What happened
Federal Reserve Bank of St. Louis researchers analyzed roughly 490,000 earnings call transcripts from 5,198 U.S. firms between 2000 and 2025 and found that while AI mentions tied to productivity rose to roughly 15% of all productivity discussion by end of 2025, approximately 95% of those mentions describe expected future gains rather than gains already realized.
Why it matters
The research reveals a structural invisibility problem: when AI makes output radically cheaper to produce, that output simultaneously becomes less valuable, causing productivity gains to cancel out in the statistics. One author, Serdar Ozkan, notes that because 'some things are going to become more abundant,' they 'are also going to become probably less valuable'—meaning real economic gains may exist but disappear into lower prices rather than show up as measured productivity growth, just as happened during the computer revolution.
What to watch
Researchers say the timeline for AI productivity payoff likely mirrors electrification and computerization, which took 'several decades' and didn't show up in productivity data until the late 1990s and early 2000s respectively. A Kansas City Fed analysis found the recent productivity pickup in official data is 'not yet broad-based,' with a small set of industries accounting for most gains even as AI adoption spreads.
Summaries like this, in your inbox every morning.
The Federal Reserve Bank of St. Louis study anchors its findings in a well-established historical pattern. When economist Robert Solow famously observed in the 1980s that 'you can see the computer age everywhere except but in the productivity statistics,' he was describing a phenomenon now known as the productivity paradox—initially named by Stanford economist Erik Brynjolfsson in a 1993 MIT paper. The current AI situation mirrors that lag: executives express near-uniform bullishness (95% describe rising productivity for AI commentary versus 75% for non-AI), yet official aggregate data show no meaningful productivity bump once capital investment is accounted for. Previous St. Louis Fed research estimated generative AI represented only a 1.1% increase in productivity by late 2024 relative to 2022, modest against the 2.3% and 1.6% overall productivity growth posted in 2024 and 2023 respectively.
The study's most striking insight is Ozkan's explanation for why measured gains may never appear: abundance destroys value. When AI makes marketing materials, animations, or news stories nearly free to produce, the output floods the market and loses its economic worth. The productivity gain—the ability to create more with less effort—gets erased not as a loss but as a permanent price decline. This mechanism is structurally invisible to standard productivity accounting, which measures output relative to input. A second constraint compounds the problem: AI cannot speed up bottlenecks that lie outside its domain. A meeting still requires two people to show up at the same time, no faster than before. Productivity is a chain, and AI has only accelerated some of the links.
Researchers emphasize that the lag is not evidence of hype but a predictable phase in technology adoption. Firms are putting real capital behind AI optimism—the pattern of positive AI mentions correlating with increased investment strengthened between their 2024 skeptical analysis ('AI Hype or Reality?') and 2025 findings. As one author, Kalyani, noted: the honest answer to when AI productivity gains will finally be visible is that 'nobody knows in advance.' Discovery happens through decentralized trial and error, firm by firm, often in ways that resist prediction—as exemplified by Google Maps' unforeseen disruption of New York taxi medallions.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
