New research from the Federal Reserve Bank of St. Louis, analyzing nearly 490,000 corporate earnings call transcripts, confirms what official data has indicated for three years: artificial intelligence has not yet produced a measurable increase in aggregate productivity. Because productivity growth is the single largest determinant of long-run wage gains and living standards, that absence has significant implications for how broadly AI's benefits will be distributed—and when. However, one of the paper's authors raised a more disquieting possibility—that AI may already be generating real gains that remain structurally invisible, because the technology itself is eroding the value of the very outputs it has made abundant.

The mechanism is straightforward. When AI radically reduces the cost of producing a given output, that output simultaneously becomes less valuable. The productivity arithmetic effectively cancels itself out: gains on one side of the ledger are offset by falling prices on the other. Anyone can now generate marketing materials, animations, or even a passable news article with a keystroke—but if everyone can, none of it commands the value it once did. The task became easier; the output became cheaper. Somewhere in that exchange, a genuine gain vanished from the statistics without ever registering as a loss. This echoes a long-recognized challenge in national accounting: GDP and productivity statistics were designed for an industrial economy where outputs are physical goods sold at market prices, and they have always struggled to capture the value of free or near-free digital services.

"Some things are going to become more abundant," said Serdar Ozkan, one of the paper's authors. "That means they're also going to become probably less valuable."

What the Data Shows

Economists Ozkan and Aakash Kalyani, along with research associate Nicholas Sullivan, scanned approximately 490,000 earnings call transcripts from 5,198 publicly traded U.S. firms spanning 2000 to 2025. They used an AI model to tag sentences referencing productivity and AI. The share of productivity commentary tied to AI climbed from near zero before ChatGPT's late-2022 launch to roughly 15% of all productivity discussion by the close of 2025.

About 95% of AI-related productivity sentences describe gains that executives expect in the future rather than gains already realized—a proportion that has remained steady since 2023. When executives do characterize AI's impact, they are nearly unanimous in their optimism: 95% describe productivity as rising, compared with 75% for non-AI-related commentary.

Researchers Say This Is Exactly What History Predicts

Ozkan said he was not surprised by the future-tense findings, given that aggregate data had already shown no meaningful productivity increase once capital investment was accounted for. He cited economist Robert Solow's well-known remark that "you can see the computer age everywhere except but in the productivity statistics," drawing a direct parallel to electrification, which he said required "several decades" to reorganize factories, retrain workers, and transform workflows before its productivity dividend became visible in the data.

Stanford economist Erik Brynjolfsson identified this phenomenon as the "productivity paradox" in a 1993 paper for MIT, and has recently described the current landscape as the modern sequel.

Kalyani, who has separately studied diffusion patterns across general-purpose technologies, said the economics profession has largely reached consensus on this point following the initial surge of post-ChatGPT enthusiasm: "The aggregate gains will be in the future, whereas what you see right now is a lot of investment and a lot of excitement and optimism for the future."

He noted that technology diffusion across regions, occupations, and firms is "extremely slow," typically unfolding over 20 to 30 years. Compressing AI's lag to just three to five years "would be a huge change" from historical precedent. Computers, consistent with Solow's observation, did not appear in productivity data until the late 1990s and early 2000s.

The Slow-Diffusion Consensus

The findings align with other 2026 Federal Reserve research. A Kansas City Fed analysis found that the recent productivity pickup in official data is "not yet broad-based," with a small subset of industries accounting for most of the gains even as AI adoption continues to spread. Separately, Fed Chair Kevin Warsh told Congress in July that AI "hasn't displaced workers" so far and has made them "a bit more productive," but cautioned that "the long term can be quite far out."

Previous St. Louis Fed research similarly estimated that generative AI represented only a 1.1% increase in productivity by late 2024 relative to 2022—modest when compared with the 2.3% and 1.6% overall productivity growth the U.S. economy posted in 2024 and 2023, respectively.

Firms Are Putting Money Behind the Optimism

The St. Louis Fed team emphasizes that the corporate commentary is not empty rhetoric. Kalyani said the researchers prioritize actions over words: "we trust what people do, not what they say." Firms that discuss AI positively have also increased research and development, capital expenditures, and overall investment—a correlation that did not exist when the team first examined AI mentions in an earlier 2024 post skeptically titled "AI Hype or Reality?" but has since strengthened.

A related San Francisco Fed study found that AI-positive firms experienced substantially higher investment and R&D growth by 2025 than other public companies, with the gains concentrated among the largest technology firms building AI infrastructure.

What the Statistics Cannot Capture

Ozkan's abundance argument is compounded by a second constraint: bottlenecks that AI cannot resolve. No matter how rapidly AI accelerates research, drafting, or analysis, two people still need to schedule and attend a meeting—and that step moves at exactly the same pace it did four years ago. Productivity is not a single metric; it is the output of an entire chain, and AI has accelerated only some of the links.

When asked what signal would ultimately confirm that AI-driven productivity gains had materialized, Kalyani acknowledged that the honest answer is that no one knows in advance. The application that ultimately matters is discovered through trial and error, spreading firm by firm and worker by worker in a process that appears almost random from the outside, even as it accumulates into something measurable in aggregate.

He offered the example of Google Maps and the taxi medallion. The mapping app briefly made owning a New York taxi medallion one of the most valuable assets in the city. Then Uber arrived and disrupted the entire system within a few years. No one predicted which navigation app would dismantle taxi monopolies. The pattern recurs with every general-purpose technology: winners and losers are determined by a chaotic, decentralized process that defies prediction, even when the technology's eventual significance is obvious in retrospect.

That is the core insight embedded in nearly half a million tagged sentences of earnings-call optimism. It is not that executives are wrong to expect AI to deliver returns. Rather, whether those returns manifest as measured productivity growth—or simply dissipate into cheaper, more abundant, and less valuable output—may remain unknowable until it has already occurred.

This story was originally featured on Fortune.com.