Guide
Grammarly AI Detector: How It Works, How Accurate It Is, and What It Flags
A plain-English guide to the Grammarly AI detector: the free checker and the paid AI Detector agent, what Grammarly's accuracy claim refers to, how Grammarly says the score should be read, and how Grammarly Authorship differs from detection.
Quick answer: The Grammarly AI detector is a free web checker at grammarly.com/ai-detector that returns a percentage estimate of how much of a passage was likely AI-generated, plus a paid “AI Detector agent” in Grammarly Pro, Business, Education and Enterprise that marks the specific sections it suspects. Grammarly says the detector “achieves 99% detection accuracy and ranks #1 on RAID’s independent benchmark”. RAID is an academic benchmark published in 2024, built from 2023-era models. Grammarly’s support documentation says the model is optimised to minimise false positives, and that its percentage “should not be used as an objective source of truth”. Turnitin, a separate detector many universities use, says it does not target Grammarly’s grammar and spelling corrections but that Grammarly’s draft-generation and paraphrasing features “will likely be flagged as AI-generated”, so a Grammarly score does not predict a Turnitin score.
This guide covers Grammarly’s detector specifically: the two products, the accuracy claim, how Grammarly says to read the score, and how Grammarly Authorship fits alongside it. For the underlying detection mechanism across all tools, see how AI detectors work. For ranked alternatives, see best AI detectors. For the tool many institutions run, see Turnitin AI detection. For Grammarly as a writing tool, see best AI grammar checkers. For the assistants students use, see best AI for students.
Who owns Grammarly
The company behind Grammarly is now called Superhuman. On 29 October 2025 the company rebranded itself as Superhuman, bringing Grammarly, Coda and Superhuman Mail under one corporate identity and launching a Superhuman Suite and an assistant called Superhuman Go. The writing product kept its name, and the company told existing customers that “nothing changes about the products they rely on”. In this guide, “Grammarly” means the product and “Superhuman” means the company.
Superhuman also owns GPTZero. It announced the acquisition on 23 June 2026 (BusinessWire), with chief executive Shishir Mehrotra describing it as “bringing the most trusted writing tool and the most trusted AI detector into one platform”. On 30 September 2026 Superhuman said that GPTZero “is now part of Superhuman” (Superhuman). Terms were not disclosed.
What the Grammarly AI detector is
There are two products sharing one name.
| Product | Where it lives | Cost | What you get |
|---|---|---|---|
| Free AI detector | The public page at grammarly.com/ai-detector | Free, no sign-up required for a basic scan | A single percentage estimate for the pasted passage |
| AI Detector agent | Grammarly Pro, Grammarly Business, Grammarly for Education, Enterprise | Included in the paid plan | The percentage plus underlined text sections and per-section bars in a side panel |
Source: Grammarly’s AI detector page and the AI Detector user guide.
The free checker returns a single score; the paid agent marks the text. Grammarly’s support guide places the agent in the side panel for individual Pro users, in Microsoft Word for team-plan members, and in docs for Enterprise, Business and Education users where an administrator has enabled it. Grammarly’s benchmark announcement refers to the agent, not the free tool.
It is a writer-facing self-check. Grammarly does not offer an instructor-side dashboard that scores a class’s submissions. The product Grammarly leads with for institutions is Authorship, covered below.
Grammarly does not publish length limits or a language list for the detector on its detector page or support guide.
How Grammarly’s detector works
Grammarly runs a proprietary in-house classifier, and its documentation describes the training corpus in two ways. The support guide states: “Our model is trained on hundreds of thousands of human and AI-generated texts.” The product page states: “Our detection model was trained on tens of thousands of texts, including both human-written and AI-generated text created before 2021.”
The product page describes the model as examining sentence structure and predictability, repetition and uniformity, metadata traces from AI tools, and comparison against known AI-generated content. For the general mechanism, see how AI detectors work.
The output is a percentage plus section-level marking. In the agent, flagged sections are underlined in the document and shown as bars in a side panel; the free checker returns the percentage alone.
What the score means, according to Grammarly
Grammarly’s support documentation makes four statements about how to read the percentage.
It is tuned to avoid false positives. In Grammarly’s words: “The AI detection model is optimized to minimize false positives, as we believe that incorrectly identifying human-generated text as AI-generated is more detrimental than failing to accurately catch AI-generated text.” This is the same trade-off Turnitin describes: a detector set this way may miss some AI text.
Outside providers’ text is more likely to score higher. Grammarly states: “The model is also more likely to assign a higher percentage of AI-generated text in a given passage if it’s been generated by an outside LLM provider (such as ChatGPT or Gemini).”
There is no score-lowering feature. Grammarly states: “Currently, we do not offer a dedicated feature to reduce the AI score in user content.”
