When AI Wears Your Name, Who Owns Your Voice?

Somewhere, a grieving student is being advised to open their essay with a quote from a dead sociologist—advice generated by an AI that has never grieved, never studied, and never actually read the sociologist in question.

That's not a hypothetical. That's roughly what happened when Grammarly quietly launched—and then just as quietly killed—a feature called "Expert Review."

What Grammarly Did, and Why It Matters

Last week, Wired reported that Grammarly, the ubiquitous writing-assistance app, had rolled out a feature that offered users AI-generated editorial feedback supposedly inspired by famous authors, working journalists, and academics. The names attached to this advice included Joan Didion, John McPhee, Susan Orlean, Stephen King, and a roster of less-famous but very real people—journalists and scholars who had never agreed to lend their names, their reputations, or their intellectual frameworks to a commercial product.

Some of those academics were recently deceased.

The feature lasted only days before Grammarly CEO Shishir Mehrotra pulled it, posting an apology on LinkedIn: "We hear the feedback and recognize we fell short on this." Shortly after, journalist Julia Angwin—one of the writers whose name had been used—filed a class-action lawsuit against Grammarly's parent company, Superhuman Platform. The company's response: the legal claims are "without merit" and it will "strongly defend against them."

The feature is gone. The questions it raised are not.

The Advice Was Terrible. That's Almost the Point.

One Atlantic journalist spent hours testing the tool before it went dark, trying to summon an AI version of herself. She couldn't. Instead, she received feedback "inspired by" Leslie Jamison, Sherry Turkle, and philosopher Amia Srinivasan. The suggestions were, without exception, useless: wordy additions, fabricated quotes, a fake story about her late grandmother inserted mid-draft, and a several-sentence capsule history of the feminist movement wedged into a paragraph that had merely mentioned the word "girlboss."

When the tool was asked to review a passage inspired by Joan Didion's The White Album, it cited Didion's most famous line—"We tell ourselves stories in order to live"—and interpreted it as an uplifting endorsement of personal narrative. Anyone who has actually read Didion knows the line is about something far darker: the self-deceptions we require just to get through the day.

The AI didn't just borrow Didion's name. It misread her entirely.

Sponsored

Advertise with Us

[email protected]

Sponsored

And yet: the tool's failure is almost reassuring in the wrong way. Because the more important question isn't whether this particular AI was bad at editing. It's what happens when one isn't.

The Anxiety Underneath

This story broke through the noise not because of what Grammarly did, but because of what it touched. There is a widespread, low-grade dread among writers, journalists, editors, and academics about what the AI economy means for their livelihoods—and more existentially, for the value of their particular kind of intelligence.

For years, people in creative and verbal professions were told to "learn to code." The taunt was cultural shorthand for a real economic shift: technical skills were ascendant, and people who worked in words seemed increasingly precarious. Now the dynamic is inverting in strange ways. Large language models are remarkably capable at coding, math, and structured reasoning. Peter Thiel, one of Silicon Valley's most influential voices, said in a 2024 interview that the post-AI labor market would be "much worse for the math people than the word people."

But the Grammarly episode complicates any triumphalism. The company didn't try to replace writers with AI. It tried to commodify writers—to extract the perceived value of their names and reputations and sell it back to users as a product feature, without paying or even asking. That's a different kind of threat. Not replacement, but appropriation.

Three Problems, One Feature

The legal and ethical issues here stack up quickly. First, there's the right of publicity—using a living person's name and likeness for commercial purposes without consent is actionable in most U.S. jurisdictions, which is presumably why Angwin's lawsuit exists. Second, there's the question of posthumous voice—what ethical framework governs the AI reanimation of scholars and writers who can no longer consent or object? Third, and most philosophically thorny, is the quality problem in reverse: the feature was bad, so the harm was limited. But if the impersonation had been convincing, the harm would have been worse—and harder to identify.

The creative and legal communities are watching closely. The Authors Guild and various journalism organizations have been pushing for clearer rules on how AI companies train their models and deploy famous names. The Grammarly case is unusual because it didn't involve training data—it involved using real people's identities as a marketing interface. That may require entirely different legal frameworks than the copyright suits currently working through U.S. courts.

What Imitation Actually Requires

There's a version of this story that ends with a simple moral: AI can't replicate human creativity, so relax. But that conclusion is too easy, and the Grammarly feature—bad as it was—points toward something more unsettling.

Learning to write has always involved imitation. You dissect the sentences of writers you admire. You feel the gap between their work and yours. That gap—the frustration, the aspiration, the slow closing of the distance—is how a voice develops. The process is inseparable from the result.

AI skips the process entirely. There is no frustration, no aspiration, no gap to close. There is only output. And when that output wears the name of a real person, it doesn't just raise legal questions. It raises a question about what we think we're paying for when we pay for writing at all.

Is the value in the words on the page? Or in the mind and experience that produced them?

For now, most readers and editors would say both. But that answer may not be stable.