The impacts of generative AI on the economy, and the labor market in particular, are hot topics. Only now are we beginning to understand those effects empirically. Anthropic’s Economic Index offered one of the first empirical views into AI work use by mapping interactions with Claude to occupations. That report showed more than one-third of use in computer and mathematical tasks.
A recent NBER study used ChatGPT prompts to identify how people use large language models. They found that generative AI activity is most common in analytical, writing, and problem-solving tasks.
But these studies measure the distribution of AI use, not which occupations use AI disproportionately relative to their size in the workforce. A large share of AI use in an occupation could simply reflect that occupation’s large employment base. Conversely, seemingly modest AI activity might represent intensive adoption if that occupation employs relatively few workers. Understanding AI’s economic impact requires adjusting for workforce size.
This analysis introduces an “AI Intensity” measure in which each occupation’s share of AI use divided by its share of total U.S. employment. An intensity score above 1.0 indicates disproportionately high adoption while a score below 1.0 indicates lower-than-expected use given workforce size.
The results reveal highly concentrated use. Writers and media professionals have the highest AI intensity of any major occupational group, using it at 12 times their employment share. Scientists, researchers, and postsecondary educators show similarly elevated rates (5-11x). Meanwhile, large employment sectors show minimal AI engagement, including healthcare practitioners at 0.2x their workforce share, business operations specialists at 0.2x, and K-12 teachers at 0.3x, despite employing 9.4%, 6.7%, and 2.2% of U.S. workers respectively.
These aren’t just differences between knowledge work and manual labor. Even among professional occupations where AI capabilities seem well-matched to job tasks, adoption varies by factors of 20-60x. Postsecondary teachers show 5.5x intensity while K-12 teachers show 0.3x. Writers show 12x intensity while lawyers show only 0.7x.
Data and Method
Anthropic’s Economic Index provides anonymized usage data from Claude.ai and its enterprise API in August 2025. Anthropic classifies each interaction by task and maps it to the O*NET taxonomy. I focus on the Claude chatbot data because Anthropic separates it out by country, making it possible to examine only the United States and thus to map it to occupation data from the Bureau of Labor Statistics. More specifically, I match the Anthropic occupational task data to 2024 BLS employment counts by 2-digit (“major” groups) and 3-digit (“minor” groups within major group) occupation codes.
For each occupation i, define:
where (Share of AI Usage)i is the share of all Claude tasks in occupation i, and (Share of Employment)i is the share of total U.S. employment in occupation i.
Values greater than one imply that the occupation uses AI more than its share of employment would suggest if AI use were evenly distributed, while values below one imply the opposite.
Results
The following table shows the top ten highest AI Intensity occupations by 3-digit codes:
Media and communication workers, writers, and authors show the highest AI intensity at 12.3x their employment share, followed by life and social science technicians (11.0x) and social scientists (9.7x). Postsecondary teachers (5.5x) and other educators also show elevated adoption.
Substantial variation exists even within knowledge work occupations. Postsecondary teachers show 5.5x intensity while K-12 teachers show just 0.3x, which is an 18-fold difference within the education sector. Healthcare practitioners, which represent 9.4% of U.S. employment, show only 0.2x intensity despite being knowledge workers who could benefit from AI-assisted diagnosis and documentation.
At the other extreme, occupations representing large parts of the labor market show minimal AI engagement. These include, not surprisingly, retail sales workers (5.4% of employment) at 0.3x intensity, business operations specialists (6.7% of employment) at 0.2x, and food service workers (7.8% of employment) at 0.01x.
Some Results Are Weird. Why?
Computer and information scientists barely crack the top ten. That seems wrong, and it probably is. Two aspects of the data are important for interpreting the results.
First, Claude may not be representative of all AI use. By one measure, Claude has one percent of market share while ChatGPT has 80 percent, with Claude users appearing to be relatively concentrated among more tech-savvy users. However, this user type should bias results away from tasks like writing, media, and communications, not towards it, lending credence to the results here.
Second, software development shows relatively low AI Intensity, which seems surprising, particularly given the bias discussed in the previous paragraph. This result most likely shows that programmers tend not to use chatbots, but instead purpose-built platforms like Github Copilot, which connect to major models via APIs.
