Artificial intelligence is increasingly influencing the economy, yet accurately measuring its effects remains a significant challenge, experts say. Policymakers, researchers, and economists highlight that without reliable data, crafting effective economic and labor policies related to AI will be difficult.
Nathan Goldschlag, director of research at the Economic Innovation Group, emphasized the importance of measurement in policy decisions. In a report published Thursday, he outlined the challenges in assessing AI’s economic impact and recommended steps for improving data collection and analysis. His call reflects growing attention in Washington, where a bipartisan group of senators introduced legislation in June to expand federal data collection and require annual reporting on AI’s influence on the labor market.
Senator Mark Kelly, a sponsor of the bill, stressed that “millions of Americans’ lives and millions of businesses” depend on informed decisions, which in turn rely on accurate data. While some government efforts exist—the Census Bureau began biweekly surveys of AI adoption in 2023 and has included related questions intermittently in its annual business survey—data remains patchy and inconsistent.
Economists currently rely on multiple AI exposure measures that analyze job descriptions to estimate which occupations are most affected. However, these indicators often provide conflicting results. A recent study by economists at Northwestern University and American University found that depending on the exposure measure used, AI’s impact on employment could appear either positive or negative. Michelle Yin, one of the study’s authors, compared the situation to receiving “three different diagnoses for the same condition.”
Much of the difficulty stems from existing economic data frameworks that predate the digital era and AI’s rapid evolution. Standard reports from the Bureau of Labor Statistics lack detailed breakdowns of technology-sector jobs or specific occupations vulnerable to automation. For example, the latest detailed occupational data currently available dates back to May 2025, limiting its usefulness amid fast-paced AI adoption.
Despite these constraints, economists value government data for tracking AI’s effects over time. Researchers at Yale’s Budget Lab are using metrics such as “occupational churn”—the rate at which job types within industries change—to serve as early indicators of AI’s influence on hiring patterns. Martha Gimbel, executive director of the lab, described these efforts as essential for “measuring this and figuring this out in real time.”
Yet, the federal statistical system faces challenges, including declining survey response rates and budget constraints. Erika McEntarfer, the former Bureau of Labor Statistics commissioner, pointed out that modest funding increases could enhance data quality and better capture economic shifts driven by AI.
In parallel, several research teams have developed tools using private-sector data. The Stanford Digital Economy Lab recently unveiled a dashboard drawing on payroll data from ADP, revealing sharp drops in entry-level jobs in AI-exposed sectors since the release of ChatGPT in 2022. Erik Brynjolfsson, the lab’s director, described this trend as a potential warning sign comparable to the Industrial Revolution’s labor market disruptions. However, other private data tell a more nuanced story. Analysis by Ramp and Revelio Labs indicated that companies heavily investing in AI were adding jobs faster than slower adopters, suggesting that the impact varies by industry and adoption strategies.
Economists generally agree that AI’s broad economic effects have been limited so far, consistent with historical patterns of technological adoption that initially reduce productivity before eventual gains. Goldschlag characterized the current period as one of experimentation, with AI tools still becoming increasingly useful.
Separating AI’s effects from other recent economic disruptions—such as the COVID-19 pandemic, inflation, shifting immigration policies, and changes in work arrangements—remains a complex task. McEntarfer noted that while data might clarify some trends over time, predicting AI’s long-term impact, especially over the next five to ten years, is unlikely.
As AI adoption grows, improving data quality and measurement methods remains a priority for researchers and policymakers seeking to understand its evolving role in the labor market and broader economy.