The number is not a verdict. From the same guide: “The percentage result should not be used as an objective source of truth, as AI detection of any kind can be prone to errors.” The detector page adds that there is “no AI detector that can conclusively or definitively determine whether AI was used”, that detection “should be just one part of a holistic approach to evaluating writing originality”, and that it “should never be used as a standalone verification method”.
How Turnitin treats Grammarly
Turnitin distinguishes Grammarly’s corrections from its generative features. Turnitin states that its detector is not tuned to target spelling, grammar and punctuation corrections, and that in testing those changes were not flagged in most cases. It says Grammarly’s draft generation, paraphrasing and summarising features “will likely be flagged as AI-generated” (Turnitin FAQs).
The two detectors are separate models with different settings, so a Grammarly score does not predict a Turnitin score on the same document. Turnitin does not let students self-check.
A court case. In Matter of Newby v Adelphi Univ. (2026 NY Slip Op 26021), decided on 28 January 2026, the academic integrity report against an Adelphi University student alleged use of Grammarly, after a professor obtained a 100% Turnitin AI score. The New York Supreme Court, Nassau County, annulled the finding as “without valid basis and devoid of reason” and ordered the record expunged, after finding the university had not followed its own conduct code or addressed the student’s evidence (Liebert Cassidy Whitmore). The case is covered in our Turnitin AI detection guide.
What Grammarly claims about accuracy
Grammarly’s detector page states that the tool “achieves 99% detection accuracy and ranks #1 on RAID’s independent benchmark, outperforming other leading AI detectors.”
The ranking claim traces to a Grammarly blog post published on 10 February 2026, Grammarly’s AI Detector Agent Ranks #1 in Quality, which says the agent “ranked #1 on RAID’s leaderboard for AI detection quality” and describes RAID as testing over 670,000 texts across writing styles and models, including attempts to fool the detectors. That post does not state a percentage. Grammarly does not publish a false-positive rate or a per-model detection rate. Its detector page also says “no AI detector is 100% accurate”.
About the RAID benchmark
RAID was published at ACL 2024 by Liam Dugan, Alyssa Hwang, Filip Trhlik, Josh Magnus Ludan, Andrew Zhu, Hainiu Xu, Daphne Ippolito and Chris Callison-Burch, and is maintained at the University of Pennsylvania (arXiv:2405.07940). The paper describes the dataset as “over 6 million generations spanning 11 models, 8 domains, 11 adversarial attacks and 4 decoding strategies”.
| Feature of RAID | Detail |
|---|---|
| Models in the corpus | ChatGPT, GPT-4, GPT-3 (text-davinci-003), GPT-2 XL, Llama 2 70B Chat, Cohere, Cohere Chat, MPT-30B, MPT-30B Chat, Mistral 7B, Mistral 7B Chat |
| Metric | Accuracy at a fixed false-positive rate, 0.05 by default |
| Corpus vintage | Dataset released May 2024, generations from 2023-era models |
| How submissions run | The submitter generates predictions locally and opens a pull request; a bot scores the submitted predictions |
Source: the RAID repository and the ACL paper.
What that means for reading the claim. Grammarly’s detector page says the tool identifies text from ChatGPT, Gemini, Claude, Microsoft Copilot and Grammarly’s own AI. RAID’s corpus does not include Gemini or Claude output, or OpenAI models newer than GPT-4, so a RAID result measures performance on the 2023-era models in the benchmark. Grammarly does not say which RAID split or false-positive rate its 99% figure refers to.
More than one vendor reports a first place on RAID. Trinka says its detector “ranked #1 for academic text, with a score (AUROC) of 0.999 out of 1.000”, and SuperAnnotate reports a #1 open-source result. RAID scores detectors across separate domains and adversarial splits, so different vendors can lead different splits. Grammarly does not publish its RAID score.
What independent testing found
The largest recent peer-reviewed study did not test Grammarly. The study by Marijke Van Vlasselaer, Filip Van Droogenbroeck and Bram Spruyt, published in the International Journal for Educational Integrity in June 2026, tested GPTZero, Copyleaks, Turnitin and Pangram on 160 academic papers (Van Vlasselaer et al.).
The only published third-party test that included Grammarly was run by a competing detector vendor.
| Test | Date | Method | Grammarly result |
|---|---|---|---|
| Pangram, 30 tools tested | 7 January 2026 | 9 AI texts (3 GPT-4o, 3 Gemini 2.0, 3 Claude 3.7 Sonnet) and 3 human texts, free tiers only; a pass required 75% or more AI on AI text and 25% or less on human text | No AI text scored at 75% or above; all 3 human texts at 25% or below |
The sample is twelve documents. Pangram sells a competing detector and placed its own tool first. The test used Grammarly’s free tier, not the paid agent that Grammarly’s benchmark claim concerns. Pangram states: “This test is not definitive of the validity of these tools.”
