Grammarly pulls its controversial AI author personas

Imagine typing out an essay, only to have Grammarly serve up critique ostensibly written by a bestselling novelist or a famed scientist, without either of them knowing their names were on the docket.

That was not a hypothetical thought experiment. It was a real feature inside Grammarly, and on March 11, parent company Superhuman officially pulled the plug after a massive backlash and a looming class action lawsuit.

Turning dead and living authors into free bot personas

The feature, dubbed Expert Review, was rolled out in August. It worked by tapping third-party LLMs and scouring publicly available web data to generate targeted feedback on your writing from influential perspectives.

Depending on what you wrote, your digital peer reviewers could range from legendary historical thinkers to contemporary fiction authors and tech bloggers. The catch? Living or dead, these writers never gave permission, nor were they ever contacted by Grammarly.

Superhuman tried to protect itself with a classic corporate disclaimer, noting that references to real people were purely informational and did not imply any official endorsement. But writers quickly noticed their names attached to AI-generated takes, and they were furious.

Superhuman initially tried to cool tempers by offering an opt-out form. That plan backfired almost instantly. Expecting busy writers to track down opt-outs, let alone expecting deceased thinkers to rise from the grave to fill out paperwork, went down predictably poorly. With an attempted class action lawsuit landing on its doorstep, CEO Shishir Mehrotra took to LinkedIn to announce the feature is now disabled while they reassess.

Why it matters

We think this is a masterclass in how not to deploy generative AI, and it exposes the legal landmines waiting for companies that treat real people as free training prompts.

For users and writers, having an AI critique your prose through different lenses is genuinely useful. Tons of people already ask ChatGPT to edit like Hemingway. But there is a massive legal and ethical gulf between emulating a tone and slapping a real living professional's byline on synthetic text without their consent.

For the broader industry, this pushes the copyright and likeness debate beyond training sets and straight into product outputs. Tech companies have spent years playing fast and loose with crawlers, but packaging living experts as automated features crosses into personality rights and trademark territory. Relying on an opt-out policy and a buried disclaimer was an astonishingly reckless gambit that no product team should ever copy.