Why Don’t Schools Like Grammarly? A Professor’s Take on AI, Detection, and Academic Integrity
A student came to my office hours last fall, frustrated and genuinely confused. She’d submitted a research essay she was proud of and it came back flagged for academic integrity review. She hadn’t used ChatGPT. She hadn’t used Claude. She’d used Grammarly to clean up her grammar and check her tone, the same way she’d been doing for three years of undergrad work without incident.
The policy had changed. The technology had changed. Nobody had clearly told her what either of those things meant for her work.
That conversation is what made me sit down and write this.
Schools that restrict or prohibit Grammarly are generally not objecting to spell-check or comma correction. They’re objecting to what Grammarly has become: a generative AI writing platform with features that go far beyond proofreading, including full paragraph rewrites, tone adjustments, AI-generated text suggestions, and a built-in AI humanizer. That’s not a theoretical question anymore. It depends entirely on which features a student is using and what the assignment is designed to measure.
Grammarly wasn’t always an AI writing tool
For most of its existence, Grammarly was a grammar and style checker. Sophisticated, yes, but fundamentally in the same category as the grammar-check function that’s been in Microsoft Word since the 1990s. Students used it. Faculty used it. Most institutions didn’t have a strong opinion either way.
The shift happened when Grammarly introduced Grammarly GO and then integrated generative AI across its platform. Users can now ask Grammarly to rewrite a paragraph entirely, generate text from a prompt, adjust the tone of an essay to sound more formal or more casual, or produce AI-generated completions sentence by sentence. The product didn’t just add features. It changed categories.
The evidence here is straightforward: if you’re asking a tool to write or substantially rewrite something for you, you’re no longer using a grammar checker. You’re using a co-author. If that co-author is acceptable in an academic context depends on the assignment, the course policy, and what the institution has decided counts as original work.
Institutions tend to get this wrong because they write policies about specific tools rather than about behaviors. “Students may not use ChatGPT” is a policy about a tool. “Students may not submit text they did not write or substantially revise themselves” is a policy about a behavior. The second version covers Grammarly’s generative features, ChatGPT, Claude, Gemini, and every AI writing tool that will exist in three years. The first one is already partial.
Why schools draw the line here, and why it’s messier than it sounds
I want to be precise about this, because the conversation tends to collapse into two camps that are both wrong.
The first camp says Grammarly is just a writing tool and restricting it is paternalistic and technologically illiterate. This position was defensible in 2018. It isn’t now, because the product that argument describes is no longer the product students are using.
The second camp says any AI assistance constitutes cheating and all of it should be banned. This position has the appeal of a clear line, but it ignores that students have always received writing help from others. Writing centers, peer reviewers, tutors, parents, and faculty all provide feedback that shapes student work. The question has never been about getting help. It’s been about what the submitted work demonstrates about the student’s own thinking, argument, and development as a writer.
The more interesting question is not “did the student use Grammarly” but “what does this submission demonstrate about this student’s learning?” AI detection tools exist precisely because that question is hard to answer just by reading the text. The problem is that the tools themselves are often insufficient for the job institutions are trying to give them.
The evidence here is that AI detection accuracy varies considerably by tool, by text type, and by the characteristics of the writer. Research published in early 2025 documented false positive rates between 8% and 17% on academic writing, depending on text genre and writer background. Non-native English writers are flagged at disproportionately higher rates than native speakers, a pattern that has real consequences in institutions that treat a detection score as evidence rather than as a probability estimate. I’ve written about this at length elsewhere, but the short version is that a detection score is not evidence of misconduct. It’s a statistical pattern that warrants a conversation, not a verdict.
What schools are using, and what the research says about it
The tools most commonly cited in institutional guidance are GPTZero, Turnitin’s AI writing indicator, and Copyleaks. All of them operate on probabilistic models. All of them produce a score, not a determination. None of them should be used as the sole basis for an academic misconduct proceeding, and the institutions using them that way are going to keep producing unjust outcomes for students whose writing happens to score high.
This is not what the research says about good detection practice. The research says detection should be one signal among several, never the primary one.
What works better is assignment design. Asking students to submit drafts, to document their writing process, to engage in brief oral defenses of their work, or to write in ways that require personal experience and reflection that no AI can supply. These aren’t workarounds. They’re what good writing pedagogy looked like before AI existed.
When detection is part of the process, sentence-level analysis is significantly more defensible than document-level scoring. A tool that identifies which specific sentences are likely AI-generated and explains why gives faculty something concrete to work with. A single number from 0 to 100 gives them almost nothing, aside from false confidence that they’ve run a proper check.
