If your hiring plan assumes AI has already delivered a productivity windfall, you are budgeting against a future tense that even public-company executives use on earnings calls.
Researchers at the Federal Reserve Bank of St. Louis (Serdar Ozkan, Aakash Kalyani, and Nicholas Sullivan) analyzed about 490,000 earnings-call transcripts from 5,198 publicly traded U.S. companies covering 2000 through 2025. They used an open-source model to flag productivity sentences, check whether those sentences also mentioned AI, and classify whether the speaker described a past, present, or future gain.
The result that should stop a freeze meeting cold: about 95% of AI-related productivity sentences referred to future gains. Roughly three-fourths of non-AI productivity sentences were also forward-looking, but AI talk is almost entirely “still to come.” Tone matches the grammar: 95% of AI-related productivity sentences described productivity as increasing, versus about 75% for non-AI sentences.
Firms are “investing, experimenting and reorganizing around AI today, while the measurable productivity effects remain mostly ahead.”
That is the St. Louis Fed team’s own summary in AI and Productivity: What Firms Are Saying on Earnings Calls.
The macro check that hiring freezes ignore
Optimism has not shown up as a broad productivity boom. The researchers point to Federal Reserve Bank of San Francisco data showing utilization-adjusted total factor productivity grew just 0.07% over the four quarters ending in the first quarter of 2026.
A related San Francisco Fed look by Kalyani and Huiyu Li found that firms with positive AI sentiment on earnings calls have posted higher capital spending and R&D growth than other public companies. That combo is concentrated among large technology firms building AI infrastructure. In other words: money is moving into AI buildout. Aggregate productivity is not yet moving with it.
Language share tells the same story of talk leading results. The share of productivity-related sentences that also mentioned AI was near zero before ChatGPT, climbed through 2023, plateaued in 2024, then accelerated again in 2025 to about
