You used Grammarly and your work got flagged

Two different products behind one toolbar

Grammarly puts things that work in completely different ways behind a single interface, and which of them you used decides how this conversation goes.

A correction fixes an error inside your sentence: a missing comma, a misspelling, a verb that does not agree with its subject. The sentence stays yours. A rewrite is not that. When you accept a suggestion to make a sentence clearer, shorter or more formal, a model generates a replacement and you take it. Do that across a paper and a real share of the submitted wording was produced by software, even though every idea in it came from you.

Generative features go further again. Anything where you asked for an introduction, a paragraph or an expansion produced text you never wrote at all, and calling that editing will not survive a follow-up question.

The report cannot name the tool

Turnitin's AI writing indicator does not say Grammarly, or ChatGPT, or anything else. It reports a share of qualifying sentences its model scored as likely machine-generated, and highlights them.

Nobody can look at that number and tell whether it came from a chatbot, a translation tool, accepted rewrites, or nothing at all except your own unusually even prose. That cuts both ways, and both halves matter to you. It is why you can be flagged without having done the thing you are accused of. It is also why the score by itself cannot establish which of those happened.

Which features actually move a score

The mechanism is not mysterious. Detectors score how predictable your word choices are and how much your sentence lengths vary. A generated replacement sentence is, by construction, a likely one. Accept enough of them and the paper flattens on exactly the two axes being measured.

• Spelling, punctuation and grammar corrections. A few words changed, your structure untouched. These do not manufacture machine-patterned prose.
• Clarity and conciseness rewrites. Your sentence replaced by a generated one. This is the feature that moves a number, and the one students forget they used.
• Tone and formality rewrites. Same mechanism, usually applied to more of the sentence at once.
• Anything generated from a prompt. That text was written by a model and no framing changes it.

Grammarly Authorship, if it was switched on

In late 2024 Grammarly added a feature called Authorship, which records how a document was produced while you write: what you typed, what you pasted, what came from its own suggestions and what was generated. It produces a report you can show someone.

The catch is the one every tool of this kind has. It only holds a record if it was running at the time. Switching it on today records nothing about a paper you submitted last month, and no setting recovers that.

If it was on, retrieve the report before you meet anyone. It is unusual evidence in these cases, because it separates typed text from pasted text, and that separation is often the exact thing in dispute.

Check the policy before you argue about detectors

A lot of academic integrity policies already permit spelling and grammar tools while prohibiting generated text, and some name Grammarly directly. If yours does, this ends quickly, and the thing that ends it is a quotation from your own institution rather than anything about how detection works.

Read for the exact boundary. Policies that allow proofreading software without defining proofreading are where accepted rewrites become genuinely contested, and that ambiguity is not your fault. Ask for the interpretation in writing, and ask before the meeting rather than during it.

Describe what you did precisely

Vagueness is what sinks people here. 'I used Grammarly' is not an answer, because it covers both a comma fix and a wholesale rewrite, and the person reading it knows that.

Name the features. If you accepted rewrite suggestions, volunteer it and say roughly how much. A student who offers 'I took its clarity suggestions on maybe a dozen sentences, mostly in the middle section' reads as honest. The identical fact, discovered afterwards, reads as concealment, and that is a much worse case to be defending.

If you accepted a lot of rewrites

Then some of the submitted wording did come from a model, and the honest position is not that the detector got it wrong. It is that you wrote the paper, used a tool for sentence-level phrasing, and did not know where your institution drew the line.

That position is stronger than it feels at 2am. The question in front of an instructor is whether you produced the work, not whether every clause is untouched. Your outline, your sources, your version history and your ability to talk through the argument all still stand, and a tidied sentence does not affect any of them.

Do not repair it by rewriting the paper now, and do not run it through anything to bring the score down. Both convert an argument about tools into an argument about your conduct after the accusation, which is the one argument you cannot win.