Official productivity statistics for the second quarter of 2026 do not yet isolate any effect of artificial intelligence. The U.S. Bureau of Labor Statistics reported that nonfarm business labor productivity rose 1.4 percent at an annual rate in that quarter and 2.2 percent over the year, while Statistics Canada reported a 1.0 percent quarterly gain for Canadian businesses. Neither agency attributes its figures to AI, and survey estimates of time saved by workers are much smaller than the claims often attached to the technology.
Key takeaways
- U.S. labor productivity has grown at an annualized rate of 2.1 percent since the fourth quarter of 2019, the same as its long-term rate since 1947 and above the 1.5 percent of the 2007 to 2019 cycle.
- A St. Louis Fed estimate from 2024 survey data put the aggregate productivity gain from generative AI at about 1.1 percent, based on self-reported time savings.
- Canadian business-sector productivity rose 1.0 percent in the second quarter of 2026 while hours worked fell 0.1 percent.
- Productivity data measure output per hour for the whole economy, so a rise cannot be traced to one technology without separate evidence.
What the 2026 productivity releases report
The Bureau of Labor Statistics published its revised second-quarter 2026 productivity release on September 3, 2026. Output in the nonfarm business sector rose 1.7 percent at an annual rate and hours worked rose 0.3 percent, which produced the 1.4 percent productivity gain. Compared with the same quarter a year earlier, productivity was up 2.2 percent. Unit labor costs rose 1.2 percent in the quarter.
The same release gives longer comparisons. Since the fourth quarter of 2019, productivity has grown at an annualized rate of 2.1 percent, with output up 2.5 percent a year and hours up 0.4 percent. Between the fourth quarter of 2007 and the fourth quarter of 2019, the rate was 1.5 percent. The long-run rate since the first quarter of 1947 is 2.1 percent. The current cycle therefore matches the postwar average after a slower decade.
Statistics Canada released its figures for the same quarter on September 3, 2026. Productivity of Canadian businesses increased 1.0 percent, hours worked fell 0.1 percent after a 0.4 percent rise in the previous quarter, and unit labor costs rose 0.6 percent.
Why a national number cannot be credited to AI
Productivity is output divided by hours. It rises when firms produce more with the same labor, but also when low-productivity jobs disappear from the average, when prices change, or when capital other than software is added. The BLS release does not break its total into causes, and neither does the Statistics Canada release. A 2.1 percent cycle average that equals the 1947 rate is consistent with many explanations and gives no way to separate them.
Economists have faced this problem before. Robert Solow remarked in a 1987 book review that computers appeared everywhere except in the productivity statistics. Jack Triplett of the Brookings Institution later reviewed the explanations offered for that gap, among them mismeasurement of output in sectors such as finance and long delays before a new technology shows up in the data. The same questions apply to AI tools now, and they are the reason aggregate data lag behind firm-level reports.
What survey estimates imply for the total
A February 27, 2025 post by Alexander Bick, Adam Blandin, and David Deming at the Federal Reserve Bank of St. Louis used nationally representative worker surveys fielded in August and November 2024. About 28 percent of U.S. workers reported using generative AI at work to some degree. Users said it saved 5.4 percent of their work hours, roughly 2.2 hours of a 40-hour week. Spread across all workers, including those who do not use it, the saving was 1.4 percent of hours, and the authors computed a rough aggregate productivity gain of about 1.1 percent.
The authors list limits. Time savings were self-reported, and saved time that workers spend on other tasks or rest does not raise measured output. Formal adoption by firms was also low at the time, so much of the use was informal. A one-time gain of about 1 percent, if real, would be small beside the 2.2 percent rise in the year to the second quarter of 2026 and would not show clearly in one quarter's data.
Reading productivity claims: a comparison table
| Level of evidence | Example in this article | What it can show | Main limit |
|---|---|---|---|
| National accounts | BLS and Statistics Canada productivity releases | Output per hour for the whole business sector | No attribution to any technology |
| Worker survey | St. Louis Fed estimate from 2024 surveys | How much time users say they save | Self-reported, tied to 2024 tools |
| Historical review | Solow (1987) and Triplett (Brookings) | Why new tools can precede measured gains | Describes an earlier technology |
Questions before crediting AI for a productivity gain
- Does the figure measure output per hour, or only time saved on a task?
- Is the comparison with a group that did not use the tool, or only with an earlier period?
- Did hours worked change, since a fall in hours raises the ratio without any new output?
- Does the source say which quarter or year the tool belongs to?
- Are costs of review, training, and licenses counted against the gain?
FAQ
Has AI raised U.S. productivity?
Official data cannot answer yes or no. Productivity has grown at the long-run rate since 2019, and survey-based estimates put the gain from generative AI near 1.1 percent as of 2024, with the caveats above.
Why do company reports show larger gains than national statistics?
Company reports usually cover a chosen task or team, while national statistics average over every business. Gains in one task can be offset by extra review work or can be absorbed in a firm without changing total output.
Do the Canadian numbers show a different picture?
Statistics Canada reported a 1.0 percent gain for the second quarter of 2026 and credited a rise in business value added. The agency does not link the figure to AI, and one quarter is a weak basis for any conclusion about technology.
When will clearer data arrive?
Revised releases and annual benchmarks arrive over time, and the BLS and Statistics Canada publish new quarterly figures on fixed schedules. Studies that follow identified groups of firms give more direct evidence than aggregates.
Related reading
- Cost savings that firms report from AI
- How companies measure savings from AI
- AI and economic growth
- AI and job opportunities
More in AI and Society.
Last verified: October 2026
Sources
- Productivity and Costs, Second Quarter 2026, Revised, U.S. Bureau of Labor Statistics, September 3, 2026 (source page)
- Statistics Canada, The Daily, second-quarter 2026 release on business-sector productivity, hourly compensation, and unit costs, September 3, 2026
- The Impact of Generative AI on Work Productivity, Bick A, Blandin A, Deming D, Federal Reserve Bank of St. Louis, February 27, 2025 (source page)
- The Solow Productivity Paradox: What Do Computers Do to Productivity?, Triplett J, Brookings Institution, March 1, 1999