OpenAI says weekly users top 900 million as AI reshapes work

OpenAI said it now has more than 900 million weekly active users. Executives and academic researchers cited the figure during a discussion on how artificial intelligence is affecting work, productivity and economic measurement.

The number surfaced as economists and labour market specialists argued that AI's rapid adoption is not yet fully reflected in conventional indicators such as productivity, gross domestic product and employment data.

Rani Chatterjee, chief economist at OpenAI, described the issue as a gap between the visible uptake of AI tools and the less clear signals in labour market statistics. She pointed to strong use by consumers and businesses, alongside uncertainty about how that activity is translating into broader economic outcomes.

Daniel Rock, assistant professor at the Wharton School, said the mismatch should not be surprising. He compared the current moment with the long lag between the spread of computers and their appearance in productivity data.

"In 1987, Bob Solow famously said that you can see computers everywhere except in the productivity statistics. We may be seeing a version of that again. At first, organisations use AI to improve the work they already do. It takes longer to discover new use cases, products and services. That is when investments in these capabilities begin to pay off over the long term," Rock said.

He added that companies often go through an initial period of disruption and investment before gains show up in the data. That process includes spending on new processes, culture and organisational change.

Firm Adoption

Gregor Schubert, assistant professor of finance at UCLA Anderson, said a key issue is the gap between theoretical exposure to AI and actual implementation inside companies. Firms with stronger technical foundations appear to adopt the tools faster and use them more effectively.

Companies are still working out how to redesign jobs and workflows around AI, he said. In many cases, that means creating new planning, validation and oversight tasks rather than handing entire processes over to software.

"AI often works in the middle of a process rather than end to end. That means adding tasks at the start, such as planning and setup, and at the end, such as validation," Schubert said.

That shift is influencing hiring. Some employers are placing more weight on subject expertise and then training those workers to use AI tools, rather than leaving deployment entirely to IT teams.

Alex Martin-Richmond, labour economist on OpenAI's economic research team, said evidence from job advertisements suggests AI is already reshaping task composition in roles linked to coding and knowledge work. Many workers, he said, are still experimenting with the tools and learning how to use them effectively.

"Usage is expanding rapidly, especially in areas where AI capabilities are improving quickly. There are things AI can do well now that it could not do six months or a year ago. We describe this as a capability overhang: many people are using the tools, but are still figuring out how to reach the frontier of use and take full advantage of them in their workflow," Martin-Richmond said.

Home and work

The discussion also focused on AI use outside formal employment. Schubert said consumer chatbot use may exceed workplace use, creating economic value that is largely invisible in GDP because it happens in households rather than paid markets.

Examples included planning trips, researching health concerns, preparing shopping lists and handling other administrative tasks. Those activities may improve welfare and save time, but they are difficult to capture in official economic statistics.

Martin-Richmond said about 30% of consumer ChatGPT usage appears to be work-related based on available signals, suggesting the line between personal and professional use is already blurred. That could mean workers are learning to use AI at home and bringing those habits into the workplace.

Rock said this consumer experimentation may amount to training that benefits employers. He also argued that companies need cultures that reward experimentation while maintaining proper checks for quality and risk.

Policy questions

The panel said policymakers should focus on how quickly workers may need to move between jobs and whether existing support systems are suited to larger occupational shifts. Martin-Richmond said unemployment insurance in the US is reasonably well designed for short-term shocks, but may need changes if AI drives bigger transitions across occupations.

Schubert called for public investment in AI training and access, arguing that many workers will need sustained support to become confident with the tools. Firms alone, he said, cannot provide that training if the goal is to prepare the wider workforce.

Rock said governments should consider small-scale policy experiments that could be expanded if labour market disruption intensifies. He also said workers should be encouraged to move toward AI-related work, not just defend themselves against it.

Chatterjee said policymakers should identify which workers are most affected, use existing institutions where possible and improve measurement of long-term outcomes. "If we can do those things, we can direct policy to the people who need it most and build the evidence needed to stay aligned with AI's broader benefits," she said.