What is generative AI's actual aggregate productivity gain today?

Using their Real-Time Population Survey data on self-reported genAI time savings, Bick, Blandin and Deming report that "Between 1 and 5 percent of all work hours are currently assisted by generative AI," with "time savings equivalent to 1.4 percent of total work hours." From November 2024 data on hourly wages, weekly hours, and weekly genAI time savings, they "estimate that genAI currently increases aggregate labor productivity by 1.9%." Applying "an AI-exposure-adjusted labor share of 0.57, following Acemoglu (2024), the implied potential productivity gain is 1.1%." They compare this to Jon Acemoglu's (2024) separate estimate of a 0.7% aggregate productivity gain, calling their own figure "slightly higher."

The authors trace the gap to methodology, not disagreement about genAI's effect: "Our estimate is based on reported genAI use and time savings, while Acemoglu (2024) uses experimental evidence on productivity gains from genAI usage and task-based predictions of genAI exposure." Reconciling inputs narrows the gap further — "If we adjusted our estimate of labor cost savings from 33% down to the 27% from the experimental studies discussed in Acemoglu (2024), we would arrive at an aggregate productivity estimate of 0.9%" — leaving "the remaining 0.2 percentage point gap" attributable to differing adoption estimates: Acemoglu's expert-assessed wage-weighted exposure share of 4.6% of tasks versus the survey's measured usage share of 5.7%. Critically, the authors flag that even this 1.1% figure is a potential gain, not an observed one: "If workers are now able to complete the same tasks in less time without their employer's knowledge, they may take their time savings as on-the-job leisure, which would increase welfare but not productivity," and they cite Bonney et al. (2024) finding "only 5.4 percent of firms had formally adopted genAI as of February 2024" — so "these potential productivity gains from genAI may not show up in productivity statistics, at least for now."

That caution connects directly to Why do most enterprise AI pilots fail to deliver returns?: both findings point to a gap between worker-level time savings or usage and firm-level P&L or productivity statistics, though the two measure different things (self-reported individual time savings versus enterprise pilot outcomes). The estimate also stands in contrast to Does AI chatbot adoption change worker pay and hours?, where administrative records find no earnings or hours effect two years out even as task structure shifts — consistent with Bick, Blandin and Deming's own prediction that productivity gains may precede any visible economic payoff. It also contrasts with Did ChatGPT's release reduce freelance writing work and pay?, a different labor-market segment (competitive freelance gig work) where exposure to genAI coincided with falling earnings rather than the productivity-side gains modeled here.

The 1.1% figure is a projection built on self-reported time savings and an assumed labor-share parameter, not a measured change in output or wages, and the paper's own adoption-speed finding, in How fast is generative AI adoption compared to earlier technologies?, underscores that this is an early-adoption snapshot. The implication the authors draw is accordingly modest: that "substantial productivity gains from generative AI are possible," not that they have yet appeared in the productivity data.