Superhuman buys GPTZero, the $30 million AI-detection startup

Edward Tian built GPTZero over winter break of his senior year at Princeton, in the weeks after ChatGPT went mainstream. It went up on a free Streamlit hosting page and crashed within a week as 30,000 people flocked to it. Less than four years later, the company had more than 20 million registered users and $30 million in annual recurring revenue — and in June, Superhuman, the productivity company formerly known as Grammarly, agreed to acquire it.

The path there ran through Tian's other life. He had worked with investigators at Bellingcat and on the BBC Eye Investigations team, and had accepted a fellowship with the New York Times' visual investigations team when GPTZero took off. He studied under John McPhee, Princeton's Ferris Professor of Journalism, and says he still visits him at home at least twice a year, where their conversations often turn to AI and writing. Cofounder Alex Cui had been working on online trust and integrity since 2017, when he built Open Mind, a browser extension aimed at "fake news." He later said the team failed to capitalize on its early viral moment with a bigger product.

GPTZero's detection work has already landed in institutional settings. ICLR became a client after GPTZero screened 300 submissions, and its "Chasing the Hallucinations" blog series has documented errors published by Deloitte (Australia), KPMG (International), EY (Canada), and PwC (Middle East). The latest finding: about one in five words spoken at the UN General Assembly's 81st session came from speeches GPTZero classed as AI-generated. Whether that scrutiny travels intact into a productivity suite used by tens of millions of writers — and whether Tian's stated hope to move past "Is this AI or not AI?" takes hold — is likely to depend on how the combined authenticity team positions detection inside Superhuman Go.