I signed up for Grammarly a few weeks ago to find out how its citation finder tool works (spoiler: it’s not great, very pushy and often produces irrelevant citations) and was surprised to see it generating expert reviews with real people’s names. Yesterday Grammarly discontinued this feature after a class action lawsuit was filed against the company by Julia Angwin, a journalist who has written extensively on privacy and technology for many years and one of the people being impersonated. Other people I’ve seen pop up include Helen Sword, Kate Crawford, TreaAndrea M. Russworm, Samantha Blackman, Adrienne Shaw, Scott Rettberg, Noah Wardrip-Fruin, Katherine N. Hayles and more. I couldn’t provoke it to generate a fake me 😉 After the Wired article, discussions about about this are all over my BlueSky feed, but there aren’t many examples of how this looks – I guess most academics I follow aren’t using Grammarly (phew). So here are some examples.

Here’s an example from when I showed Kishonna Gray how it worked when she was visitng last week. First we asked Grammarly to generate a 1000 word essay on black game studies, then we asked for the Expert review. Unsurprisingly, Kishonna shows up, as shown above. Here’s another example from a text generated about electronic literature.

As you can see, each expert gives pretty bland, run-of-the-mill advice. We clicked on “Show example” for fake-Kishonna’s advice and got an explanation of why they suggested it. But while the advice isn’t bad – be more precise about the time – it’s not well tied to the supposed reason they gave it. In fact, the supposed reason is that Kishonna’s work is situated, for instance by discussing how Black folks’ experience as gamers relates to what’s happening at the time, like Black Lives Matter or Gamergate. Situated writing is of course something that LLMs are extremely bad at. And just adding “In the early 2000s….” isn’t really helping with that.

If you click “Show example” it suggests exactly how to rewrite your text and you can click to insert the revision. This is frictionless not-writing for students and academics. It would honestly be difficult to use this tool and not just accept what it suggests.

This is just one of many AI tools Grammarly now provides. Here is the full menu:

The irrelevant citations it suggests are particularly annoying to me as a peer reviewer and reader of academic articles that are often clearly at least partly AI-generated. Scholarslop, as David Berry has called it.

Here is an example of how Grammarly suggests sources to cite. You select the “Citation finder” and it identifies statements in your text that it predicts need a citation. It provides a few possible choices and offers a big green button labelled “Insert in-text citation”. Here are two screenshots taken on 18 February 2026 showing how it does this – the second is for the same statement but I clicked on one of the other suggested sources to cite.

If you actually read the statement Grammarly says needs a citation, and compare this to the sources suggested, you’ll see they don’t match. Both sources are irrelevant citations. The claim is that promotional writing that is not intended to be primarily factual makes up a large part of the training data of LLMs. The sources offered are about how much LLMs are used in scientific writing, and about how many people have used LLMs. But it takes a lot more cognitive effort to read this and decide that no, they’re not good sources, than it does to just trust Grammarly and click that big green button and move on.

I’ve got a draft article in review that addresses some of this more extensively. For now, if you have students who use Grammarly I highly recommend signing up and checking it out so you know how to talk with them about it. You may also be interested in checking how Grammarly grades a paper as if it is a specific professor (like you?) if you type in the instructor name, the class code, the university and ideally upload the grading matrix.

Grammarly is a lot more than a tool to help non-native speakers check their spelling and grammar. It has become a full-scale plagiarism machine. I read some comments on Bluesky from NLP scholars asking what happened – apparently Grammarly was founded in Ukraine and was an NLP darling whose developers attended all the NLP conferences. But then, as summaried at Wikipedia, they bought other companies (a document editor, an AI-enabled email tool), got a lot more funding, and starting in October 2025 rolled out a lot of AI tools. Grammarly now seems more focused on enabling click-button AI plagiarism than on helping people become better writers.

You get a free week’s subscription, but have to give them your credit card. Maybe not a great idea. When I cancelled after 6 days they gave me an extra free week. So if you’re disciplined enough to remember to cancel in time, you could try it for yourself.