The mechanism is simple. When AI makes some output radically cheaper to produce, that output simultaneously becomes less valuable, and the productivity math cancels itself: Gains in one column get erased by falling prices in another. Anyone can now generate marketing materials, animations, even a passable news story with a keystroke. But if everyone can, none of it commands what it used to. The task got easier; the output got cheaper. Somewhere in that trade, a real gain disappeared from the statistics without ever showing up as a loss.

“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 roughly 490,000 earnings call transcripts from 5,198 publicly traded U.S. firms between 2000 and 2025, using an AI model to tag sentences about productivity and AI. The share of productivity commentary tied to AI rose from near zero before ChatGPT’s late-2022 debut to roughly 15% of all productivity discussion by the end of 2025.

Approximately 95% of AI-related productivity sentences describe gains executives expect in the future, not gains already realized, a share that has held steady since 2023. When executives do describe AI’s effect, they’re almost uniformly bullish: 95% describe productivity as rising, compared with 75% for non-AI commentary.

Researchers say this is exactly what history predicts

Ozkan said he wasn’t surprised by the future-tense findings, since aggregate data already showed no meaningful bump in productivity once capital investment was accounted for. He invoked economist Robert Solow’s famous quip that “you can see the computer age everywhere except but in the productivity statistics,” drawing a direct line to electrification, which he said took “several decades” to reorganize factories, retrain workers, and change workflows before its productivity payoff showed up in the data.

Stanford economist Erik Brynjolfsson called this the “productivity paradox” in a 1993 paper for MIT, but lately has taken to describing the current situation as the modern sequel.

Kalyani, who has separately studied diffusion patterns across general-purpose technologies, said the profession has largely reached consensus on this point after the initial post-ChatGPT excitement faded: “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, and compressing AI’s lag to just three to five years “would be a huge change” from historical precedent. Computers, per Solow’s observation, didn’t show up in productivity data until the late 1990s and early 2000s.

The slow-diffusion consensus

The pattern lines up with other 2026 Fed research. 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 of the gains even as AI adoption spreads. Separately, Fed Chair Kevin Warsh told Congress in July 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 generative AI represented only a 1.1% increase in productivity by late 2024 relative to 2022—modest compared to the 2.3% and 1.6% overall productivity growth the economy posted in 2024 and 2023, respectively.

Firms are putting money behind the optimism

Crucially, the St. Louis Fed team says this isn’t empty talk. Kalyani said the researchers weight actions over words: “we trust what people, do not what they say,” noting that firms discussing AI positively have also increased R&D, capital expenditures and investment—a correlation that didn’t exist when the team first studied AI mentions in an earlier 2024 post skeptically titled “AI Hype or Reality?” but has since strengthened.

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

What the statistics can’t capture

Ozkan’s abundance argument sits alongside a second constraint: bottlenecks that AI simply cannot dissolve. However fast AI accelerates research or drafting or analysis, two people still need to schedule and show up to a meeting—and that step moves at exactly the same speed it did four years ago. Productivity is not one number. It’s the output of an entire chain, and AI has only sped up some of the links.

Asked what signal would finally prove AI’s productivity gains had arrived, Kalyani said the honest answer is that nobody knows in advance. The application that ends up mattering gets discovered through trial and error, spreading firm by firm and worker by worker, in a process that looks almost random from the outside even as it adds up to something real 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 came along and punctured the entire system within a few years. Nobody predicted which navigation app would end taxi monopolies. The pattern repeats with every general-purpose technology: Winners and losers get decided by a chaotic, decentralized scramble that resists prediction, even when the technology’s eventual importance is obvious in hindsight.

That’s the real takeaway buried in nearly half a million tagged sentences of earnings-call optimism. It’s not that executives are wrong to believe AI will pay off. It’s that whether the payoff shows up as measured productivity growth—or simply evaporates into cheaper, more abundant, less valuable output—may be unknowable until it’s already happened.