If I Use Grammarly, Will It Show Up as AI? What Turnitin and Grammarly Actually Say
By Ashikur Rahman16 min readQuick answer: not for ordinary grammar fixes, but likely yes for rewrites, and both companies say so in writing. Turnitin's FAQ answers "No" when a student uses Grammarly for grammar checks, then says content from Grammarly's generative features, "including draft generation, paraphrasing, summarizing", "will likely be flagged as AI-generated". Grammarly's own AI Detector guide agrees: red and blue underline corrections "should typically not impact the percentage score", while rewrites from its agents "come from our LLM". The line is correcting versus rewriting. The proof that you wrote it is Grammarly Authorship, if you switch it on before you start.
What does Turnitin say about Grammarly?
Turnitin says no for grammar checks and likely yes for anything Grammarly generates. The question appears word for word in Turnitin's AI writing detection FAQ:
"If students use Grammarly for grammar checks, does Turnitin detect it and flag it as AI? No. Our detector 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. Based on tests we conducted on human-written documents with no AI-generated 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. Please note that 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."
Two phrases carry the weight. "In most cases" means not every case, and Turnitin publishes no sample size or rate for its Grammarly test. And the exclusion names paraphrasing, the feature most students reach for when a sentence feels clumsy. The same FAQ says Turnitin can also spot text changed by "AI paraphraser or bypasser (also called humanizers) tools to evade detection", in English submissions only.
What does Grammarly say about its own edits being flagged as AI?
Grammarly gives the same answer from the other side. Its AI Detector user guide says "traditional, non-generative corrections in the form of Grammarly red and blue underlines should typically not impact the percentage score." Using its writing agents or its generative assistant "to meaningfully rewrite sentences and full paragraphs will raise the percentage score." Then comes a note that is blunter than most articles on the subject:
"Note: If you are working with other Grammarly agents to rewrite your text, it is likely that the AI Detection agent will flag your content as AI-generated. That is because rewrites from the Grammarly, Paraphraser, and Humanizer agents come from our LLM. If you are concerned about being flagged for inappropriate AI use, we suggest that you refrain from using our agents to rewrite your content; instead, focus on using agents to guide feedback that you manually incorporate."
That last sentence is the practical rule. Let the tool point at a problem, then fix it in your own words. Grammarly publishes no test of how much its underline corrections move its own score, and does not name the model behind its rewrite agents.
Which Grammarly features change your words, and which only correct them?
Grammar, spelling and punctuation fixes only correct. Full-sentence rewrites, AI prompts, the Paraphraser and the Humanizer change the wording, and that is the output both companies expect to be flagged.
| Grammarly feature | Effect on your text | What the detector pages say |
|---|---|---|
| Spelling, grammar, punctuation (red) | Corrects only | Turnitin: "in most cases" not flagged. Grammarly: "should typically not impact" its score. |
| Clarity and word choice (blue) | Small changes Grammarly calls non-generative | Grouped with red underlines. No detector publishes a separate test. |
| Tone detection | No change | Nothing to detect. |
| "Rewrite full sentences with a click" (Pro) | Changes wording | Grammarly: meaningful rewrites "will raise the percentage score". |
| AI prompts (draft, improve, summarize) | Generates text | Turnitin: "will likely be flagged as AI-generated". |
| Paraphraser agent | Rewrites whole paragraphs | Grammarly: "come from our LLM". Turnitin lists paraphrasing as likely flagged. |
| Humanizer agent | Rewrites to sound "more natural" | Grammarly: likely flagged. Turnitin detects humanizer tools. |
| AI Rewriter agent | Swaps phrases "AI often uses" | No detector publishes a test. |
| Authorship | No change; records the process | The evidence Grammarly, GPTZero and universities point to. |
The Paraphraser offers six presets, "Humanize, Academic, Professional, Streamlined, Creative, or Simple". The Humanizer guide says it tunes "AI-generated writing to sound more natural, sincere, and distinctly human", and it is available on Free as well as Pro and Plus. So a Free account does not mean grammar-only help. What matters is which button you press.
Grammarly's page on turning off AI says generative features "are not turned on by default" and any user can turn them off in account settings.
How do Turnitin's rules apply to a real essay?
Turnitin needs at least 300 words to produce an AI report and shows any AI share below 20% as an asterisk with no number. Those two rules decide what an instructor sees.
| Turnitin rule | What the page says | What it means for you |
|---|---|---|
| Length | "at least 300 words of prose text"; "must not exceed 30,000 words" | Shorter pieces get no AI report. |
| Low scores | FAQ: "1% to 19%". Report guide: "between 0 and 20" | Below 20% shows "*%", no number. |
| False positives | "under 1% for documents with over 20% of AI writing" | About one fully human paper in 100 wrongly flagged. |
| Use of the score | "should not be used as the sole basis for action" | A score is not a misconduct finding. |
| Visibility | "not visible to students" | Instructors can share the PDF with you. |
The minimum is on Turnitin's file requirements page. Using the AI Writing Report explains the asterisk: "there is a higher incidence of false positives when the percentage is between 0 and 19." Because the two pages word the band differently, this article says "below 20%".
