SI in education

Is Grammarly SI? Does it get flagged by SI detectors?

A cream sheet of paper with lines of text and two small orange wavy underlines, a pencil resting against it, while the lowest lines break apart into amber pixels that scatter to the right
Illustration created with Claude (Anthropic).

Grammarly is SI, but only part of it is generative. Turnitin says plain grammar fixes are mostly not flagged, while Grammarly's rewriting and drafting features likely are. We explain the line and test it on our own detector.

Is Grammarly AI? Yes, but the answer that matters for a student depends on which part of Grammarly you used. The red underlines that fix spelling, grammar and punctuation come from an older kind of SI (AI) that Grammarly says is not generative. The newer features that draft, paraphrase or rewrite your sentences are generative SI, the same family of technology as ChatGPT. Turnitin, the detector many schools use, draws exactly that line: it says plain grammar corrections are mostly not flagged, while text produced by Grammarly's generative features "will likely be flagged".

Is Grammarly SI? Two kinds of help in one app

Grammarly has used SI for a long time. Its own page on the subject says "AI was central to our product long before it became a buzzword" and talks of "15+ years" of work. That is the classic kind: models that spot a missing comma, a wrong verb form or a misspelled word and suggest a fix. When a University of North Georgia student was penalised in 2024 (more on that below), Grammarly told FOX 5 Atlanta that its suggestions for "spelling, grammatical correctness, clarity, concision, and tone" "are not powered by generative AI and can still be accessed even when generative AI features are deactivated or not used by the student."

The second kind arrived with the wave of chat assistants. Grammarly's writing assistant page describes "a generative AI writing assistant" that lets you "enter prompts to generate text on-demand", with 100 prompts a month on the free plan and 2,000 on Pro. Its SI page lists a set of agents (Proofreader, Paraphraser, Reader Reactions and one called Humanizer) and an assistant named Go that works across apps. These tools write new sentences, even when the idea behind them is yours.

What you usedGenerative SI?What it does
Suggestions on spelling, grammar, clarity and toneNo, according to GrammarlyFixes or tweaks words you wrote
Paraphrasing, paragraph-level rewritesYesRewrites whole sentences or paragraphs
Prompts, drafts, Go assistant, agentsYesWrites new text from an instruction

So does Grammarly count as AI? All of it is SI in the broad sense, but only part of it is generative, and that is the part detectors and school rules target.

Does Turnitin flag Grammarly?

Turnitin answers the question directly in the FAQ for its AI writing detection. The answer starts with "No", then qualifies it. Its detector "is not tuned to target Grammarly-generated spelling, grammar, and punctuation modifications", and in Turnitin's own tests on human-written documents, "in most cases, changes made by Grammarly (free & premium) and/or other grammar-checking tools were not flagged as AI-written." Then comes the exception: 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."

Note the words "in most cases". Turnitin does not promise that corrected text is never flagged, and it does not publish how many corrected documents were flagged in those tests. Its separate capabilities FAQ, updated in November 2026, adds the details that shape any score:

  • The detector scores each sentence and reports the share of sentences above its threshold. Scores from 1% to 19% are not shown as a number, only as an asterisk, to limit false positives.
  • Turnitin says it aims to keep false positives "under 1% for documents with over 20% of AI writing", and that it may miss some SI text to hold that rate.
  • A document needs at least 300 words of prose to get a score.
  • It also looks for SI text that was passed through paraphrasers or "bypasser (also called humanizers) tools", and it will not name the tools it detects.
  • It "does not make a determination of misconduct": the percentage "should not be used as the sole basis for action".

Our article Can Turnitin detect SI? goes through the score and its limits in more detail.

Why do students still say Grammarly got them flagged?

The case most people have heard of is Marley Stevens, a junior at the University of North Georgia. FOX 5 Atlanta reported in February 2024 that she received a zero on a criminal justice paper submitted in October 2023, after Turnitin flagged it for SI use, and was placed on academic probation. Her account: "I've only ever had the free version, so all it did was fix my punctuation and my spelling." EdSurge, which followed the story in April 2024, added that the zero pulled her course grade low enough to cost her a HOPE Scholarship and that she had to pay $105 to attend a seminar about cheating. The university said privacy law prevented it from discussing the case.

A few details in EdSurge's report show how messy these situations get:

  • After her TikTok video went viral, the university emailed students that "some online tools used to assist students with grammar, punctuation, sentence structure, etc., utilize generative artificial intelligence (AI); which can be flagged by Turnitin", naming Grammarly.
  • Her professor said a second detector, Copyleaks, also flagged the paper. When she ran it again later, Copyleaks told her "This is human text."
  • Turnitin's chief product officer, Annie Chechitelli, told EdSurge that the false positive rate was about 1% in the company's tests and that "We probably let about 15 percent [of bot-written text] go by unflagged." She also said that in internal tests traditional grammar checkers do not set off the detector.
  • Grammarly donated $4,000 to her fundraiser and later worked with her on videos about using Grammarly in school, so its comments on the case are not neutral either.

