AI Is Probably Not (Yet) the Reason for Labor Market Weakening

On Friday we will learn how the labor market fared in April. Regardless of exactly where payroll growth and the unemployment rate land, what’s clear from the recent data is that the labor market has cooled substantially from where it was three years ago. Payroll employment growth has been relatively weak over the prior year, at only about 20,000 net new jobs per month.1 From its post-pandemic low of 3.4% in April 2023, the unemployment rate has risen to 4.3% in March 2026—a higher though still healthy level. Both layoffs and hires have been low, with unemployed job seekers having an especially difficult and worsening experience.  

As we all try to make sense of this labor market, many have suspected that it bears the imprint of widespread AI adoption by employers. While this would probably not explain low layoffs, it is a reasonable conjecture for at least two reasons. First, the capabilities of LLMs are advancing at an impressive, though increasingly difficult to measure pace and have already become useful for many workplace tasks. Second, the timing of labor market cooling appears, on the surface, to align with the emergence of increasingly powerful LLMs.2

However, the weight of the evidence—to date—does not support this conjecture. In new analysis released today, The Budget Lab uses the microdata underlying monthly employment reports to search for any labor market effects of AI. A relatively recent methodological innovation, called synthetic differences-in-differences, helps to distinguish any genuine impact of AI on exposed occupations from the persistent differences between exposed and unexposed jobs. 

Those differences are important to reckon with. Some are clear from prior analysis: workers in AI-exposed occupations tend to be more highly educated and are more likely to be women. Others are less well-studied, like the sensitivity of exposed jobs to the normal business cycle. We find that, prior to the pandemic, AI-exposed occupation employment is considerably less cyclical than unexposed employment. An econometric strategy that accommodates this kind of difference is helpful for truly understanding the effects of AI.

When we apply our preferred strategy, we find no strong evidence of impacts as of yet. The full analysis has more detail, but two key graphs are excerpted below. Figure 1 shows the impact of AI on the employment of the average exposed occupation, expressed as a share of the population.3 The estimate is close to zero and cannot be distinguished from it, statistically speaking. The same is true for Figure 2, which shows impacts on inflation-adjusted hourly wages. 

These results complement earlier analysis by The Budget Lab showing no unusual rise in occupational “churn”, i.e., the degree to which workers report that their occupations have changed over time. If AI were displacing a substantial number of workers by automating their jobs, one would expect this to result in many displaced workers eventually taking new jobs in new occupations. One feature of that previous analysis is that it does not rely on AI exposure metrics and is not subject to critiques of those measures. However, it does not formally attempt to identify effects of AI. We consider it helpful to take a variety of different methodological approaches to the same basic questions about AI impact. 

The Budget Lab will continue to process incoming data from the monthly employment report to understand how the economy is evolving. AI seems quite likely to eventually leave its mark on the labor market, even if it has not already. As policymakers consider their options in the post-AI economy, understanding exactly when and to what extent AI is affecting the economy will be particularly important.