Artificial intelligence (AI) is changing how businesses operate and how workers perform their jobs. But policymakers still lack the data needed to understand the full effects of these changes on employment, wages, productivity, and the broader economy. As AI adoption accelerates, federal statistical agencies should improve how they measure these changes while making better use of increasingly valuable private-sector data.
The federal government has made a useful start. The Census Bureau’s 2026 AI supplement to its Business Trends and Outlook Survey measured AI adoption across firms, business functions, and individual worker tasks. It found that 18 percent of businesses used AI in at least one business function, representing 32 percent of U.S. employment. Among businesses adopting AI, 57 percent used it in three or fewer functions, most commonly sales and marketing, business development, and information technology. Only a small share reported AI-related changes in overall employment.
These statistics provide an important snapshot, but not a comprehensive picture of how AI is changing work. Knowing whether a business uses AI does not reveal which workers use it, which tasks it changes, how much time it saves, or how it affects hiring and wages. AI may automate some tasks, augment others, and create demand for new skills. Its effects will vary across industries, occupations, firms, and workers, making aggregate employment figures insufficient to capture the changes.
Recent research illustrates the measurement challenge. A Stanford study using ADP payroll records found that employment among workers ages 22 to 25 in highly AI-exposed occupations was 19 percent below the level it would have reached had it followed the employment trend among their less-exposed peers. The researchers found that this pattern primarily reflected reduced hiring rather than increased layoffs. A separate study by Ramp and Revelio Labs, linking AI-related business spending to workforce records at more than 21,000 firms, found that companies making larger AI investments subsequently experienced stronger employment growth, including growth in entry-level positions.
These findings are not necessarily contradictory. The Stanford research examines employment trends across occupations with different levels of AI exposure, while the Ramp research examines employment changes at firms with different levels of actual AI investment. Neither establishes that AI alone caused the observed outcomes. Together, they demonstrate why policymakers need better information about technology adoption and its effects on businesses and workers.
Private-sector data are increasingly valuable for this purpose. Payroll processors, recruiting platforms, business software providers, and financial technology companies can observe changes in hiring, compensation, job requirements, and technology spending in greater detail than traditional government surveys often allow. But their data is frequently proprietary and limited to particular customer populations. They cannot substitute for consistent, representative national statistics.
Federal statistical agencies should build on existing efforts and take three steps to close these gaps.
First, make Census Bureau AI data collection recurring and link it to employment outcomes. The Census Bureau should continue its AI supplement to the Business Trends and Outlook Survey and expand its ability to track how AI adoption changes over time. Where legally permissible, it should link business responses to existing longitudinal employment and earnings records. The Longitudinal Employer-Household Dynamics (LEHD) program, for example, connects employer and worker records to provide information about employment, earnings, and worker movement across firms and industries.
The Confidential Information Protection and Statistical Efficiency Act (CIPSEA) provides a framework for designated federal statistical agencies to share confidential business data for statistical purposes, subject to strict confidentiality safeguards. Agencies should make greater use of this authority to improve economic statistics. Linking AI adoption data with employment records could help researchers determine whether businesses introducing AI in particular functions subsequently change their hiring, workforce composition, or wages, and whether they use AI to complement employees or substitute for particular tasks.
Second, add AI-related questions to the Bureau of Labor Statistics’ (BLS) Job Openings and Labor Turnover Survey (JOLTS). JOLTS already tracks job openings, hires, and separations. Adding questions about AI’s effects on hiring plans could reveal whether employers leave positions unfilled, reduce planned openings, change job qualifications, or create new roles. This would be particularly useful for understanding changes in entry-level hiring, occupational demand, and the skills employers seek.
Third, improve measurements of how AI changes the tasks workers perform, and the time required to complete them. Employment statistics alone cannot capture productivity gains when workers accomplish more in the same amount of time or when AI changes a job without eliminating it. The BLS American Time Use Survey offers one avenue for better measurement. Agencies should explore adding consistent questions about AI use, the tasks it supports, and the time it saves. This would help policymakers distinguish automation from task augmentation and assess whether productivity improvements translate into higher wages, reduced working hours, or expanded business output.
Federal agencies should complement these efforts through partnerships with private-sector data providers. Payroll processors could provide timely employment and compensation information, recruiting platforms could identify shifts in job requirements, and financial technology firms could offer insight into AI investment. With appropriate privacy protections and consistent methodologies, these sources could supplement federal surveys and administrative records. Prioritizing existing surveys, administrative data linkages, and reliable private-sector information would also help agencies modernize statistical infrastructure without creating entirely new data systems.
AI is likely to reshape the economy in ways that traditional employment statistics were not designed to capture. Better federal statistics, complemented by private-sector data, would give policymakers, employers, workers, and researchers a more accurate and timely understanding of these changes. That information is essential to developing policies that encourage innovation, support workers, and help the U.S. economy realize the full benefits of AI.
Image credit: Generated with DALL-E
