• AI use at work ranges from under 10% in some states to nearly 25% in others, with sharp gaps even between neighbors.

• Census data shows AI adoption depends less on geography and more on industry mix, workforce age and firm size.

• Regions where AI use spreads broadly instead of narrowly are more likely to see shared economic gains, according to new research.

Montanans are using AI for work at more than twice the rate of neighboring South Dakota. Nearly a quarter of Vermonters report using AI on the job, while fewer than 1 in 10 Louisianans do.

This striking state-by-state divergence comes from newly released US Census Bureau Household Pulse Survey data, which estimates the share of adults using artificial intelligence in their work in June 2025. The numbers range from roughly 9% in Louisiana to nearly 25% in Vermont, with sharp contrasts even among states that share borders.

States vary widely in their industry mix, firm size, workforce age and job roles, all of which strongly shape whether AI tools are even relevant to daily work.

At first glance, patterns broadly track expectations: states with large tech workforces and higher rates of educational attainment tend to report higher AI use, with lower use reported in states with older populations or labor concentrated in sectors like agriculture, hospitality or traditional manufacturing.

But the magnitude of the gaps, especially between neighbors, seems jarring. 

When survey data shows such stark differences across adjacent states, I’ve been taught to assume something might be off in the methodology. In this case, however, the explanation appears less about faulty measurement and more about who is being measured. 

States vary widely in their industry mix, firm size, workforce age and job roles, all of which strongly shape whether AI tools are even relevant to daily work.

The most recent surveys don’t really fit regional tech stereotypes.

Among the country’s 25 biggest regions, Charlotte, San Diego and Orlando show the highest usage in summer 2025 data, the most recent collected by the Census Bureau. Boston and Los Angeles had among the lowest usage, alongside Baltimore and Philadelphia, with Detroit at the very bottom.

Differences seem less about economic dynamism than industry and population makeup. Even those factors don’t explain it all. Finance is one of the heaviest industry uses of AI, yet New York City is toward the bottom in reported adoption. Fewer than 1 in 5 tech-heavy San Francisco businesses reported using AI. St. Louis was closer to 1 in 4.

This matters. A recurring finding in the technology-diffusion literature is that when new tools spread slowly beyond early adopters, productivity gains concentrate in a small set of frontier firms and regions. When diffusion is broader, the benefits are more likely to spread. Anthropic’s September 2025 Economic Index report explicitly documents that early AI adoption is concentrated geographically and by task, showing that logic in real time. 

The trend points to more benefits for current economic winners, worsening inequality. It doesn’t have to be this way.

Wider spread of AI adoption → greater economic impact

Regions where new technologies become normalized across many roles, instead of just elite technical ones, are more likely to see productivity gains translate into wider wage growth and resilience. 

For more residents to benefit from an emerging technology, a 2013 paper argued, leaders need to remove “friction” by incorporating programming and shaping culture. People need to want this future. Alas, public opinion research from Ipsos shows large demographic and occupational gaps in how Americans perceive and use AI, with comfort levels varying sharply by education, income and job type.

This helps explain why raw adoption percentages are only a starting point. Two states might report similar overall AI usage while seeing very different outcomes depending on whether those tools are confined to white-collar knowledge work or embedded across small businesses, construction firms, logistics operations and frontline services.

It also complicates national narratives about AI “stalling” or “plateauing.” Much of the debate relies on surveys — including the Census Bureau’s Business Trends and Outlook Survey — that capture behavior unevenly across firm size and sector. Many workers may not even realize they are using AI as those features are quietly absorbed into everyday software.

The takeaway from the state-level data is not that some regions are “ahead” and others are “behind” in a simple sense. It’s that AI’s economic impact will likely depend less on headline adoption rates and more on how evenly those tools are diffusing through local labor markets.

In that sense, the map is less a scoreboard than a warning. Where comfort with AI remains narrow, so too may the benefits.