OpenAI study finds ChatGPT Enterprise token use rose sevenfold in nine months
- ChatGPT Enterprise output tokens grew roughly sevenfold from June 2025 to March 2026; firms that adopted between January 2024 and June 2025 nearly quadrupled usage, accounting for about half of the total growth.
- The six-month post-adoption worker sample covers more than 1,500 organizations and 17 million messages, linking Enterprise records with job titles and message-level task classifications under privacy-preserving methods.
- Among U.S. public companies, ChatGPT Enterprise adopters are larger and more valuable than non-adopters, with higher R&D and SG&A spending before adoption.
- Use within adopting firms spans job functions and seniority levels, while early-career workers have especially high usage intensity.
- Messages cover writing, technical work, communication, and information synthesis; the paper says firms still vary widely in how quickly and broadly they integrate AI into workflows.
Hacker News opinions
I initially balked at the 69-page length, but the main text is only 23 double-spaced pages. The rest is tables, figures, and references.
I wish OpenAI's frontier-lab papers read less like stakeholder marketing. DeepMind's AGI to ASI paper was better written, even if I have problems with its execution.
This is a standard working-paper preprint for the field, not a white paper. Wait for journal review if that is the standard you want.
I've noticed a recent surge in articles telling me AI is doing great. It feels like there is anxiety about getting the story out before investors cash out.
Large companies may appear to adopt Enterprise more because they have controls against mixing personal and corporate AI use. My girlfriend gets real value from a personal ChatGPT account for lesson plans, slides, test-data analysis, and culturally representative materials because her district has no AI program or budget.
I hope all PII is stripped before student testing data goes into a personal account.
My wife uses AI similarly as a teacher. I am skeptical of AI as the teacher, but AI as a teacher assistant seems more plausible.
The teachers I know see AI as dishonest in lesson planning and risky for grading or data analysis because of privacy. They mostly deal with students using it for homework, and think it has made grading worse and test scores lower.
This reads like an ad saying enterprises need to buy more AI until they discover what it is good for. Calling the product general purpose avoids making a sharper case against an established alternative.
My takeaway is that there is no measurable ROI yet. Enterprises and OpenAI still do not know how to measure it, so the pitch is to keep buying while everyone figures it out.
I am worried about the adopter comparison: the authors randomly sample matched adopters but use all Compustat U.S. companies as non-adopters without reporting the sampling rate. They also may count 10-license pilots as Enterprise adoption.