Getting AI’s Workforce Impact Right Starts With Better Data

For centuries, policymakers and the public have worried that automation could replace human workers. Yet past waves of automation have done more than eliminate jobs; they have reshaped how workers perform tasks and created opportunities for higher-skilled work. These concerns, once centered on steam-powered machines or vehicle manufacturing, now focus on artificial intelligence (AI), which plays an increasingly important role in every industry.

In response, Congress is considering legislation to study AI’s impact on the U.S. workforce, while states such as New York have proposed bills that would require employers to report how they use AI and how it affects hiring. Collecting this data is an important step toward understanding AI’s economic effects, which go well beyond hiring and layoffs. Rather than imposing new reporting requirements on businesses, policymakers should commission studies and expand existing workforce surveys to better capture AI’s full impact on the workforce, not just changes in headcount.

Much of today’s legislative discussion surrounding AI’s workforce impact focuses on whether AI creates or eliminates jobs. Those outcomes matter, but they often emerge later in the adoption process than AI’s earliest effects. In most organizations, AI first changes individual tasks, redistributes responsibilities, and alters workflows. For example, a marketing analyst may use AI to generate a report’s first draft, while an engineer may use it to accelerate software development. Those employees can then use the time saved to complete more tasks or take on new types of work. Measuring only hiring and layoffs overlooks the operational changes and productivity gains that often precede shifts in employment.

To address this gap, studies and statistical surveys should capture whether AI automates routine tasks, augments existing responsibilities, or enables employees to shift toward higher-value work. Surveys should also include clearer occupational classifications that distinguish between jobs in which AI primarily augments workers’ duties and those in which it automates a substantial share of tasks. Studies and surveys should capture AI’s primary functions, such as decision support, content generation, and customer interaction, to show how adoption patterns vary by industry. Without consistent definitions of AI use across occupations, comparisons among firms, industries, and regions become difficult and can obscure how AI is actually reshaping work.

Studies should also examine how AI affects different demographic groups within the workforce. Workers ages 18 to 22 who are just entering the labor market may experience AI adoption differently from more experienced professionals because entry-level roles often rely more heavily on routine tasks that AI can augment or automate. Tracking these differences could help policymakers understand how AI affects career entry, advancement, and workforce development over time. These distinctions would reveal more about AI’s long-term labor market effects than employment counts alone.

Additionally, statistical agencies should collect better information on how organizations adopt AI. Two companies in the same industry may report similar employment outcomes despite using the technology in fundamentally different ways. One firm may deploy enterprise AI systems integrated into daily operations, while another may experiment with a limited pilot in a single department. Surveys should capture AI adoption maturity, including whether deployment remains experimental, is limited to specific functions, or has become organization-wide. They should also identify which employees have access to AI tools, how organizations govern their use, and whether workers receive formal training before rollout. Without this context, workforce statistics reveal little about why AI affects employment or why some firms achieve greater productivity gains than others.

A common blind spot in the conversation around AI’s impact on the workforce is that AI adoption often begins from the bottom up rather than through formal corporate initiatives. Employees increasingly use publicly available AI tools to draft content, summarize documents, or analyze data before employers formally implement enterprise AI systems. As a result, company-level data alone may significantly understate AI adoption across the workforce. Statistical agencies should therefore complement employer surveys with worker surveys that capture how employees independently incorporate AI into their daily responsibilities. Measuring both organizational deployment and employee-driven adoption would provide a more comprehensive view of AI use across the economy.

Finally, any workforce legislation should recognize that successful AI adoption depends as much on investments in people as on investments in technology. Many firms purchase AI software, but realizing its benefits often depends on whether employees receive training to integrate these tools into existing workflows. Collecting data on workforce upskilling, AI training, and human capital investments would help policymakers understand why some organizations achieve greater productivity gains than others. Policymakers should use these findings to encourage workforce training alongside AI investment rather than focusing primarily on technology adoption.

As AI continues to improve, policymakers need better information to understand how the technology impacts industries, firms, and workers, particularly so they can identify opportunities to improve U.S. competitiveness and productivity through effective AI adoption. That requires moving beyond measures of hiring and displacement toward data that captures how AI transforms tasks, organizational practices, employee behavior, and workforce skills. By collecting richer, more detailed data, legislators can build a stronger foundation for future workforce policy while giving businesses, workers, and governments a clearer picture of AI’s evolving economic impact.