Editorial summary. This is our text summary of an article. Charts, figures, and the author’s full voice are at the original — read it there .
Licence. The original is licensed under CC BY 4.0. This page adapts it: the summary and analysis are ours, not the authors’.
Editorial verdict
Landmark study. The representative sample of 18,000 workers linked to administrative register data gives this research unusual methodological rigour; the 16-percentage-point gender gap in ChatGPT adoption and the finding that employer restrictions and the need for training — not beliefs — are the primary adoption barriers are robust findings that people-development practitioners cannot ignore.
Executive summary
This paper examines adoption patterns of ChatGPT among workers in occupations identified as exposed to generative AI, addressing who uses the tool, how workers perceive its impact on their jobs, and what prevents broader uptake. Using a representative survey of approximately 18,000 Danish workers across 11 occupations, conducted between November 2023 and January 2024, and linked to administrative register data on earnings, wealth, education, and labour market histories, the authors document that 41% of workers in exposed occupations had used ChatGPT for work within roughly one year of its launch. Key findings include a pronounced gender gap — women are 16 percentage points less likely to have used ChatGPT than men in the same occupation — and the observation that higher-earning, younger, and less experienced workers adopted the tool earliest, suggesting inequality-reinforcing dynamics. Workers estimate ChatGPT can halve working times in approximately one-third of their tasks but report that employer restrictions and the need for training are primary barriers to adoption. A randomised information experiment confirmed that shifting beliefs about productivity gains does not translate into changed behaviour, implying structural and organisational frictions rather than informational deficits are the binding constraint on adoption.
Key insights
- 1Women in the same occupation and workplace, performing the same job tasks, are 12–16 percentage points less likely to have used ChatGPT for work than men, a gap not explained by beliefs, task composition, or other observable characteristics.
- 2Employer restrictions and the perceived need for training — not fears of job loss or lack of awareness of productivity potential — are the primary barriers preventing workers from adopting ChatGPT, as confirmed by a randomised experiment showing information provision alone does not shift behaviour.
- 3Although younger and less experienced workers adopt ChatGPT faster (consistent with the technology substituting for expertise), workers who adopted it already earned slightly more before its arrival, indicating that ability-related advantages may be compounding rather than offsetting existing inequalities.
Practical takeaways
- Employer policies — whether restrictive or permissive — are the dominant factor shaping ChatGPT adoption among workers; organisational decisions about access, guidelines, and training carry more weight than individual attitudes or awareness.
- Women more often report adoption frictions related to training, while uncertainty about ChatGPT's benefits explains little of the gender gap (11.8 to 10.4 percentage points), pointing to the nature of training provision as a lever for more equitable uptake across the workforce.
References
- Eloundou et al. (2023).GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.
- Carvajal et al. (2024).Survey experiment on ChatGPT adoption among university students in Norway.
- Otis et al. (2024).High-performing entrepreneurs and GPT-powered AI assistants in Kenya.
- Eisfeldt et al. (2024).Stock price responses of firms exposed to ChatGPT.
- Agrawal et al. (2024).Model of organisation-wide adoption of AI across job tasks.
- Autor et al. (2024).Tasks, automation, and the rise in US wage inequality.
- Acemoglu and Restrepo (2024).Robots and jobs: Evidence from US labor markets.
- Acemoglu (2024).The simple macroeconomics of AI.
Source & Provenance
Not specified
Anders Humlum, Emilie Vestergaard
December 30, 2024
Research Study
Europe
Original source metadata is preserved. AI analysis is generated separately.
Like this? Get the Monday Decision Brief — free, every week.
No spam, unsubscribe anytime.
Rate this article
Want the full article? Read it at the original source — free, no paywall.
Read original article