Anthropic’s own data is consistent with this hypothesis. While Anthropic only released global API use without breaking it down by countries, comparing that to global Claude use shows that while software development was 38 percent of Claude and writing 14 percent, software development was 45% of API use, followed by data processing at 11 percent, and then writing at 8 percent. In other words, software developers rely heavily on generative AI, but not on chatbots, per se.
Still, setting aside software, unless some occupations rely disproportionately on ChatGPT or any AI other than Claude, these results should be broadly accurate.
Conclusion
This analysis reveals that AI adoption, when adjusted for workforce size, is more concentrated than raw usage statistics suggest. While Anthropic’s Economic Index shows computer and mathematical tasks accounting for the largest share of AI chatbot use, normalizing by employment reveals a different picture: writers, scientists, and researchers in relatively small occupations use AI at rates 5-12 times higher than their workforce share would predict.
This concentration has important implications for how AI’s economic benefits and disruptions will be distributed. The occupations showing highest AI intensity employ a small fraction of American workers, perhaps 5-7% combined. Meanwhile, large employment sectors including healthcare (9.4% of employment) and business operations (6.7%) show little AI engagement, as, of course, do retail sales (5.4%) and food service (7.8%).
The pattern mirrors the NBER study results that large language models currently complement “elite cognitive work” rather than automating routine tasks. My empirical results confirm this using observed usage data, but add a crucial labor market dimension: AI’s current economic footprint is deep but narrow. Productivity gains, as well as disruption, from AI will likely accrue disproportionately to already-high-wage knowledge workers.
Whether this represents the early stages of broad diffusion or a persistent divide depends on factors beyond technology alone. Healthcare, education, and frontline services face sector-specific barriers like regulatory constraints, workflow integration challenges, and infrastructure gaps that won’t be solved by better AI models. The “missing middle” of office workers, business professionals, and administrators shows low adoption despite apparent task compatibility, suggesting barriers related to awareness, organizational change, and tool design.
Future research should track whether AI intensity patterns persist or broaden as capabilities improve and specialized applications proliferate. Longitudinal analysis linking AI intensity to wage and employment outcomes will help distinguish between augmentation and automation effects. And comparative studies across different AI tools, such as comparing chatbot usage to specialized applications like GitHub Copilot or industry-specific platforms, would provide a more complete picture of AI’s labor market footprint.
Appendix: Full Results
| 3-Digit Occupation | AI Intensity | Claude Share (%) | Employment Share (%) |
| Media and Communication Workers, Writers, and Authors | 12.3 | 6.8 | 0.5 |
| Life, Physical, and Social Science Technicians | 11.0 | 2.5 | 0.2 |
| Social Scientists and Related Workers | 9.7 | 2.8 | 0.3 |
| Mathematical Science Occupations | 7.5 | 1.7 | 0.2 |
| Life Scientists | 6.1 | 1.9 | 0.3 |
| Postsecondary Teachers | 5.5 | 4.8 | 0.9 |
| Physical Scientists | 5.0 | 0.9 | 0.2 |
| Librarians, Curators, and Archivists | 4.6 | 1.3 | 0.3 |
| Other Teachers and Instructors | 4.1 | 2.5 | 0.6 |
| Computer and Information Research Scientists | 3.3 | 25.7 | 7.7 |
| Art and Design Workers | 3.0 | 1.3 | 0.5 |
| Agricultural Workers | 2.8 | 0.5 | 0.2 |
| Drafters, Engineering Technicians, and Mapping Technicians | 2.3 | 1.0 | 0.4 |
| Entertainers and Performers, Sports and Related Workers | 1.9 | 0.9 | 0.5 |
| Other Sales and Related Workers | 1.6 | 0.5 | 0.3 |
