A new study published in the Proceedings of the National Academy of Sciences has found that the adoption of ChatGPT in the workplace has been widespread but uneven. Younger, less experienced, and higher-earning workers are more likely to use the tool, while lower-income workers and women have adopted it at significantly lower rates. Even among workers in the same occupation handling similar job tasks, men are much more likely to use ChatGPT than women. The findings highlight how barriers to adoption may be reinforcing existing inequalities in the labor market.
ChatGPT is one of the most widely used generative artificial intelligence tools, capable of producing human-like text in response to user prompts. Since its launch, it has been integrated into various workplaces to assist with tasks such as writing, coding, and research. Given its potential to transform how work is performed, researchers were interested in understanding who is using ChatGPT, how they expect it to impact their jobs, and what factors influence adoption.
While generative AI has the potential to help workers by automating repetitive tasks and increasing efficiency, some studies suggest it could benefit less experienced workers the most by leveling skill gaps. However, adoption patterns are not always uniform across different demographics. By examining how workers across various occupations use ChatGPT, the researchers aimed to uncover whether certain groups were being left behind in the adoption process.
“I’m a labor economist studying how technological change affects employment and wages. ChatGPT marks the rise of generative AI, and we wanted to measure which workers have adopted this frontier tool to better understand its potential impact on the labor market,” explained Anders Humlum, an assistant professor of economics and Fujimori/Mou Faculty Scholar at the University of Chicago.
To investigate ChatGPT’s adoption in the workforce, the researchers collaborated with Statistics Denmark to survey 18,000 workers from 11 occupations between November 2023 and January 2024. The survey targeted jobs that experts considered highly exposed to ChatGPT, meaning that the tool could significantly reduce the time required to complete key job tasks. These occupations included marketing professionals, financial advisors, software developers, journalists, and legal professionals.
Survey responses were linked to national labor market records, which provided detailed information about each worker’s earnings, education, work history, and demographic background. The researchers analyzed which workers were using ChatGPT, their expectations about how it would impact their work, and the barriers preventing some employees from adopting the tool.
Humlum and his colleagues found that ChatGPT has been widely adopted in exposed occupations, with 41% of workers using it for job-related tasks. However, adoption rates varied significantly by profession. Marketing professionals and journalists—who often engage in writing-intensive tasks—were the most likely to use ChatGPT, with 65% reporting that they had used the tool. By contrast, financial advisors, whose work involves handling sensitive information, had the lowest adoption rate at just 12%.
When examining demographic differences, the researchers discovered a gender gap. Women were 16 percentage points less likely than men to use ChatGPT, even when comparing workers in the same occupation with similar job responsibilities. This gender disparity persisted across all measured indicators of adoption, suggesting that women face additional barriers to using the technology.
“The staggering gender gap in the adoption of ChatGPT — even among workers in the same occupations handling similar job tasks — was a big surprise for us,” Humlum told PsyPost.
Age and experience also played a role in adoption. Younger and less experienced workers were more likely to use ChatGPT, with each additional year of age associated with a 0.7 percentage point decrease in the likelihood of using the tool.
Interestingly, despite having less tenure in their fields, workers who used ChatGPT tended to earn slightly more before the tool became available. This suggests that higher-achieving individuals within their professional peer groups were more likely to experiment with and adopt the technology.
“Our key finding is that substantial inequalities have emerged in the adoption of ChatGPT,” Humlum said. “Younger, higher-earning men are significantly more likely to use these tools compared to other workers within the same occupation.”
Workers in occupations exposed to ChatGPT saw significant productivity potential in the tool, estimating that it could cut working times in about a third of their job tasks. However, employees with greater expertise in their fields tended to report smaller productivity gains, indicating that ChatGPT is more beneficial for those who have less specialized knowledge.
Despite recognizing the tool’s efficiency, many workers were hesitant to change how they allocated their time. About 40% of respondents stated that even if ChatGPT saved them time on certain tasks, they would not increase the amount of work they did in those areas.
One of the study’s most surprising findings was that workers’ expectations about time savings did not strongly predict whether they actually used ChatGPT. For example, among those who believed the tool could cut task completion time in half, only 23% planned to use it in the next two weeks.
Instead, adoption was often hindered by structural barriers, such as employer restrictions on AI use or the perceived need for training. When researchers randomly informed workers about ChatGPT’s time-saving potential, it did not significantly change their likelihood of using the tool, reinforcing the idea that external factors are limiting adoption.
The survey focused on Danish workers, so the findings may not fully reflect adoption patterns in other countries. Additionally, while the study identifies trends in adoption, it does not explore whether ChatGPT use leads to measurable improvements in job performance or career outcomes.
“The patterns we document reflect short-run effects, capturing the unequal adoption of ChatGPT in its first year,” Humlum noted. “These inequalities could narrow or widen over time as more workers adopt the tools and firms begin reorganizing workflows around this new technology.
Future research could investigate whether early adopters gain long-term advantages in earnings or promotions and whether targeted training programs can help bridge existing gaps in adoption. “Our long-term goal is to understand how generative AI is reshaping labor markets,” Humlum said.
The study, “The unequal adoption of ChatGPT exacerbates existing inequalities among workers,” was authored by Anders Humlum and Emilie Vestergaard.