Grammarly Authorship
Grammarly’s other product in this area records the writing process, and it is free.
Grammarly Authorship records how a document was written. It categorises text by origin — typed by the writer, pasted from a website, generated with AI assistance, or pasted from a non-web source — and produces a shareable report that can replay the document’s construction (Grammarly Authorship). It works in Google Docs, Microsoft Word, Superhuman docs and Canvas (Grammarly).
It is free on Grammarly Free as well as Pro.
The writer controls disclosure. Grammarly’s page states: “You decide when to share your Authorship Report. Grammarly doesn’t auto-send it to your professor or anyone else.”
Since 10 March 2026 it can name which AI agent was used. Grammarly added agent-specific attribution, so a report can show how a writer used particular Superhuman agents such as Reader Reactions, Citation Finder or Proofreader. Grammarly reported more than 5 million Authorship reports generated since the feature entered beta in October 2024.
It records from when it is switched on. Because Authorship is opt-in and the writer decides whether to share the report, it is evidence a writer can offer about their own process.
Grammarly’s faculty-facing material explains its case for process evidence. Its post states that detectors “can be biased against non-native English writers and neuro-diverse learners” and that “detection models frequently fail to accurately identify AI-generated text from the latest models” (Grammarly for institutions, 15 October 2024).
The same approach is visible elsewhere, from Turnitin Clarity to provider-side AI watermarking.
How Grammarly’s detector differs from Turnitin and GPTZero
This is a comparison of how the tools are used, not a ranking. For ranked accuracy, see best AI detectors. QuillBot also offers a free AI detector.
| Grammarly AI detector | Turnitin AI writing detection | GPTZero | |
|---|---|---|---|
| Who can run it | Anyone; free basic check | Instructors and administrators at institutions licensing Turnitin Originality | Anyone; free basic check |
| Can a student self-check | Yes | No | Yes |
| Where the result appears | Percentage on the web page; underlines and side-panel bars in the paid agent | Inside the Similarity Report | gptzero.me |
| Minimum length | Not published | 300 words of prose | — |
| Low-score handling | Percentage shown as given | No score or highlights between 1% and 19% | Percentage with a confidence band |
| Owner | Superhuman | Turnitin | Superhuman |
Sources: Grammarly AI Detector user guide, Turnitin FAQs, Superhuman.
The main difference is who runs it. Grammarly’s detector is a self-check a writer runs on their own text. Turnitin’s is run by an instructor, and the student does not see it.
What it costs
| Plan | Price | AI detection included | Notes |
|---|---|---|---|
| Free web checker | $0, no account needed | Yes, basic percentage only | |
| Grammarly Free | $0 | Authorship yes; AI Detector agent no | 100 AI prompts per month |
| Grammarly Pro | $12 per month billed annually ($144/year); $30 billed monthly | Yes, AI Detector agent, plus plagiarism detection | 2,000 AI prompts per member per month |
| Grammarly Business | Not published on the plans page | Yes | Administrator-controlled |
| Grammarly for Education | Not published on the plans page | Yes | Administrator-controlled |
| Enterprise | Custom, contact sales | Yes | Adds data loss prevention, custom roles, BYOK encryption |
Source: Grammarly plans. Prices in USD. Grammarly Authorship is free on every tier.
What to do, by situation
A writer checking work before submitting. The free Grammarly checker costs nothing and needs no account. It does not predict what Turnitin will say.
A student who wants to be able to show they wrote something. Grammarly Authorship, switched on before you start, records your process as you go. Pair it with your platform’s own version history.
A student who has been accused. Preserve document version history, drafts, notes, outlines and timestamps. Ask which tool produced the score and what the exact figure was. Check whether your institution’s policy permits detector output as sole evidence — Grammarly’s detector page says detection “should never be used as a standalone verification method”, and Turnitin says its percentage “should not be used as the sole basis for action”.
An instructor deciding whether to rely on it. Grammarly’s detector page states that detection “should never be used as a standalone verification method”. For classroom tooling that does not rest on detection, see best AI for teachers.
An institution that wants process evidence. Authorship, Turnitin Clarity and comparable trackers record behaviour rather than infer from style. They work when students know in advance what is expected, which makes this a policy design question as much as a software one.
If you write in English as a second language, or you have a diagnosed learning difference, put that in writing early. In UK Office of the Independent Adjudicator cases analysed by HEPI, the outcome turned on institutions’ procedures, including whether detection reliability for non-native speakers was considered.
Summary
- Two products: a free web checker that returns one percentage, and a paid AI Detector agent that marks suspected passages inside documents.
- The accuracy claim: Grammarly says “99% detection accuracy” and “#1 on RAID”. RAID is built from 2023-era models, and Grammarly does not publish its RAID score or which split the 99% refers to.
