Artificial Intelligence (AI) may be poised to raise productivity across various domains, including writing (Noy and Zhang 2023), programming (Peng et al. 2023), and research and development (Toner-Rodgers 2024; Korinek 2023). However, understanding the extent to which AI -- and generative AI in particular -- has been adopted as part of the production process remains an open question. This note reviews the extant surveys on AI adoption at both the employee and firm levels. Surveys of firms show a wide spread of adoption rates, ranging from 5 percent to about 40 percent. Surveys of workers show between 20 and 40 percent of workers using AI in the workplace, with much higher rates in some occupations like computer programming. While estimates of the level of AI uptake vary, measurement considerations partly explain the differences; more importantly, the available time series data all suggest rapid growth in adoption.
Our analysis begins by identifying and reviewing 16 surveys on AI adoption. These surveys originate from government agencies, NGOs, academics, and some private companies. The surveys were generally fielded from late 2023 to mid-2024. For each survey, we reviewed the results to extract the relevant data on AI adoption. This includes recording the number of respondents and distinguishing whether the survey was conducted at the firm level or the individual/employee level. We also document whether the surveys cover broad AI or specifically generative AI (genAI), given the recent interest in the latter technology. In addition, we record the timing of each survey, identifying whether the data collection occurred at a specific point in time or over a range of dates. Notably, only a few of these surveys were conducted on a recurring basis. A standout example is the U.S. Census Bureau's Business Trends and Outlook Survey (BTOS), which is conducted every two weeks. We mostly focus on the BTOS statistics from late 2023 and early 2024, when Census included additional detailed questions on AI adoption.
Table 1 provides a summary of the surveys included in our analysis. We have six distinct firm-level surveys, which are the first seven rows of the table. With the exception of BTOS and the Bar Association, the results cluster between 20 percent and 40 percent. Almost all the firm-level surveys only ask about AI adoption in general, though the Dallas Federal Reserve also asks about generative AI use. The U.S. Census Bureau BTOS survey reports the lowest AI adoption rate, with fewer than 5 percent of firms using AI over the previous two weeks (first row of Table 1). The gap between BTOS and the other surveys is not due to sampling error: the Census Bureau estimates the standard error of their estimates at about 0.05 percentage points, and a binomial calculation implies standard errors for the other surveys of about 3 percentage points or less. The second row of Table 1 shows an alternative estimate of 20 percent AI uptake based on BTOS data; this is similar to some of the other firm level estimates. This estimate is employment-weighted instead of firm-weighted, meaning that large firms (which tend to have higher uptake) influence the estimate more. In addition, the 20 percent figure is based on a question that asks for AI use over the previous six months rather than the two-week lookback period used in the other question. Bonney et al. (2024) find evidence of "de-adoption" -- firms experimenting with AI but then abandoning it -- consistent with higher measured adoption when using longer lookback periods. Finally, the question lists many specific AI applications, rather than simply asking if AI was used. It is possible that listing specific applications causes respondents to recall their in-scope uses. Note that the Census Bureau's transparency and thoroughness in the released data and documentation makes analysis and evaluation much easier.
Based on the results in Bonney et al. (2024), most of the gap between the 5 percent and 20 percent figure is due to employment weighting with the changes in the survey question accounting for less of the gap. While the other firm surveys are not explicitly employment-weighted, it is likely they skew towards larger firms. For example, documentation from the Richmond Federal Reserve shows that establishments with fewer than 10 workers account for about 20 percent of their sample, while these establishments make up about 80 percent of the population. In contrast BTOS has an explicit objective to be representative by firm size, can likely reach a representative sample (due to the Census's Bureau's expertise and resources), and explicitly weights estimates to be representative by firm size.
To summarize, there are good reasons to favor the relatively low Census estimates of firm-weighted AI adoption, among them the availability of a representative sampling framework and a well-documented methodology. However, the uptake rates are not necessarily at odds with each other. If we tentatively treat the other firm-level surveys as being approximately employment-weighted -- due to practical sampling considerations and the cost of contacting very small firms -- then almost all the estimates of AI adoption fall between roughly 20 percent and 40 percent.
Turning to surveys of individual workers (rows 8 through 17), we also see many estimated adoption rates in the 20 to 40 percent range. Surveys of specific occupations show some variability, but surveys of computer programmers (Github and Jetbrains) show extremely high uptake rates. Interestingly, most worker-level surveys focus on generative AI, while firm-level surveys on broad AI adoption are more common. This may reflect the view that generative AI tools are fairly easy for a wide range of workers to adopt, whereas non-generative AI (the vast majority of applications prior to 2022) more often would require specialized training and a production process organized around the tool. If worker-level surveys asked about all AI use presumably we would see higher measured adoption rates.
We see a fairly wide range of adoption estimates, but to what extent are these surveys mutually consistent? It is hard to make comparisons between the firm-level surveys and the individual-level surveys. Depending on how AI users are distributed across firms, the average firm-level adoption rate might be higher or lower than the individual adoption rate. For example, if AI-using workers are distributed uniformly across firms, firm AI adoption will be higher than worker-level adoption. On the other hand, if AI-using workers are concentrated in particular firms, then firm-level uptake could be lower than worker-level uptake. However, one relationship that should hold is that the employment-weighted firm adoption rate should be greater than the individual adoption rate. This is because the employment-weighted firm adoption rate effectively treats everyone at a firm as an AI user if anyone at the firm is an AI user. Note that even this relationship breaks if individuals are using AI at work without the knowledge of their employer. Indeed, the Conference Board survey found that for 29 percent of respondents management was not aware of the worker's AI use. The upshot is that a wide range of firm-level and worker-level adoption rates are potentially consistent, especially with the reporting and monitoring difficulties of generative AI.