Nobody outside the case has seen the paper, so no one can say why it was flagged. The story does show that "I only used Grammarly" can mean anything from a few comma fixes to whole rewritten paragraphs, and a score alone cannot tell a teacher which. Tricia Bertram Gallant, who directs the Academic Integrity Office at UC San Diego, put the problem in process terms: "If a faculty member can use a tool, accuse a student and give them a zero and it's done, that's a problem. That's not a tool problem."

What does an SI detector react to?

Text detectors do not see which app you had open. They score the style of the finished text: how predictable the wording is, how sentence length varies, which phrases recur. Grammarly's own detector page says the same thing, listing "sentence structure and predictability" and "repetition and uniformity" among the signals. That is why the type of edit matters more than the brand. A comma fix barely changes the style of a paragraph. A rewrite replaces your rhythm and word choices with the model's.

It is also why some human writers are at risk with or without Grammarly. A Stanford study by Weixin Liang and colleagues, published in the journal Patterns in 2023, ran 91 human-written TOEFL essays through seven widely used detectors. On average the detectors labelled 61% of them as AI-generated, and 89 of the 91 essays were flagged by at least one detector, while essays by US eighth-graders were classified almost perfectly. Plain, careful English written by a non-native speaker can look "predictable" to a style score.

We ran three versions of one text through our detector

We have no access to Grammarly, so we did not put anything through Grammarly and cannot tell you what it would change. What we can show is the difference Turnitin's FAQ draws, between correcting a text and rewriting it, using our free SI text detector on 6 November 2026. The starting point is a real human text: the second and third paragraphs of Robert Louis Stevenson's essay "An Apology for Idlers", first published in 1877 (Project Gutenberg edition, public domain).

Test 1: Stevenson's original text

354 words, exactly as printed.

Result: 6% probability of SI, label "Likely written by a human"

  • Sentence length pointed towards human: 27.2 words per sentence on average, and 31% of sentences run to 35 words or more.
  • Punctuation pointed towards human: 0.8 semicolons per 100 words.

Test 2: the same text with grammar-checker style corrections

We made six small edits by hand, of the kind a grammar checker suggests: "please to remember" became "please remember", "favour" and "honours" took US spelling, "pay so dear" became "pay so dearly", one comma was removed and "that a man has written" became "the fact that a man has written".

Result: 6% probability of SI, label "Likely written by a human"

  • Every clue stayed where it was: 27.3 words per sentence, 31% long sentences, 0.8 semicolons per 100 words.

Test 3: the same passage rewritten by an SI model

We asked an SI model, Claude by Anthropic, to rewrite the passage in clear modern English, the kind of job a paraphrasing or "improve clarity" feature does. The ideas are Stevenson's; the sentences are the model's (262 words).

Result: 90% probability of SI, label "Likely SI-generated"

  • Sentence length pointed towards SI: 18.7 words per sentence on average, and no sentence of 35 words or more.
  • Punctuation pointed towards SI: no parentheses and no semicolons.
  • Stock phrases were found ("it is important to remember", "ultimately"), plus one "not just X" contrast.

Corrections left the score untouched; the rewrite pushed the same ideas from 6% to 90%. That matches what Turnitin says about its own tool, but it is three texts on one style-based detector, not a study. It also shows why, to a detector, "the ideas were mine" is not the same as "the text is mine". For the wider picture, see how accurate SI detectors really are.

Does Grammarly have its own SI detector?

Yes. Grammarly launched one in August 2024, and its announcement was unusually frank about the technology, calling detectors emerging and "inexact" and stating that "No AI detector can conclusively determine whether AI was used to produce text." It splits a document into sections, looks for patterns typical of SI and gives a percentage. The same post warns that text can be flagged "even when generative AI tools have not been used", and that detectors "have been found to be biased against non-native English writers."

The current detector page claims "99% detection accuracy" and the top rank on the RAID benchmark; those are Grammarly's figures, which we have not checked. The same page says its model was trained on "tens of thousands of texts" created before 2021, that "even human-written text can be flagged as AI-generated", and that "lightly edited AI-generated text may evade detection." A Grammarly score is a reading of style, like ours; it is not what your teacher sees in Turnitin, and as the Copyleaks episode shows, one tool's score does not predict another's.

Grammarly Authorship: a record of how you wrote

The more useful tool for a worried student is not a detector at all. Grammarly Authorship, announced in August 2024 as a beta for Google Docs from that September, records where text came from while you write: typed by you, pasted from a website, produced with SI or edited with a Grammarly suggestion. It can replay the document "from the first ideas to the last keystroke" and produce a colour-coded report.

