Why AI’s job market impact remains so hard to measure

The Fed estimated that about 18 percent of U.S. firms had adopted AI by year-end 2025, while a separate survey-based estimate found 41 percent of individuals reporting work-related generative AI use as of November 2025. One dataset points to layoffs, another to retraining, and a third to broad worker use, yet none of them are measuring exactly the same thing. The public debate keeps swinging between panic and optimism while policymakers and employers make decisions with partial visibility.

The measurement problem is the story

The Federal Reserve put the core problem plainly: AI adoption can look sharply different depending on the survey question, the respondent, and whether researchers ask firms or workers. Those differences may reflect sampling, framing, information gaps, and even social desirability bias, which means people may answer in ways that do not fully match what they do on the job.

Employer and worker surveys already tell very different stories. The same economy can look relatively cautious from the employer side and much more saturated from the worker side.

The Fed also estimated that 78 percent of the labor force works at firms that have adopted AI, and about 54 percent works at firms that use large language models. A technology can spread through workplaces even when firm-level adoption appears modest. The technology’s current use is concentrated in cognitive and analytical work, especially in professional services and financial sectors, which makes it harder to detect through broad labor statistics that are designed to track the entire economy at once.

Job postings are not flashing a clear AI alarm

Labor-demand data have not produced a clean signal either. A New York Fed analysis using Lightcast job-posting data and an occupational AI-exposure measure found little indication of a distinct AI-driven decline in labor demand. Overall hiring has slowed since ChatGPT’s late-2022 release, but postings for AI-exposed occupations did not fall disproportionately.

A slowdown in hiring is not the same as AI-driven displacement. In the New York Fed’s Second District, more firms reported retraining workers in AI-exposed occupations than reducing hiring, which points to adjustment inside jobs rather than broad destruction of them. The pattern fits a labor market in transition, but not one with a single obvious AI break point.

The Budget Lab at Yale reached a similar conclusion in its October 2025 assessment. Measures of exposure, automation, and augmentation showed no sign of being related to changes in employment or unemployment, and the broader U.S. labor market had not experienced a discernible disruption since ChatGPT’s release 33 months earlier.

Exposure is not the same as displacement

International comparisons reinforce how easy it is to misread AI risk. The International Labour Organization updated its global exposure index in 2025 by combining task-level data, expert input, and AI-model predictions, and concluded that globally one in four workers is in an occupation with some generative AI exposure. It also found that 3.3 percent of global employment sits in the highest-exposure category.

That does not mean those jobs are disappearing. The ILO’s earlier analysis found the dominant effect of generative AI likely to be augmentation rather than automation, especially in clerical work.

The gender split in the ILO data also shows why broad averages can hide important differences. Female employment is more exposed than male employment in the highest-exposure category, at 4.7 percent versus 2.4 percent.

Workers are using AI, even if layoffs are harder to pin down

Household data suggest the technology is already inside the workday. The Federal Reserve’s 2025 household well-being report, released in May 2026, found that one in four workers used generative AI at work in the prior month. Eighty-one percent of users reported that it saves them time, and users were more likely to say it improves their careers than to worry it will replace their jobs.

The same report showed a labor market that is still solid, but less comfortable than before. Concerns about finding or keeping a job increased, layoffs ticked up slightly, and more adults under 30 said not being able to find a job kept them from working.

The Fed’s April 2026 reporting also found that one in four workers had used generative AI at work in the prior month.

Layoff trackers catch headlines, not always causation

Private layoff trackers add more noise to the picture. Challenger, Gray & Christmas recorded 60,620 U.S.-based job cuts in March 2026, up 25 percent from February, with AI-related cutting leading all reasons that month. Its May 2026 report put job cuts at 97,006, with AI leading layoff reasons for the third consecutive month.

Those figures are employer-cited reasons, not a direct accounting of AI-caused displacement. A company can cite AI when it is reorganizing work, cutting costs, or consolidating teams, and it can also avoid saying AI when the technology is doing most of the restructuring. These reports can undercount or overstate the real effect depending on how firms describe the cut.