The Idea
Over New Year's 2022, Edward Tian, a 22-year-old Princeton senior writing his thesis on AI detection, was sitting in a Toronto coffee shop during winter break with too much free time on his hands. ChatGPT had launched a month earlier, and his classmates were already using it to write essays. Tian, a computer science major minoring in journalism, kept circling one belief: there were aspects of human writing a machine should never be able to fake, and people deserved to know when the words in front of them weren't human.
So over a few days, using GPT-2 and a laptop, he built a tool that scored text on two metrics, perplexity and burstiness, to guess whether a human or a bot wrote it. On 2 January 2023 he tweeted out a rough beta, expecting a few dozen people to poke at it. He went to sleep. By morning, thousands had signed up and the app had crashed the platform hosting it. This was the beginning of GPTZero.
The Execution
- January 2023: Tian tweeted the beta on 2 January. The post racked up over 7 million views, the app pulled 30,000 uses in its first week, and Streamlit had to allocate extra servers to keep it from collapsing under the traffic.
- January 2023: Two weeks after launch, Tian recruited a few Princeton alumni and shipped GPTZeroX, adding sentence-level highlighting of suspected AI text. The API followed in February.
- May 2023: Tian turned down VC calls from A16Z, Menlo and others before raising $3.5m in seed funding led by Uncork Capital, teaming up with high-school friend Alex Cui as CTO and going full-time after graduating.
- 2023: The criticism arrived fast. The Washington Post and Ars Technica flagged false positives, noting the tool could wrongly accuse students, and that highly regular human text (even the US Constitution) sometimes got tagged as AI. Tian leaned into it, building "Origin", a feature letting students prove they wrote their own work.
- June 2024: GPTZero raised a $10m Series A led by Footwork's Nikhil Basu Trivedi, with Reach Capital, Jack Altman's Alt Capital and Neo joining. The kicker: it was already profitable. Total raised across its life was just $13.5m.
- 2024–2025: The product widened well past an essay checker into hallucination detection, citation verification and a feature called Replay that records a document keystroke by keystroke. Tian claimed a false-positive rate under 1% across 20 languages.
- June 2026: Superhuman, the company Grammarly formed after buying the email client and rebranding, acquired GPTZero. Terms weren't disclosed, but Tian said the company had hit 19 million registered users and $30m in ARR.
The lesson?
Tian never went looking for a market. He had a week of free time, a thesis topic, and one tweet, and the demand found him overnight. The best founders don't manufacture urgency, they spot a problem the whole world is suddenly anxious about and ship the rough version before anyone else. GPTZero raised just $13.5m, stayed profitable, and exited into the arms of the very kind of AI-writing tool it was built to police. When your product becomes the antidote to the thing everyone's building, the people building it become your buyers.