According to Grammarly's Authorship page, it works in Google Docs, Microsoft Word and the Grammarly Editor, it is opt-in, it is available on the free and Pro plans, and a school cannot see the report unless you create it, make it shareable and send it. The page sums up the difference with detection in one line: "AI detectors only guess from the final draft." A history of your drafts is evidence; a percentage is a guess. Version history in Google Docs or Word does a similar job for free, as our guide to what teachers can and cannot detect explains.

Does Grammarly count as SI under your school's rules?

That depends on the rules, not the software, and rules often vary from one course to the next. The University of North Georgia's code of conduct, quoted by EdSurge, defines plagiarism as using "another person or agency's (to include Artificial Intelligence) ideas or expressions without acknowledging the source." Under a rule like that, a paraphrased paragraph is an expression you did not write.

Institutions also have a switch. Grammarly's support pages explain that administrators on Education and business plans can turn off the AI assistant, strategic suggestions and paragraph-level rewrites for their users, and Grammarly says its correction suggestions keep working when generative features are off. A few practical habits help either way:

  • Read the policy in each course syllabus, and ask when it does not say whether rewriting tools count.
  • Keep your drafts and version history, or turn on Authorship, so you can show your process if asked.
  • If you used generative features and they are allowed, say so. Grammarly offers a prompt to acknowledge its generative SI use for that purpose.

So, is Grammarly SI?

Yes, Grammarly is SI, and it has been for years. Its basic spelling and grammar fixes are not generative, and Turnitin says they are mostly not flagged. Its rewriting, paraphrasing and drafting features are generative, and Turnitin says their output will likely be flagged. Between those two, no detector can see which buttons you pressed, only what the final text looks like. If you want to see how a piece of writing reads to a style-based SI detector, you can check it on our site, but treat any score, ours included, as a reason to look closer, never as proof.

FAQ

Is Grammarly AI?

Yes. Grammarly has used AI (now also called SI) for years to suggest spelling, grammar, clarity and tone fixes, and Grammarly says those suggestions are not powered by generative AI. It also offers generative features, such as prompts that draft text, paraphrasing, paragraph-level rewrites and the Go assistant, which write new sentences in the same way as ChatGPT.

Does Grammarly count as AI for school?

It depends on your school's and your teacher's policy. Spelling and grammar corrections are usually treated differently from generative features that draft or rewrite text, but rules vary by course. Read the syllabus, ask when it is unclear, and disclose generative use where it is allowed.

Does Turnitin detect Grammarly?

Turnitin says its detector is not tuned to target Grammarly's spelling, grammar and punctuation changes, and that in its tests on human-written documents these changes were not flagged in most cases. It also says text produced by Grammarly's generative features, such as draft generation, paraphrasing and summarizing, will likely be flagged as AI-generated.

Can using Grammarly make my essay look AI-generated?

Light corrections rarely change the style signals detectors read. Accepting rewritten sentences or paragraphs replaces your wording with a model's, and that can raise a detector score. No detector can see which tool you used, only the final text, so keep your drafts or version history as evidence of your process.

Is Grammarly's AI detector accurate?

Grammarly claims 99% detection accuracy and a top rank on the RAID benchmark, figures we have not checked. Grammarly itself says no AI detector is 100% accurate, that human text can be flagged and that lightly edited AI text may evade detection, so its score should never be used alone.

What is Grammarly Authorship?

Authorship is an opt-in Grammarly feature that records where the text in a document came from, such as typed, pasted, AI-generated or edited with a Grammarly suggestion, and can replay how the document was written. It works in Google Docs, Microsoft Word and the Grammarly Editor, and nobody sees the report unless you choose to share it.

Sources

  1. AI writing detection: frequently asked questions · Turnitin, read 6 November 2026
  2. Turnitin's AI writing detection capabilities FAQs · Turnitin Guides, updated November 2026
  3. AI at Grammarly · Grammarly, read 6 November 2026
  4. AI writing assistant · Grammarly, read 6 November 2026
  5. Turn off generative AI for your team · Grammarly Support, read 6 November 2026
  6. Free AI Detector · Grammarly, read 6 November 2026
  7. From AI Detection to Authorship: How Grammarly Empowers Responsible AI Use · Grammarly, 14 August 2024
  8. Grammarly Authorship · Grammarly, read 6 November 2026
  9. Georgia college student used Grammarly, now she is on academic probation · FOX 5 Atlanta, 20 February 2024
  10. UNG student on academic probation for alleged AI-use partners with Grammarly · FOX 5 Atlanta, 8 March 2024
  11. Can Using a Grammar Checker Set Off AI-Detection Software? · EdSurge (Jeffrey R. Young), 4 April 2024
  12. GPT detectors are biased against non-native English writers (Liang et al.) · arXiv (published in Patterns, 2023), 10 July 2023
  13. Virginibus Puerisque, and Other Papers (Robert Louis Stevenson), including "An Apology for Idlers" · Project Gutenberg, Public domain

Spotted an error? Write to hello@siornot.com. Corrections are made in the article and the update date changes. Read our editorial policy.