- How to read it: Grammarly says the model is optimised to avoid false positives, is more likely to score outside providers’ text higher, and should not be treated as an objective source of truth.
- Turnitin: says it is likely to flag Grammarly’s generative features, but not its grammar corrections.
- Authorship: a free, opt-in record of how a document was written, which the writer chooses whether to share.
- Ownership: Grammarly is a product of Superhuman, which also owns GPTZero.
Frequently asked questions
Does Grammarly have an AI detector?
Yes. Grammarly offers a free AI detector at grammarly.com/ai-detector that returns a percentage estimate for pasted text, with no sign-up required for a basic scan, and a paid AI Detector agent in Grammarly Pro, Business, Education and Enterprise plans that also underlines the sections it suspects. Both are self-check tools for writers. The product Grammarly markets to institutions is Grammarly Authorship.
Is Grammarly’s AI detector accurate?
Grammarly’s detector page says it “achieves 99% detection accuracy and ranks #1 on RAID’s independent benchmark”. RAID is an academic benchmark built from 2023-era models that reports accuracy at a fixed false-positive rate, and Grammarly does not publish its RAID score. The largest recent peer-reviewed study did not test Grammarly. The only published third-party test that included it, by competitor Pangram in January 2026, used twelve documents and the free tier; none of the AI texts reached Pangram’s 75% pass mark, and all human texts were scored at 25% or below.
Is Grammarly’s AI detector free?
Yes, in its basic form. The web checker at grammarly.com/ai-detector is free and needs no account for a basic scan. The AI Detector agent, which marks individual sections inside your document, requires a paid plan; Grammarly Pro is $12 per month billed annually, or $30 billed monthly. Grammarly Authorship is free on every tier.
Does Grammarly’s AI detector detect ChatGPT?
Grammarly’s page says the detector identifies text from ChatGPT, Gemini, Claude, Microsoft Copilot and Grammarly’s own AI tools. Grammarly does not publish accuracy figures for each model. The benchmark it cites, RAID, includes 2023-era ChatGPT and GPT-4 text.
Will using Grammarly get flagged as AI by Turnitin?
It depends on which Grammarly features you used. Turnitin states its detector is not tuned to flag spelling, grammar and punctuation corrections, and that in testing those changes were not flagged in most cases. Turnitin says Grammarly’s draft generation, paraphrasing and summarising features “will likely be flagged as AI-generated”. Turnitin does not let students self-check before submitting.
Does using Grammarly count as cheating?
That is set by your institution’s policy. Most policies distinguish between correcting your own writing and generating text. Read the assignment brief, disclose the tools you used, and keep the drafts that show your own work. Grammarly Authorship is designed to make that disclosure straightforward.
Can Grammarly detect its own AI writing?
Grammarly’s page says the detector identifies text from Grammarly’s own AI tools. Its support documentation also says the model “is also more likely to assign a higher percentage of AI-generated text in a given passage if it’s been generated by an outside LLM provider (such as ChatGPT or Gemini)”. Turnitin says Grammarly’s generative features are likely to be flagged by its own detector.
What is the difference between the Grammarly AI detector and Grammarly Authorship?
The detector inspects a finished piece of text and estimates whether a machine wrote it. Authorship records how the document was produced as you write it, categorising text as typed, pasted from the web, pasted from elsewhere, or AI-assisted, and can replay the construction. Authorship is free on all tiers, opt-in, records from when it is switched on, and is shared only when the writer chooses.
How do I lower my Grammarly AI detection score?
Grammarly says: “Currently, we do not offer a dedicated feature to reduce the AI score in user content.” If the text is genuinely yours, the useful step is evidence rather than editing: enable Authorship, keep version history, and raise the result as a possible false positive.
Why does Grammarly say my own writing is AI-generated?
Any detector can produce false positives. Grammarly states its model is optimised to minimise them and that “the percentage result should not be used as an objective source of truth”, and its institutional material notes that detectors “can be biased against non-native English writers and neuro-diverse learners”. Preserve your drafts and version history, and if the score is being used in a decision, ask what your institution’s policy says about detector output as sole evidence.
Is Grammarly’s AI detector better than GPTZero?
The two have not been tested against each other in the same published study. GPTZero was tested in the 2026 Van Vlasselaer study and Grammarly was not. Both are now owned by Superhuman. For ranked accuracy across the market, see best AI detectors.
Did Grammarly buy GPTZero?
Superhuman, the company formerly named Grammarly, announced on 23 June 2026 that it would acquire GPTZero, and said on 30 September 2026 that GPTZero “is now part of Superhuman”. The deal value was not disclosed. GPTZero said at the announcement that “nothing changes for our daily users”. The company renamed itself from Grammarly to Superhuman on 29 October 2025 while keeping Grammarly as the name of the writing product.