in the writer's own words, and it never tells a person that their own writing is not theirs.
What Should a Student Do When the Syllabus Says Nothing?
Ask in writing, then keep the reply. A syllabus that says nothing about editing tools hasn't given permission for the generative half. Until an instructor answers, corrections are the low-risk act. Full-paragraph rewriting changes enough of the text to be worth declaring. Two sentences in an email can save a term of guessing.
Institutional policy is the clearer layer. In the same HEPI survey, 80 percent of respondents called their institution's policy on AI clear 7. That still leaves open whether the person who set the assignment counts a sentence rewrite as AI use.
So make the question specific.
Is grammar and spelling correction acceptable? Does sentence-level rewriting count as generated text? Those are two different answers, and an instructor asked for both will usually give both.
If the answer is silence, one line at the end of the submission puts the use on the record, though it authorizes nothing: Grammarly was used for spelling and grammar corrections, including punctuation, and no sentences or paragraphs were generated or rewritten by it.
That line is specific enough to check against the draft trail, and short enough to add to every assignment. MLA's guidance on generative AI already asks for functional uses of such a tool to be acknowledged 5. It is silent on a non-generative spell-check, so the line is a courtesy there and an expectation where the rewrite tools ran.
Declaring an edit has a cost, and pretending otherwise would be dishonest. A student who says nothing risks a question later with no good answer. One who declares a rewrite invites a conversation about how much of the paper is theirs, but that conversation can be had on evidence. That's the argument for keeping the three acts apart from the first draft on.
Common questions
Does Grammarly's spell-check count as AI use under a course policy?
Rarely as generative AI use. Grammarly's support pages call its correction underlines traditional and non-generative, and say they shouldn't, as a rule, move Grammarly's own AI detection score 1. A policy aimed at generative tools is aimed at text a model produced. The underlines don't produce any. Some courses ban every writing aid, spell-checker included, and there the correction engine is covered too. That's a question for the syllabus and the instructor, and no general rule settles it.
Is Grammarly's rewrite feature allowed in college work?
The answer sits with the course, and with whether the rewrite produced sentences the student didn't write. Grammarly states that its writing agents and generative assistant raise Grammarly's own AI detection score when they meaningfully rewrite sentences and full paragraphs 1. It also says those features stay off until somebody switches them on 2, so a course that permits editing help but not generated prose sits right on that line. Ask first, then declare the use.
Can a professor tell whether Grammarly was used on a paper?
Not from the finished file alone. A corrected paper and a rewritten one arrive as the same document with the same timestamp. What separates them sits outside the text: the version history, Grammarly's own record of typed against pasted material 8, and what the student can say about the argument. A detector percentage isn't that record, and Grammarly's guide tells students not to treat its score as a prediction of what an instructor's tool will report 1.
Does using Grammarly need a citation?
Generated text does. MLA's 2023 guidance on generative AI goes further than most students expect: it asks writers to acknowledge all functional uses of a generative AI tool, editing prose and translating words among them, in a note or the text or some other suitable place 5. It applies to Grammarly's generative features, not to a non-generative spell-check. The guidance is a style-guide expectation rather than law, and a course policy can be stricter or looser. One sentence naming the tool and the stage of the work it touched is usually enough.
What happens if a paper is flagged after only spell-check was used?
It becomes a claim to be tested, and the testing is done with evidence. Bloomberg Businessweek's 2024 reporting followed Marley Stevens at the University of North Georgia. She said she was put on academic probation for a year after Turnitin flagged her essay and a disciplinary hearing decided she had cheated. She also said she used only Grammarly's standard spell-checking and grammar features 6. Her account is hers rather than a finding either way. What answers a flag is the draft trail, plus the student's grasp of the argument.
References
- 1.AI Detector User Guide Grammarly Support, 2026. support.grammarly.comGrammarly's own split between non-generative correction underlines and generative rewriting, its warning that its score does not predict an instructor's detector, and its statement that it offers no score-reduction feature.
- 2.How Can I Turn Off AI in the Grammarly Product Offerings? Does the Use of Grammarly Trigger AI Detectors? Grammarly Support, 2026. support.grammarly.comSupports the claim that Grammarly's generative features are off by default and available on demand.
- 3.Academic Integrity Policy The City University of New York, 2022. cuny.eduSupports the definition of cheating as unauthorized use, with AI systems named, and the treatment of unattributed AI content as plagiarism even when paraphrased. The page dates the policy June 27, 2011, amended and replaced June 27, 2022, with a later explanation of the AI revision.
- 4.Generative AI in Academic Writing The Writing Center, University of North Carolina at Chapel Hill, 2023. writingcenter.unc.eduAdvises students to consult each syllabus for guidance or requirements on AI tools.
- 5.How Do I Cite Generative AI in MLA Style? MLA Style Center, 2023. style.mla.orgSupports the claim that MLA asks for functional uses of an AI tool, such as editing prose, to be acknowledged.
- 6.AI Detectors Falsely Accuse Students of Cheating—With Big Consequences Bloomberg Businessweek, 2024. bloomberg.comDavalos and Yin, October 18, 2024. Supports the finding that Grammarly's 'improve' and 'make it sound academic' features turned work read as 100 percent human into 100 percent AI while spelling and grammar suggestions had a marginal impact, the Texas A&M essay test, and Marley Stevens's account that she was put on probation after a disciplinary hearing. Bloomberg's page is paywalled; read via the course-reading copy at https://www.cs.utexas.edu/~pstone/Courses/309fall24/preclass/readings/week13__Do_AI_Detectors_Work__Students_Face_False_Cheating_Accusations___Bloomberg.pdf
- 7.HEPI/Kortext AI survey shows explosive increase in the use of generative AI tools by students Higher Education Policy Institute, 2025. hepi.ac.ukSurvey of 1,041 full-time UK undergraduates, fieldwork December 2024, published February 2025: 53 percent named being accused of cheating and 51 percent false results as what put them off AI; 80 percent called their institution's policy clear.
- 8.Introducing Authorship Grammarly Support, 2026. support.grammarly.comSupports the claim that Grammarly Authorship distinguishes typed text from pasted text, including text pasted from a source it cannot see.
- 9.Harris v. Adams, Memorandum and Order on Motion for Preliminary Injunction U.S. District Court for the District of Massachusetts (text via FindLaw), 2024. caselaw.findlaw.comDecided November 20, 2024; the motion for a preliminary injunction was denied. The opinion records: 'According to Mr. Hoey, RNH admitted that he had used Grammarly to "generate ideas" and "create[ ] portions" of the draft script, copying and pasting language directly from Grammarly into his submission.'
9 sources, numbered by first appearance.
General guidance for teachers, administrators and students. What holds at one institution, on one assignment, may not transfer to another.
Human checks whether a piece of writing can be verified as written by a person, and shows the sentences behind the result. It does not report a probability that a person used AI, it does not check for plagiarism, and no number it produces stands for a student's honesty. A flag should be reason to look closer and not verified proof of cheating.