'We don't want to be a policeman of the Internet': How a John McPhee student turned GPTZero into a $30 million AI-detection business

Edward Tian built GPTZero over winter break in his senior year at Princeton, right after ChatGPT went mainstream. He put it on a free Streamlit page and, within a week, the site was crashing under traffic from 30,000 people. The pitch was simple: show how much of a piece of online writing looked machine-made. That simple pitch turned into a real business fast.

Less than four years later, GPTZero says it has more than 20 million registered users and $30 million in annual recurring revenue. In June, Superhuman, the productivity company once known as Grammarly, agreed to buy it for undisclosed terms. PitchBook valued the company at more than $88 million, and someone familiar with the deal said the sale returned at least 10 times the $13.5 million GPTZero had raised. Tian and cofounder Alex Cui, friends since high school in Toronto, now run a combined “authenticity” team inside Superhuman.

Tian, now 26, didn’t start out trying to build a startup at all. He wanted to be a journalist. Before GPTZero took off, he had worked with Bellingcat, spent time on BBC Eye Investigations, and accepted a fellowship with the New York Times visual investigations team. He also learned from John McPhee at Princeton, where he says the writer emphasized drafts, editing, and process. Tian still visits McPhee at home and says their conversations often circle back to AI and writing. When he showed McPhee an AI-generated version of one of his articles, Tian says the response was blunt: it was terrible.

Cui had his own run at online trust problems years earlier. As a Caltech undergraduate in late 2017, he helped build Open Mind, a browser extension aimed at fake news and propaganda. It got attention on Fox & Friends and NPR, and he presented it to Facebook’s misinformation team and Senate staff. But it never scaled into a bigger product. When Tian’s project started getting traction, Cui was working on a master’s in computer science and machine learning at the University of Toronto and decided it was time to drop everything and launch.

Now GPTZero is being used far beyond classrooms. The company says it found more than 100 hallucinated citations in papers accepted at NeurIPS, and scanned 300 of the roughly 20,000 submissions to ICLR, finding more than 50 more. ICLR later became a client. GPTZero has also flagged errors at Deloitte, KPMG, EY, and PwC, and said about one in five words spoken at the UN General Assembly’s 81st session came from speeches it classed as AI-generated. Inside Superhuman, the company’s tools are being folded into a broader authenticity push that reaches beyond detection and into plagiarism checking and writing replay. Tian seems keen to move the debate on. He doesn’t want to be the internet police. He wants people to ask better questions than just whether something was written by AI.