Proofademic’s sentence-level detection does exactly this: it flags individual sentences, assigns each one a probability score, and explains what triggered the flag. That’s more honest about what detection can and can’t tell you than a document score, and it’s more useful when faculty need to make a decision.
Does Grammarly count as AI? How to think about it clearly
Students ask me this question regularly. Usually they’re trying to figure out what’s allowed. I’m more interested in helping them understand why the question matters in the first place.
I want to be precise about this, because the answer depends on what Grammarly is doing to your writing in a given instance.
Grammarly’s grammar and style-checking features, the ones that flag comma splices and passive voice overuse, function more like a proofreader than a co-author. Using them is not meaningfully different from asking a writing center tutor to mark up your draft. Most institutional policies that restrict AI use aren’t trying to prohibit that.
Grammarly’s generative features, the ones that rewrite your paragraphs, adjust your tone, or produce text from a prompt, are a different category. Submitting the output of those features as your own academic work is, functionally, delegating the writing to an AI. It doesn’t matter that the interface looks friendlier than ChatGPT.
Proofademic’s blog has a useful breakdown of how the Grammarly AI question plays out in academic contexts, including which features tend to fall inside versus outside most institutional policies. Worth reading if you’re advising students or trying to write a clearer policy yourself.
The question to put to students isn’t “is Grammarly on the approved tools list.” It’s “what is this tool doing to your writing, and is the result still yours?” If it’s catching errors you would have caught with more time, that’s editing. If it’s producing prose you couldn’t have written yourself, that’s delegation.
What academically appropriate AI use looks like
I don’t want to end this without something constructive, because this newsletter is about policy and practice, not enforcement.
The evidence here is consistent across a significant body of research on academic integrity: students who understand why integrity policies exist comply at meaningfully higher rates than students who only know what the rules are. If a student’s only reason to avoid AI is the fear of being caught, we haven’t built academic integrity. We’ve built an adversarial surveillance relationship.
The institutions doing this well are doing a few things differently. They’re redesigning assignments so that original thinking, personal reflection, and process documentation are built into the assessment, not bolted on as an afterthought. They’re having explicit conversations about which tools are appropriate for which purposes and why the distinction matters for a student’s development as a thinker and writer. And when they use detection, they’re using it as one piece of evidence in a broader conversation, not as a verdict that forecloses one.
Grammarly’s place in all of this is a useful teaching case. It’s a product that changed what it is, in a policy environment that hasn’t kept up. The institutions that use that gap as an occasion to have clearer conversations about what academic writing is for are the ones that’ll come out of this moment with stronger academic cultures, not weaker ones.
The technology will keep changing. The underlying question, what does this student’s submitted work demonstrate about their learning, won’t.
Frequently asked questions
Does Grammarly count as AI for academic purposes?
It depends on which features you’re using. Grammarly’s grammar and spell-check functions are broadly analogous to proofreading and are generally not what institutions are prohibiting. Grammarly’s generative features, including paragraph rewrites, tone adjustments, and AI-generated text, fall within what most institutional AI policies restrict. If the policy is unclear, ask the instructor directly before submitting.
Why don’t schools like Grammarly?
Many schools don’t restrict basic Grammarly use. What they restrict is the use of its generative AI features. The concern is the same as with any AI writing tool: when the tool is rewriting or generating your academic work, the submission no longer fully reflects your own thinking and development, which is what academic assessment is supposed to measure.
How do professors check if an essay was written by AI?
Some use probabilistic detection tools like GPTZero or Turnitin’s AI writing indicator. Others notice patterns across a student’s work over the course of a semester: inconsistencies between in-class writing and submitted assignments, significant shifts in vocabulary or analytical sophistication, or unusual fluency on topics the student hasn’t otherwise engaged with. No single method is reliable in isolation. Process documentation and draft submission requirements are often more informative than post-submission detection.
What AI detectors do colleges use?
Turnitin’s AI writing detection and GPTZero are the most commonly cited in institutional guidance documents. Some institutions are beginning to evaluate tools with sentence-level detection, which produce more interpretable results than document-level scores and are more defensible in a misconduct conversation.
Can you use Grammarly and not get flagged for AI?
Grammarly’s basic grammar and style features are unlikely to trigger AI detection because they don’t substantially alter the text. The generative rewrite and text-generation features may trigger detection, and more importantly, using them to produce submitted work may constitute academic misconduct regardless of detection. The question isn’t whether the detection catches it. It’s whether the work is yours.