Generative AI now reaches 80 percent of occupations and more than 40 percent of tasks, but in most of these jobs fewer than half of workers actually use it, according to a new study published by economists in September 2026. The paper, titled "What Work Does Generative AI Do?" and authored by researchers from the Federal Reserve Bank of St. Louis, Vanderbilt University, and Harvard University, concludes that while adoption is wide-ranging, it remains shallow across most professions. The findings challenge vendor-published estimates of AI's workplace penetration by measuring how workers actually employ generative AI rather than relying on chat log analysis.
As of May 2026, 45 percent of US adults between ages 18 and 64 used generative AI for work, while 55 percent used it for non-work purposes, bringing overall adoption to 62 percent. Four out of five detailed occupations show adoption rates above 20 percent, but only one in six occupations exceeds 70 percent adoption. Usage is highest in management and professional roles, particularly those related to finance, business, and computers, where about 15 percent of occupations had adoption rates above 70 percent. Personal service occupations and jobs requiring manual activity or interpersonal interaction show the lowest adoption. When measured by tasks rather than occupations, only 2.8 percent of tasks had adoption rates above 50 percent, and none surpassed 70 percent.
The researchers highlight a significant measurement gap between their findings and those published by AI companies like Anthropic, Microsoft, and OpenAI. "GenAI is used in many occupations and tasks, but few of these exhibit very high adoption rates," the authors state. The discrepancy stems from chat log classifiers that link conversations to generic task descriptions such as "Edit written material or documents." When OpenAI data suggests 15 percent of chats involve such editing, those numbers don't align with the US Department of Labor's O*NET database, where only 2.4 percent of workers hold jobs that include that task. The result, according to the study, is that vendor exposure figures likely overstate how much generative AI is genuinely relevant to people's jobs.
The paper explains that chat log classifiers associate conversations with tasks that have generic, activity-based descriptions, creating a mismatch with actual occupational task databases. By analyzing Real-Time Population Survey data instead, the researchers measured how workers truly use generative AI in their roles rather than what AI companies infer from usage logs. This method reveals that most occupations show relatively low adoption rates despite AI's broad reach. The computer-oriented professions that dominate the high-adoption category represent a small fraction of the overall workforce, meaning the technology's practical impact remains concentrated rather than evenly distributed.
One key driver of AI adoption is experience, the study finds. Workers who begin using it in one domain tend to adopt it in other areas as well. The researchers argue it's at least as important to understand why some workers adopt AI while others don't as it is to identify which tasks the technology can handle. That insight points to a future where workplace AI integration depends less on technical capability and more on user familiarity and willingness to experiment. The broader implication is that companies betting on rapid, universal AI transformation may need to rethink their timelines, focusing instead on targeted adoption in high-value roles where the technology already shows traction. For business leaders, the challenge shifts from deploying AI everywhere to cultivating the conditions under which employees will actually choose to use it.