Grammarly Launches New AI-Powered Editor

Yesterday, Grammarly sent out a press release announcing its “new agent-powered doc” feature, which incorporates eight different AI agents to help with various aspects of the writing process.

According to Grammarly, the new tools are designed to help both students and professionals write better work.

The launch coincides with the start of the 2025 school year. Grammarly is already a divisive tool among educators, and the additional AI features of its writing platform are likely to generate even more concerns.

As such, it’s worth taking a moment to understand exactly what Grammarly is launching, who it is for and what issues it may raise.

While the update puts AI front and center, it doesn’t offer a significant difference from what the company was able to do previously. However, that doesn’t mean that the new features will be without controversy, especially from the instructors and experts it is basing its opinions on.

The Key Features

Disclosure: I use Grammarly’s non-generative tools to help with proofreading and editing. However, for this post, I am using the new editor and will play with the new tools more than normal.

According to the press release, Grammarly’s new editor has eight different agents in addition to its regular AI chatbot. Those agents are:

1. Reader Reactions Agent: Predicts what readers will likely learn or understand from your text. (Note: Though mentioned in the release, this agent was not available for me to test.)
2. AI Grader Agent: Predicts what grade you are likely to get on a paper. Can either provide generic feedback or feedback tailored to the assignment and instructor.
3. Citation Finder Agent: Finds citations for various claims made in a work, also seeks out refuting evidence.
4. Expert Review Agent: Offers automated feedback “inspired by experts.” It has dozens of experts listed, though Grammarly is not affiliated with those individuals.
5. Proofreader Agent: Offers inline suggestions to improve one’s writing.
6. AI Detector Agent: Determines whether a work is likely to have been generated by AI.
7. Plagiarism Checker Agent: Checks if the content was copied from other sources.
8. Paraphraser Agent: Adapts your writing style for a particular tone or audience.

According to Grammarly, all the agents in question, except for the AI Detector and Plagiarism Checker Agent, will be available to all users, including those using the free version (albeit with fewer AI prompts).

The features are currently only available in their doc feature, but will be available across all websites eventually.

Those familiar with Grammarly will note that many of these features have been available for some time.  The AI and Plagiarism Checker features have long been available, and its proofreader agent has long been the service’s core function. It also already offers assistance with tone, similar to its Paraphrasing Agent.

As such, much of the focus has been on its AI Grader and Expert Agent. Both have raised concerns regarding how they work and how they were trained.

The New Concerns

The two biggest new features are the AI Grader and Expert Review Agents.

The AI Grader is straightforward. It can either work with “basic guidelines,” or you can add a rubric and an instructor. If you take the latter route, you can either upload or paste the assignment guidelines and then list the instructor’s name, school, and class.

According to Grammarly, it will then check “publicly available instructor information to deliver tailored recommendations and estimated grades.”

The Expert Review Agent works similarly. But, instead of targeting a specific instructor, it pulls advice from different experts on writing. However, Grammarly makes it clear that references are for “informational purposes only” and that they “do not indicate any affiliation with Grammarly or endorsement by those individuals.”

In both cases, Grammarly’s AI is essentially trying to predict what a particular person would say in response to a paper. However, in all cases, it lacks an official connection with the person and bases its guesses on available information.

This raises a serious risk: That the AI could get it drastically wrong. While it is right to attribute the information to its source, there’s no guarantee that the suggestions will accurately represent that person’s advice. This risks not only the user, but also the reputation of the experts and instructors.

By tying suggestions to specific humans, it risks those humans being blamed for dubious advice. In that regard, it could be considered a case of reverse plagiarism. Attributing suggestions to a human that never would have given them in the first place.

The Academic Implications

From an academic integrity standpoint, there isn’t much new here. Grammarly launched its generative AI features in March 2023, and this doesn’t expand upon them.

Instead, Grammarly has launched a series of new AI agents designed for proofreading/editing, and has repackaged its features with a new interface. In many ways, this is an update in presentation more than capability.

Still, I can imagine that many instructors won’t be pleased that Grammarly is tailoring writing advice to them. That’s going to create some difficult conversations when that advice is wrong or the predictions of Grammarly’s AI turn out to be untrue.

And instructors are the most difficult ones for Grammarly to get right. With the other experts, there’s at least a large amount of content for the AI to train on. While it’s unclear if Grammarly had permission for that training, at least the content exists.

For instructors, the amount of available content is going to vary wildly. Some will have large amounts for the AI to examine, while others will have almost nothing. Similarly, the advice and guidance in the rubric may be helpful or not.

In short, this system is likely to struggle with making decisions and may wildly overestimate or underestimate grades. Instructors need to be aware of that and may want to anticipate being asked about it over the school year.

Bottom Line

Grammarly is already divisive in academia, and this is unlikely to change that. Grammarly is retaining its generative AI features but introducing more editing/proofreading tools and putting its AI features front and center.

My primary concern is that these tools could lead to human-written work being mistakenly flagged as AI-generated. As we discussed in June, AI polishing can cause otherwise human-written text to be flagged. As such, any AI editing could be problematic.

However, here Grammarly sells the solution to the problem it (in part) creates. Grammarly Authorship monitors the writing environment and tracks when work is pasted in. However, it is only available at schools with which it has partnerships.

To me, the biggest issue is that Grammarly is trying to be the perfect writing improvement tool for everyone. However, the needs of instructors, students, professional writers and everyday people are very different. This is especially true when it comes to AI.

Trying to be a tool for everyone means that it struggles to be a tool for anyone. I can’t help but feel that these changes will only further alienate it from academia, even as many of its new features target students.