Worked example: a 1,000-word essay
Assume 1,000 words of ordinary prose, all counted as qualifying text. Turnitin may exclude lists or headings, so treat these as round numbers.
Step 1, is it long enough? 1,000 is more than 300 and less than 30,000, so it gets an AI report.
Step 2, where is the asterisk line? 0.20 x 1,000 = 200 words. If the detector attributes fewer than 200 words to AI, the instructor sees "*%" and no highlights.
Step 3, grammar fixes only. Turnitin says such changes were "in most cases" not flagged. Expect a low score or the asterisk, but Turnitin publishes no rate, so no one can promise it.
Step 4, two 150-word paragraphs through the Paraphraser. 2 x 150 = 300 words, and 300 / 1,000 = 0.30, or 30%. That is above the 200-word line, so if the detector catches it, a number and highlights appear.
Step 5, the score can understate. Turnitin says "if we identify that 50% of a document is likely written by an AI tool, it could contain as much as 65% AI writing." A reported figure is closer to a floor than a ceiling.
Step 6, a 250-word reflection. 250 is less than 300, so no AI report under the current rules. The FAQ warns that with "only a few hundred words, the prediction will be mostly "all or nothing"".
Is the line between correcting and rewriting perfectly clean?
No. Originality.ai, a detector vendor, published a small Grammarly study using "ten known human IELTS essay answers" and "Originality Standard 2.0", a model the article says "is retired". "Lightly edited" meant it "only accepted Grammarly grammar suggestions". "Heavily edited" meant it accepted all suggestions, "including rephrasing and rewriting suggestions."
| Version of the 10 essays | AI scores, from the linked sheet | Plain reading |
|---|---|---|
| Original human essays | 0% on 9, 1% on 1 | Read as human |
| Grammar suggestions only | 69%, 78%, 0%, 0%, 0%, 0%, 7%, 1%, 3%, 0% | 8 of 10 at 7% or less; 2 flagged hard |
| All suggestions, incl. rewrites | 94%, 77%, 80%, 1%, 99%, 100%, 100%, 100%, 100%, 0% | 8 of 10 at 77% or higher |
The article's summary says light edits left "almost all of the examples continuing to display as human-written". That glosses over two grammar-only essays that scored 69% and 78%. For those writers, grammar fixes alone produced a score most instructors would notice.
The limits are real: ten essays, one retired model, lengths of 125 to 298 words (all under Turnitin's 300-word floor), no test date, and features that predate Grammarly's current agents. It is a vendor study, not peer reviewed. It still supports the direction both vendors state: rewrites move scores a lot, grammar fixes usually do not, and "usually" is not "always".
Transparency note: CriticNest has no affiliate or commercial relationship with Grammarly, Turnitin, GPTZero, Originality.ai or OpenAI. Nothing here is a paid placement. Every quotation and figure was read from those companies' own pages, and from the two university pages cited, on 28 September 2026. That is the position today; if that changes, the disclosure will move to the top of this article.
Is Grammarly's AI detector accurate?
Grammarly claims high accuracy but publishes no false-positive rate, and its own support guide says the result "should not be used as an objective source of truth". The free AI Detector page claims it "Achieves 99% detection accuracy" and ranks first on RAID's benchmark. Grammarly's RAID blog post says RAID uses "over 670,000 texts". The RAID leaderboard was not read for this article, so the ranking is Grammarly's claim.
The support guide is more careful: the feature "is not 100% accurate and should not be used as a definitive assessment". It is "optimized to minimize false positives", and "We expect these instances to be very rare", with no number. It adds that scores "may differ from those of other solutions like Turnitin, GPTZero, Copyleaks".
| Detector | Accuracy claim (vendor's words) | Not published on the pages read |
|---|---|---|
| Grammarly | "99% detection accuracy", first on RAID | False-positive rate; test of its own grammar edits |
| Turnitin | False positives "under 1%" for documents over 20% AI | Sample or rate for its Grammarly test; rate for non-native writers |
| GPTZero | "99% accuracy rate"; on RAID, "95.7% of AI texts" caught | Test of Grammarly grammar-only edits |
| OpenAI (withdrawn) | Caught 26% of AI text; flagged human text 9% of the time | Pulled "due to its low rate of accuracy" |
Grammarly and GPTZero both claim to lead RAID; neither claim was checked. OpenAI's classifier page is the useful history: "As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy." For a wider comparison, see our review of AI detector tools and where each one fails.
What does GPTZero do with Grammarly-edited text?