| Counselors, Social Workers, and Other Community and Social Service Specialists | 1.5 | 2.8 | 1.8 |
| Electrical and Electronic Equipment Mechanics, Installers, and Repairers | 1.5 | 0.4 | 0.3 |
| Metal Workers and Plastic Workers | 1.4 | 0.7 | 0.5 |
| Other Educational Instruction and Library Occupations | 1.3 | 1.6 | 1.3 |
| Media and Communication Equipment Workers | 1.1 | 0.1 | 0.1 |
| Other Healthcare Practitioners and Technical Occupations | 1.1 | 0.2 | 0.1 |
| Supervisors of Building and Grounds Cleaning and Maintenance Workers | 0.8 | 0.2 | 0.2 |
| Other Office and Administrative Support Workers | 0.8 | 2.2 | 2.8 |
| Advertising, Marketing, Promotions, Public Relations, and Sales Managers | 0.7 | 0.6 | 0.8 |
| Lawyers, Judges, and Related Workers | 0.7 | 0.4 | 0.6 |
| Financial Specialists | 0.7 | 1.8 | 2.7 |
| Architects, Surveyors, and Cartographers | 0.7 | 0.1 | 0.1 |
| Other Food Preparation and Serving Related Workers | 0.6 | 0.2 | 0.3 |
| Sales Representatives, Wholesale and Manufacturing | 0.6 | 0.7 | 1.2 |
| Other Production Occupations | 0.6 | 0.3 | 0.5 |
| Secretaries and Administrative Assistants | 0.6 | 1.3 | 2.2 |
| Engineers | 0.6 | 1.0 | 1.8 |
| Sales Representatives, Services | 0.5 | 0.7 | 1.5 |
| Health Technologists and Technicians | 0.3 | 0.4 | 1.1 |
| Preschool, Primary, Secondary, and Special Education School Teachers | 0.3 | 0.7 | 2.2 |
| Other Management Occupations | 0.3 | 0.7 | 2.5 |
| Retail Sales Workers | 0.3 | 1.5 | 5.4 |
| Other Healthcare Support Occupations | 0.3 | 0.2 | 0.8 |
| Operations Specialties Managers | 0.2 | 0.7 | 2.7 |
| Supervisors of Office and Administrative Support Workers | 0.2 | 0.3 | 1.1 |
| Information and Record Clerks | 0.2 | 1.0 | 4.1 |
| Health Diagnosing and Treating Practitioners | 0.2 | 2.1 | 9.4 |
| Business Operations Specialists | 0.2 | 1.5 | 6.7 |
| Legal Support Workers | 0.2 | 0.1 | 0.3 |
| Tour and Travel Guides | 0.2 | 0.1 | 0.3 |
| Law Enforcement Workers | 0.2 | 0.1 | 0.5 |
| Construction Trades Workers | 0.2 | 0.1 | 0.7 |
| Other Construction and Related Workers | 0.1 | 0.0 | 0.1 |
| Other Personal Care and Service Workers | 0.1 | 0.5 | 3.6 |
| Top Executives | 0.1 | 0.3 | 2.8 |
| Food Processing Workers | 0.1 | 0.0 | 0.3 |
| Supervisors of Sales Workers | 0.1 | 0.1 | 1.0 |
| Supervisors of Food Preparation and Serving Workers | 0.1 | 0.1 | 1.0 |
| Cooks and Food Preparation Workers | 0.1 | 0.2 | 2.2 |
| Vehicle and Mobile Equipment Mechanics, Installers, and Repairers | 0.1 | 0.0 | 0.1 |
| Financial Clerks | 0.1 | 0.1 | 1.5 |
| Other Installation, Maintenance, and Repair Occupations | 0.0 | 0.1 | 1.5 |
| Material Recording, Scheduling, Dispatching, and Distributing Workers | 0.0 | 0.1 | 3.2 |
| Grounds Maintenance Workers | 0.0 | 0.0 | 0.7 |
| Nursing, Psychiatric, and Home Health Aides | 0.0 | 0.0 | 1.0 |
| Motor Vehicle Operators | 0.0 | 0.1 | 2.1 |
| Food and Beverage Serving Workers | 0.0 | 0.1 | 7.8 |
| Building Cleaning and Pest Control Workers | 0.0 | 0.0 | 0.6 |
Scott Wallsten is President and Senior Fellow at the Technology Policy Institute and also a senior fellow at the Georgetown Center for Business and Public Policy. He is an economist with expertise in industrial organization and public policy, and his research focuses on competition, regulation, telecommunications, the economics of digitization, and technology policy. He was the economics director for the FCC's National Broadband Plan and has been a lecturer in Stanford University’s public policy program, director of communications policy studies and senior fellow at the Progress & Freedom Foundation, a senior fellow at the AEI – Brookings Joint Center for Regulatory Studies and a resident scholar at the American Enterprise Institute, an economist at The World Bank, a scholar at the Stanford Institute for Economic Policy Research, and a staff economist at the U.S. President’s Council of Economic Advisers. He holds a PhD in economics from Stanford University.
