Grammarly has a serious identity problem. The writing tool’s “Expert Review” feature is generating AI feedback under the names of real journalists, authors, and academics, none of whom gave consent, according to a report by The Verge AI published Wednesday.
The Verge AI’s own staff discovered the issue firsthand. Editor-in-chief Nilay Patel, editor-at-large David Pierce, and senior editors Sean Hollister and Tom Warren all appeared in the feature as AI-generated “experts” whose writing style supposedly inspires the feedback. None of them gave Grammarly permission to use their identities.
What the Feature Actually Does
Launched in August, Grammarly’s Expert Review promises to “sharpen your message through the lens of industry-relevant perspectives.” When users click the button in the sidebar, the AI analyzes their text and surfaces suggestions supposedly “inspired by” real-world experts.
The roster spans well-known names like Stephen King, Neil deGrasse Tyson, and Carl Sagan, alongside dozens of working journalists and editors. The Verge AI identified tech reporters from Bloomberg, Wired, The New York Times, The Atlantic, and multiple gaming outlets, all included without notification or consent.
Superhuman’s Defense Falls Flat
Grammarly’s parent company, Superhuman, offered a narrow justification. Vice president Alex Gay told The Verge AI: “The experts in Expert Review appear because their published works are publicly available and widely cited.” Superhuman also clarified that the feature “doesn’t claim endorsement or direct participation from those experts.”
But there are several problems with that position:
- Inaccurate profiles: Some expert descriptions contain outdated job titles, errors that could have been fixed by simply asking the people involved.
- Broken sourcing: The feature crashed frequently, and “source” links often led to spammy copies of legitimate sites, unrelated pages, or content not written by the named expert at all.
- Misleading presentation: In Google Docs, the AI suggestions render visually similar to real user comments, creating the impression of direct edits from the named expert.
The Imitation Problem
The Verge AI’s own testing exposed a deeper flaw: AI imitation of a person’s writing style doesn’t translate to understanding how that person edits. A suggestion generated under senior editor Sean Hollister’s name recommended adding an explanatory parenthetical, the kind of redundancy the real Hollister actively edits out.
Training an AI on someone’s published work can approximate their voice. It cannot replicate their editorial judgment. And presenting that approximation under a real person’s name, with a checkmark badge and the word “expert,” misrepresents what the feature actually delivers.
Why This Matters
This isn’t just a product quality issue. It raises real questions about consent, identity, and the emerging norm of using public figures’ names to add credibility to AI outputs. The professionals named in this feature had no say in how their identities are being used or what advice is attributed to them.
What stands out here is the scale: this isn’t one or two high-profile names borrowed for marketing. It’s dozens of working professionals, some of whom are still actively publishing, whose professional reputations are now attached to AI-generated feedback they’ve never reviewed.
For AI practitioners and product teams, this story signals a pressure point that’s only going to grow. Using publicly available content as training data is one thing. Putting a real person’s name on the output, and calling it expert guidance, is another.
The full investigation, including screenshots and Superhuman’s complete statement, is available at The Verge AI.