GPTZero added a label for this case. Its "AI Polished" post says the tag "identifies text originally written by a human and later improved by AI", filed under the "Mixed" class. It names Grammarly as an example of "writing enhancement" and says the label "indicates the work is original, unlike texts generated entirely with AI." GPTZero expects the tag "to appear on 4.4% of user submissions."
Why do universities say a detector score is not proof?
Because a small false-positive rate becomes many students at scale. Vanderbilt University said in August 2023 it would "disable Turnitin's AI detection tool for the foreseeable future", noting it "submitted 75,000 papers to Turnitin in 2022" and "around 750 student papers could have been incorrectly labeled". The sum is 75,000 x 0.01 = 750. Whether Vanderbilt has since re-enabled it was not verified.
The University of Pittsburgh Teaching Center says detectors "are not accurate enough to prove that students' have violated academic integrity policies" and that it "has disabled the AI-detection tool in Turnitin." Neither page mentions Grammarly.
The same arithmetic works for one class. At "one out of every 100 fully-human written documents", a lecturer marking 150 human essays could expect 150 / 100 = 1.5, so one or two wrongly flagged papers. Turnitin itself says it "does not make a determination of misconduct".
How does Grammarly Authorship prove you wrote it?
Authorship records where each piece of text came from as you write, then lets you share a report and a replay. The Authorship page says: "AI detectors only guess from the final draft. Authorship records your actual writing process - typed, pasted, AI-assisted - so you'll always have receipts."
The Introducing Authorship guide says it separates typed text, pasted text and AI-generated text, and also text "typed and then edited using traditional, non-generative features, such as grammatical error correction". That category settles the Grammarly question: the words were yours, and the tool fixed them.
| Authorship feature | Free | Paid plans |
|---|---|---|
| Unlimited reports and sharing | Yes | Yes |
| "Typed by a Human" analytics | Yes | Yes |
| Colour-coded report and replay | Yes | Yes |
| "Copied from a source" analytics, incl. AI share | No | Yes |
| AI and plagiarism detection in reports | No | Pro, Premium and Plus |
Know the limits. In Google Docs and Word it "does not track your writing activity" by default, so it only covers writing done after you switch it on. Word tracking has "a 20-page limit". Grammarly says it "may not capture 100% of cases accurately" and publishes no accuracy rate. Report data "remains on our servers for up to 10 years." You decide when to share; Grammarly "doesn't auto-send it to your professor".
The same page also suggests pairing the AI Rewriter "to refine flagged language". Rewriting flagged text to lower a score is the bypasser behaviour Turnitin says it detects. Authorship is the honest route; that one is not.
If I use Grammarly, will it show up as plagiarized?
Grammar corrections copy nobody's text, so they are not plagiarism, and the AI check and the plagiarism check are separate. An AI score estimates whether text looks machine-written. A plagiarism check looks for text matching another source. Grammarly Pro lists "Catch accidental plagiarism" apart from "Detect AI generated text". Authorship also shows whether sentences "came from a website", so a report shows where a quotation came from. The pages read did not cover Turnitin's similarity checker. Quote and cite anything you did not write.
Do you need Grammarly Pro, or is Free enough?
For grammar help and Authorship, Free is enough. Pro adds the features that change wording, plus AI and plagiarism checks. These lines come from the Grammarly plans page on 28 September 2026.
| Feature | Free | Pro |
|---|---|---|
| Price shown | "$0 USD / month" | "$12 USD" on the card |
| AI prompts | "100 prompts / month" | "2,000 prompts / member / month" |
| Rewrite sentences, adjust tone | Unavailable | Yes |
| Detect AI text, catch plagiarism | Unavailable | Yes |
| Authorship | Human-typed analytics | Plus source analytics |
The card shows "$12 USD" with no "billed annually" label. The page's embedded plan data lists $30 billed monthly and $144 billed annually, and $144 / 12 = $12, so the card is the annual price per month. The request was geolocated outside the US but returned USD; US pricing was not verified from a US connection. For alternatives, see our comparison of AI writing tools for bloggers.
How can you use Grammarly and still show the work is yours?
Use it to correct, not to write, and keep a record. This is about working honestly, not beating a detector.
- Read your course or client policy first. If generative AI is banned, turn it off in Feature customization.
- Turn on Authorship before the first word. It cannot record work done while it was off.
- Accept red and blue underline corrections. Both vendors say they should typically not move a score.
- Keep the Paraphraser, Humanizer and full-sentence rewrites away from assessed work unless your rules allow them.
- When a suggestion flags a weak sentence, rewrite it yourself, as Grammarly advises.
- Keep drafts. GPTZero tells instructors to ask for "drafts, revision histories, or brainstorming notes".
- Disclose any generative help your rules permit. Grammarly encourages writers to "provide proper attribution when they use AI."