While most surveys lack longitudinal data, those that do field multiple waves help us understand the trajectory of adoption. On the firm side, the Chamber of Commerce recorded a 73 percent annualized growth rate between 2023 and 2024. The Census BTOS survey shows a 78.4 percent annualized growth rate. Lastly, the American Bar Association reported a 38 percent annualized growth rate. Among individual-level surveys, Pew is the only source showing changes over time, with an annualized growth rate of 145 percent. Overall, these findings suggest that regardless of measurement differences in the levels adoption is rising very rapidly both at the individual and firm-level. These high growth rates cannot be sustained and will have to tail off in coming years; whether the saturation point is close to complete adoption remains to be seen.
This note documents a number of interesting patterns in surveys of AI adoption. As the BTOS data show, firm-level adoption may be measured as quite low or fairly high depending on the details of the question asked and the weighting used. While the topline Census number of 5 percent seems much lower than other surveys, employment weighting and the phrasing of the question closes much of the gap. Treating the other firm-level surveys as approximately employment-weighted, estimates of AI adoption range from about 20 percent to 40 percent. Individual level surveys of AI use at work find a similar range adoptions rates. Similar levels of individual AI adoption and employment-weighted firm AI adoption can be mutually consistent if some workers are using AI at work without the knowledge of their managers. Regardless of measured AI adoption levels, all available surveys show adoption growing rapidly.
The more important economic questions going forward will hinge on how AI is used in the workplace and how much it is used within each firm. It will be increasingly important for surveys to gauge intensity and novelty of AI use: whether it is simply a better autocomplete or whether it is automating large ranges of worker tasks. Productivity and employment may be affected if AI adoption leads to reorganization of production processes or increased automation of research and development; better measurement of these margins would be of great value.
Abel, Jaison R., Richard Deitz, Natalia Emanuel, and Benjamin Hyman.(2024)."AI and the Labor Market: Will Firms Hire, Fire, or Retrain?," Federal Reserve Bank of New York Liberty Street Economics,September
Bick, Alexander, Adam Blandin, and David J. Deming. (2024). "The Rapid Adoption of Generative AI (PDF)," National Bureau of Economic Research, September.
Bonney, Kathryn, Cory Breaux, Catherine Buffington, Emin Dinlersoz, Lucia Foster, Nathan Goldschlag, John Haltiwanger, Zachary Kroff, and Keith Savage. (2024). "Tracking Firm Use of AI in Real Time: A Snapshot from the Business Trends and Outlook Survey (PDF)," U.S. Census Bureau, March.
Bright, Johnathan, Florence E. Enock, Saba Esnaashari, John Francis, Youmma Hashem, and Deborah Morgan.(2024). "Generative AI is already widespread in the Public Sector", January.
Cañas, Jesus, and Emily Kerr. (2024). "Texas Firms Using AI Report Little Impact on Employment." Dallas Fed Economics, June.
Conference Board (2024). "Majority of US Workers Are Already Using Generative AI Tools -- But Company Policies Trail Behind".
Corcoran, Emily Wavering and Sonya Ravindranath Waddell. (2024). "Automation and AI: What Does Adoption Look Like for Fifth District Businesses?" June.
Daigle.(2024). "Survey: The AI wave continues to grow on software development teams". September.
Digital Education Council. (2024). "Digital Education Council Global AI Student Survey 2024," August. https://www.digitaleducationcouncil.com/post/digital-education-council-global-ai-student-survey-2024
He, May. (2024). "State of Workers 2024 Report." https://pro.morningconsult.com/analyst-reports/state-of-workers-2024
Korinek, Anton.(2023). "Generative AI for economic research: Use cases and implications for economists." Journal of Economic Literature 61.4 (2023): 1281-1317.
Korst, Jeremy, Stefano Puntoni, and Mary Purk. (2024). Growing Up: Navigating Gen AI's Early Years (PDF). AI at Wharton, November.
McClain, Colleen. (2024). "Americans' use of ChatGPT is ticking up, but few trust its election information." Pew Research Center, March 26.
Noy, Shakked, and Whitney Zhang.(2023). "Experimental evidence on the productivity effects of generative artificial intelligence." Science 381.6654 (2023): 187-192.
Peng, Sida, et al.(2023). "The impact of ai on developer productivity: Evidence from github copilot." arXiv preprint arXiv:2302.06590 .
Perkowski, Patryk and Ales Marsal.(2024). "Generative AI at Work: Survey Evidence from Three Central Banks," https://ssrn.com/abstract=4957562 or http://dx.doi.org/10.2139/ssrn.4957562
Sergeyuk, Agnia, Yaroslav Golubev, Timofey Bryksin, and Iftekhar Ahmed. "Using AI-based coding assistants in practice: State of affairs, perceptions, and ways forward." Information and Software Technology 178 (2025): 107610.
Toner-Rodgers, Aidan.(2024). "Artificial intelligence, scientific discovery, and product innovation." arXiv preprint arXiv:2412.17866.
U.S. Chamber of Commerce.(2024). "The Impact of Technology on U.S. Small Business," September. https://www.uschamber.com/technology/artificial-intelligence/the-impact-of-technology-on-u-s-small-business