AI usage data underscore rapid uptake, but long-term impact remains uncertain

New analyses from OpenAI and Anthropic offer a snapshot of how people are using today's AI tools, showing rapid uptake and a range of applications that defy easy early predictions.

OpenAI says ChatGPT grew from 1 million registered users in December 2022 to 100 million people using it weekly by November 2023, and now counts more than 750 million weekly active users. The two firms describe complementary patterns: ChatGPT’s top uses are practical advice and tutoring (about 28.3%), editing or generating text (28.1%), and information-seeking queries (21.3%). For Claude, usage centers on computing and math tasks (36.9%), with an increasing share in educational instruction and library work (12.7%).

Geography and income matter: richer economies show higher uptake, and middle-income countries such as Brazil use ChatGPT nearly as much as the United States. The data provide a snapshot, not a forecast, but they illustrate how users are experimenting with the technology across tasks today.

Economists and historians alike have long warned that early signals can mislead. The IBM parable is invoked to remind readers that early budgeting and forecasting often miss where technology will end up. In 1955, SAGE and other military contracts accounted for the lion’s share of IBM’s computer revenue, while private administrative computing lagged. Yet within a few years, private-sector revenue surged, and by the early 1960s it dwarfed military spending at the same company. The AI data presented here show a different tempo—diffusion that appears unusually rapid for a general-purpose technology, but with many unknowns about how it will translate into productivity and employment markets.

Scholars often cite diffusion dynamics to explain why technology spreads unevenly within and across regions. Zvi Griliches’s classic analysis of hybrid corn showed a fast uptake within states but slower diffusion across state lines. By contrast, the OpenAI and Anthropic data suggest diffusion of AI today is happening at a historically fast clip across many economies, which could compress the window for adaptation. That acceleration raises both opportunities and risks for workers, firms, and policymakers, and it also means traditional lagged benefits may appear sooner than expected.

Taken together, the reports underscore what we can learn—and what we cannot—from current usage data. For ChatGPT, roughly half of all messages touch a relatively small, clearly defined set of work activities such as documenting information and making decisions or solving problems. At the same time, humans perform many tasks beyond those categories, which AI tools may or may not automate in the near term. Claude’s usage, meanwhile, shows substantial automation in some contexts, even as augmentation remains significant elsewhere. These patterns suggest AI could function as a complement in many roles, while also enabling more autonomous task execution in others.

Ultimately, the key takeaway is not a single forecast but a cautionary note about interpretation. The authors of both reports stress that their findings describe present usage and narrow historical comparisons, not long-run effects on jobs, wages, or growth. Still, the rapid diffusion observed today stands in contrast to many past general-purpose technologies, which took longer to diffuse and adapt to. As with the IBM case decades ago, empirical data can illuminate how AI is used now, while leaving open the fundamental questions about its broader social and economic impact.