The launch comes as AI-generated writing becomes increasingly common across the internet, raising questions about how organizations should identify machine-generated content and how much confidence they should place in detection systems that can still produce incorrect results.
Chief executive Max Spero and co-founder Bradley Emi started Pangram in 2023 after meeting as undergraduates at Stanford, where they studied in the university's artificial intelligence lab. Following graduation, Spero joined Google while Emi worked at Tesla under AI researcher Andrej Karpathy before the pair decided to build technology focused on identifying AI-generated content after ChatGPT's release.
Spero said they believed widespread AI-generated material would become an unavoidable part of online communication and wanted to create tools for "the world that this technology will bring forth."
According to a recent analysis by digital marketing firm Graphite cited by Pangram, roughly half of news articles and blogs published online today are AI-generated, underscoring the demand for tools that can evaluate the origin of digital content.
Pangram's platform allows users to paste text into a chatbot-style interface that classifies it as AI-generated, human-written or AI-assisted. The company said it trained its model on millions of text samples, enabling it to improve its ability to distinguish between human and AI writing over time.
A 2025 study from the University of Chicago found Pangram achieved near-perfect accuracy when evaluating texts of roughly 500 words or longer, outperforming competing services including Originality.ai and GPTZero. The company acknowledges its models have historically been less effective on shorter passages and on content that combines human and AI writing, though it says Pangram 4.0 is designed to improve performance in those areas.
The study challenged a broader perception that AI detection systems are unreliable, a view that gained traction after OpenAI discontinued its own AI detection tool in 2023 because of accuracy limitations.
Pangram said most of its revenue comes from individual subscriptions priced at $20 per month. It has also signed agreements with universities, publishers and online platforms seeking to identify AI-generated content. Quora was among its early customers, and Substack recently partnered with Pangram to give readers the option to filter out AI-generated posts.
The company said its annual revenue run rate has grown to 35 times what it was a year ago. While Pangram did not disclose revenue figures, it said monthly users increased to 120,000 in June from 2,700 a year earlier.
In addition to text detection, Pangram's new AI image detection model is intended to broaden what the company describes as "cognitive security." Pangram says the model identifies fully AI-generated images with 99.5% accuracy, and the company plans to introduce an AI video detection tool later this year.
Menlo Ventures partner Deedy Das said he sees AI detection becoming a widespread requirement as generative AI becomes more common. "I see A.I. detection as a kind of universal need that everyone has," he said.
Despite the company's progress, Pangram acknowledges its technology is not infallible. Mixed human-and-AI content remains one of the more difficult scenarios for detection models, creating the possibility of incorrect classifications. Spero said Pangram 4.0 improves its ability to identify partially AI-generated text, including an example recently highlighted by Substack author Freddie deBoer.
Spero also drew a distinction between fully AI-generated content and AI-assisted writing, saying he believes it is "completely OK" for people to use AI as an assistant, even though he personally chooses not to. How schools, publishers and other organizations ultimately decide to handle AI-assisted content remains an open question.
This analysis is based on reporting from The New York Times.
Image courtesy of Pangram.
This article was generated with AI assistance and reviewed for accuracy and quality.
