Hardly a day goes by without a story about generative AI and what it means for jobs and economic growth. Yet the relentless focus on AI in the workplace may be obscuring something equally important: how the technology is changing the way people get things done at home.
A new working paper by Miao “Ben” Zhang, assistant professor of finance and business economics at USC Marshall School of Business, along with co-authors Michael Blank of Stanford Graduate School of Business and Gregor Schubert of UCLA’s Anderson School of Management, is among the first studies to examine this largely overlooked dimension of the AI revolution. Using Internet browsing data from more than 200,000 U.S. households tracked between 2021 and 2024, the researchers find compelling evidence that the home — not the office — is currently where generative AI is delivering its largest productivity gains.
The study focuses on ChatGPT adoption and its effect on what the researchers call “productive” digital tasks — activities undertaken online that are not purely for enjoyment, such as searching for a job, planning travel, or managing household finances. Their findings show that ChatGPT users completed these tasks between 76% and 176% more efficiently than non-users. The saved time did not, however, flow into education or skill-building. Instead, users spent their newfound hours on leisure: posting to social media, streaming videos, and other recreational activities.
“The large efficiency gains might sound surprising at first — but they actually make a lot of sense,” Zhang said. “Unlike at work, people doing tasks at home aren’t trained professionals. They’re jumping between websites for various tasks, piecing things together. ChatGPT can replace all of that with a single conversation.”
Alongside those productivity gains, the researchers identify a troubling trend: a widening gap in who is actually using these tools. Younger and wealthier Americans have been far quicker to adopt generative AI than their older and lower-income counterparts. More concerning still, the researchers’ data suggest that this “GenAI digital divide” is not closing; it is widening.
That disparity carries real economic stakes. Much of the public optimism surrounding generative AI rests on the idea that tools like ChatGPT can function as a great equalizer, a “PhD in everyone's pocket” that gives lower-skilled workers access to capabilities once reserved for the highly educated. But that promise only holds if lower-income individuals are actually using these tools. The study’s evidence suggests that many are not, which may mean they are missing out on the potential economic benefits of AI-assisted job searching, financial planning, or skills development.
The researchers emphasize that one of the central hopes for generative AI has been its potential to extend capabilities across the socioeconomic distribution, particularly for those at the lower end. Their findings, however, suggest that the people who stand to benefit most are currently the slowest to adopt these tools.