Freelancers answer to clients the same way. For one documented AI-assisted workflow, see how 50 SEO posts were published with Claude Code.
How this was researched
Every page quoted was read at source on 28 September 2026: Grammarly's plans, AI Detector and Authorship pages, two Grammarly blog posts, and its support guides on the AI Detector, turning off AI, Authorship, prompts, the Paraphraser, Humanizer and AI Rewriter; Turnitin's AI detection FAQ, report guide, file requirements and release notes; GPTZero's homepage, FAQ and two blog posts; Originality.ai's study and its data sheet; OpenAI's classifier page; and the Vanderbilt and Pittsburgh pages.
Pages were fetched as plain HTML, with a headless browser where needed, and through the public help-centre API for support articles. Quotes are verbatim.
Not verified: the RAID leaderboard; any Grammarly false-positive rate, which Grammarly does not publish; any rate for Turnitin's Grammarly test, which Turnitin does not publish; whether Turnitin's bypasser detection names Grammarly's Humanizer; US pricing from a US connection. No detector was run and no plan was bought for this article. The author's role is careful reader of the fine print, not tester.
The verdict: correct freely, rewrite yourself, record everything
Using Grammarly to fix spelling, grammar and punctuation should not show up as AI. Turnitin answers "No" in its FAQ, and Grammarly says those corrections "should typically not impact the percentage score". Using Grammarly to write or rewrite is different. Drafts, paraphrases, summaries and Humanizer output come from a language model, and both companies say that output will likely be flagged.
The line is not perfectly clean. In Originality.ai's small test on a retired model, two of ten grammar-only essays scored 69% and 78%. Detectors disagree, and Grammarly publishes no false-positive rate. So process evidence matters more than any score.
The rule: accept corrections, rewrite weak sentences yourself, switch on Authorship before you start, keep drafts, and disclose any generative help your rules allow. If a score is questioned, the record of how you wrote is your answer.
Frequently asked questions
If you use Grammarly, does it show up as AI?
Grammar, spelling and punctuation fixes usually do not. Turnitin's FAQ answers "No" for grammar checks and says Grammarly's changes were "in most cases" not flagged in its tests. Grammarly says its underline corrections "should typically not impact" its own score. Text generated or rewritten by Grammarly's AI features is different: both companies say it will likely be flagged.
Does Turnitin detect Grammarly?
Turnitin says it is not tuned to target Grammarly's spelling, grammar and punctuation changes. It says content from Grammarly's generative features, "including draft generation, paraphrasing, summarizing", "will likely be flagged as AI-generated". It also detects text changed by AI paraphrasers and humanizer tools in English submissions.
Does Grammarly count as AI?
Partly. Grammarly says its correction engine uses AI but "does not change the substance of the writing". Its generative features, such as prompts, the Paraphraser and the Humanizer, produce new wording from its language model. Whether that counts as AI use depends on your course or client policy, so read it first.
If I use Grammarly AI, will it show up as plagiarized?
Plagiarism checks look for text matching other sources, which is separate from AI detection. Grammar fixes copy nothing. Generated text is not copied from a source either, but it may break your course's AI rules, which is a different problem. Grammarly Pro offers "Catch accidental plagiarism" as its own feature.
Will the Grammarly Paraphraser get flagged by Turnitin?
Probably. Turnitin lists paraphrasing among Grammarly's generative features whose output "will likely be flagged". Grammarly's own guide says Paraphraser rewrites "come from our LLM" and its own detector will likely flag them.
Is Grammarly's AI detector accurate?
Grammarly claims "99% detection accuracy" and a first-place RAID ranking, which this article could not verify. Its support guide says the result "should not be used as an objective source of truth". Grammarly publishes no false-positive rate and warns its scores may differ from Turnitin's and GPTZero's.
What is Grammarly Authorship?
Authorship records how a document was written: text you typed, text pasted from the web or an unknown source, AI-generated text, and text edited with Grammarly's corrections. It produces a colour-coded report and a replay you choose whether to share. It works in Google Docs, Word and Grammarly's docs, only while tracking is on.
Is Grammarly Authorship free?
The core is. Grammarly says "Grammarly Authorship is available in Grammarly Free and Grammarly Pro". Free users get unlimited reports, sharing and "Typed by a Human" analytics. Paid subscribers also see "Copied from a source" analytics, including the share of AI-generated text, plus AI and plagiarism detection inside reports.
Can Turnitin see a score below 20%?
No percentage is shown. Turnitin displays an asterisk with no number for low AI scores, because its testing found more false positives in that band. The FAQ says "1% to 19%" and the report guide says "between 0 and 20", so the safe reading is anything below 20%.
What should I do if Grammarly edits got my essay flagged as AI?
Ask to discuss it and bring evidence of your process: your Authorship report if it was on, earlier drafts and version history. Turnitin says its score "should not be used as the sole basis for adverse actions against a student".
