We’re adding AI detection to Grammarly to empower our customers with tools for responsible AI usage. Our focus is on providing transparent information that end users, especially students, can use to make more informed decisions about the content they submit.
AI detection is a part of our plagiarism checker feature and is currently available when using Grammarly for Mac and Grammarly for Windows in Microsoft Word and Pages, the Grammarly Editor, and the Grammarly browser extension in Google Docs.
To check your document for plagiarism and to see if some of the content may be flagged as AI-generated, please open the Grammarly widget and click the quotation marks button:
Grammarly will then show, at the bottom of the assistant window, the probability of your text being flagged as AI-generated:
The AI detector shows the percentage of text that appears AI-generated and offers guidance on interpreting the results responsibly. Although it doesn't explain why the text is flagged, it helps users gauge, before submission, the likelihood of their text being flagged. The percentage result should not be used as an objective source of truth, as AI detection of any kind can be prone to errors. Grammarly users who use AI detection should use the results as a data point in deciding how they may want to edit, rephrase, and attribute sources properly to reduce the risk of inappropriate AI use.
Detailed FAQs
What does the % score mean?
The score represents the % of scanned text that is likely AI-generated, based on our extensive machine-learning training model. Our model is trained on hundreds of thousands of human and AI-generated texts. Grammarly breaks your text into smaller sections, checking each section against the model for patterns that are typically present in AI text (language patterns, syntax, and complexity). Grammarly then returns a % score, indicating how much of the provided text appears AI-generated. It assesses whether text resembles AI-generated writing; like other AI text checkers, it cannot provide a definitive conclusion.
How accurate is the score?
- The AI detection score is an averaged estimate of the amount of AI-generated text that is likely contained in a given document or piece of writing. Rigorous testing was done to ensure that the model is directionally accurate at identifying AI-generated text. Shorter passages are slightly harder to measure for an accurate score versus longer patterns.
- Like all AI detectors, Grammarly’s AI detection is not 100% accurate and should not be used as a definitive assessment of whether AI-generated text is present. The score itself should be viewed as an average estimate rather than a definitive percentage assessment.
I have not used AI-generated text in my document, but Grammarly is still saying that I have AI text in my writing.
Grammarly’s AI detection model is optimized to avoid false positives, as we believe that falsely identifying human-generated text as AI text is more detrimental than not accurately capturing AI-generated text. That said, there may be instances in which the human-generated language patterns, syntax, and complexity mirror that of AI-generated text more closely. We expect these instances to be very rare, but as with any AI detector, the results should always be viewed within broader context and not taken as an objective source-of-truth. We share more context in this article.
If my professor uses a different AI detector to scan student writing, will Grammarly’s % score be the same?
Grammarly’s AI detection uses a proprietary in-house model, so the scores may differ from those of other solutions like Turnitin, GPTZero, Copyleaks, and others. You should not view certain score with Grammarly’s AI detection as a clear indication that your professor will see the same score when they run it through their preferred AI detector. However, Grammarly’s model is trained similarly to some of the other most accurate and used detectors, so we believe the scores should be directionally aligned.
How does Grammarly’s AI detection distinguish between AI generated content, content that is modified by generative AI, and human-written text that is edited with traditional Grammarly features?
Because Grammarly’s AI detection model has been trained solely on human-generated and AI-generated text from LLM providers, traditional, non-generative Grammarly corrections (spelling, grammarly tone, clarity, delivery) typically should not impact the % score. 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 LLM provider (ChatGPT, Gemini, etc.) than text that was written by a human but modified by Grammarly’s generative AI (using prompts like Improve It, Make it Sound Academic, etc.). That said, paraphrasing edits with Grammarly and other providers are performed by LLMs, so those actions should still lead to some percentage of AI-generated text being triggered in Grammarly’s Ai detection.