Research check · 17 October 2026
AI use reaches many jobs, but that is not a layoff count
A U.S. worker survey found generative AI use in more than 80% of occupations. That means at least one in five workers in those occupations reported work use; it does not mean most tasks or jobs were replaced.
What was measured
Bick, Blandin, Deming and Schumacher surveyed nearly 14,000 U.S. workers across four quarterly waves from August 2025 to May 2026. Respondents identified which of their occupation's ten most important O*NET tasks they perform and which tasks generative AI regularly helps them complete. In more than 40% of measured tasks, at least one in five workers who perform the task reported using AI for it. Fewer than 3% of tasks had a majority of workers reporting AI use.
Why the denominator matters
These are shares of workers within occupations or within tasks, not shares of jobs lost. An occupation can have many AI users while each uses it for different parts of the job. The authors found exposure scores predicted some, but far from all, of the differences in adoption. Reported use is a better starting point for studying outcomes than capability alone, but this survey did not estimate an effect on hires, headcount, wages or hours.
What remains open
The survey relies on self-report and a selected list of ten tasks per occupation. It does not show how often AI was used, whether output improved, or how an employer changed staffing. We inspected the authors' detailed public explanation and working-paper abstract, not the full PDF. The next test is to link adoption to labor outcomes in comparable workers or firms, with alternative explanations addressed.