The short answer
We cannot tell you, and we do not think anyone outside Grammarly can yet. Grammarly claims 99% accuracy on the RAID benchmark. We found no peer-reviewed study that tests this detector, so that figure stands unchecked. Grammarly itself says the score is an estimate and should never be used alone.
What Grammarly claims
Its AI detector page says the tool achieves 99% detection accuracy and ranks first on RAID’s independent benchmark. The same page limits that claim:
- It says no AI detector is 100% accurate.
- It says no detector can conclusively determine whether AI was used to produce text.
- It says detectors do not track authorship, so human writing can be flagged.
- It says lightly edited AI text may evade detection.
- It says you should never rely on the results of an AI detector alone.
Its user guide says the model is tuned to minimize false positives, because Grammarly considers wrongly flagging human text the more harmful error. It says that should be very rare. It gives no false positive rate on the pages we read. Other makers also cite RAID for their own tools, and we have not checked any of these results.
Independent evidence
We searched journals and preprint servers on 1 November 2026 and found no independent study of Grammarly’s AI detector. That does not mean it is good or bad. It means nobody we could find has measured it in public. The first results for this question are forum threads and reviews, which are one person’s few samples. The wider evidence on detectors is in do AI detectors work?
What the score means
Grammarly’s user guide calls the score the percentage of scanned text that is likely AI-generated. It splits your text into sections and checks each against a model. It describes the result as an averaged estimate and says shorter passages are harder to score. Its developer page says the check cannot provide a definitive conclusion, needs at least 30 words and takes up to 100,000 characters.
So 30% does not mean a person or a model wrote 30% of the text for certain. It means the model estimated that share. Grammarly also says its model is in-house, so scores may differ from other tools.
What raises the score, according to Grammarly
This matters if you are worried about your own use of Grammarly. Its user guide says traditional corrections, the red and blue underlines, should typically not affect the score. Rewrites from its AI agents come from a language model and can raise it. So can text from tools such as ChatGPT or Gemini. Whether any of this is allowed is up to your course. See does Grammarly count as AI?
Why a Grammarly score is not proof
Grammarly says it itself: detection should never be a standalone check. That applies to the score you get and to any score your teacher gets. If a result worries you, keep your drafts and see how to prove you didn’t use AI.
Three things hold for any AI detector, this one included:
- It reads the finished text. It does not see how the text was written, so it can only estimate.
- Its figures depend on the test. An accuracy number holds for the texts it was measured on. Your text may be shorter, edited or in another language.
- A small error rate is still real people. Liang and colleagues (2023) found seven detectors, which the paper does not name, flagged 61.3% of TOEFL essays by non-native speakers on average. Detectors have changed since, but plain, predictable writing remains the risk.
Sources
Read 1 November 2026.
- Grammarly: AI detector, user guide and AI Detection API.