Workers’ age and AI adoption

Numerous studies have analysed the effects of AI on productivity, growth and employment. Few of these focus on the factors that promote or hinder the uptake of AI. Drawing on the results of a European survey, this column analyses these factors, including the role of the age structure of the workforce. The authors find evidence of an inverted U-curve between workforce age and AI adoption, with adoption being weaker where the age distribution is tilted towards young or elderly workers. However, higher AI exposure in an industry moderates these demographic effects, perhaps due to a skill-levelling effect of the new technology.

There is rising consensus that AI is a new general-purpose technology (GPT) whose diffusion could boost lingering productivity growth in both advanced and emerging economies (among numerous papers, see Aghion et al. 2019, Brynjolfsson et al. 2019, Filippucci et al. 2024). A vast body of literature has been devoted to analysing the effects of AI on productivity and employment, with findings that are sometimes contrasted (Yuting Fan and Nguyen 2026, Filippucci et al. 2026, Baslandze et al. 2026, Aghion and Bunel, 2024). The adoption of AI is progressing rapidly within firms, more than for previous GPTs overall, but little research has been carried out yet into the factors that promote or hinder this adoption (Bick et al. 2026, Allen 2026, Cerutti et al. 2025, Lindenlaud et al. 2026).

AI adoption factors

In a recent paper (Cette et al. 2026), we provide an empirical investigation of the industry-specific drivers of AI adoption across a large set of European countries. We draw on successive waves of the Eurostat Survey on “ICT usage and e-commerce in enterprises”, which since 2021 includes information on the share of firms adopting a range of AI technologies by industry as well as the perceived obstacles to such adoption. We supplement this database with additional information concerning structural, policy, and other characteristics that may affect adoption. The resulting panel covers 15 EU countries and 24 industries, with adoption and perceived obstacles spanning the 2021-25 period. 

The Eurostat Survey reports adoption of seven different types of AI technologies by firms in each industry: autonomous robots, image recognition/processing, machine learning, natural language generation, process automation/decision support, speech recognition, and text mining. An eighth technology was added in the 2025 survey, termed “Generative AI”. We focus on three adoption outcomes: the share of companies using no AI, the first principal component (PC1) of AI usage shares and the share of companies using at least three AI technologies. PC1 captures both the fact of using AI and the extent of AI use. 

We find robust negative statistical associations of both adoption and the extent of AI use (at least three AI technologies) with mismatch obstacles (data availability, expertise, technological incompatibility, or AI’s perceived lack of usefulness for the firm), the significance of which has increased over time, as well as with costs and personal concerns (legal uncertainty, ethical and privacy concerns), the significance of which has decreased over time. Human capabilities (such as ICT skills and organisational capital) and innovation policies are positively related with AI adoption, while policies curbing either workforce flexibility or competition in industries that are key providers of intermediate inputs are inversely related with adoption.

The impact of workers’ age

We also provide the first systematic cross-country/cross-sectoral analysis of the link between AI adoption and the workforce age distribution. So far, only two other studies have linked adoption to demographics: André and Schief (2026) relate AI exposure (measured via the OECD Survey of Adult Skills) to demographic characteristics of the workforce; Bick et al. (2026) report individual adoption choices (collected via a specific multi-country worker-level survey) by age group.

Generally speaking, the link between age and use of new technologies is conceptually ambiguous and the empirical evidence is inconclusive. AI adoption is no exception: it is closely related to expected returns when replacing or complementing (and augmenting) tasks executed by workers (Lindenlaub et al., 2026) and, in turn, task bundles are often related to workers’ age in complex ways. Thus, the issue is eminently empirical.

We find that the share of prime-age workers (aged 25–49) impacts positively and significantly on adoption and negatively and significantly on non-adoption. Conversely, adoption declines when either young (15-24) or senior (50+) workers predominate – sketching an inverted-U curve. Additional regressions, which interact the age variable with a country-invariant measure of sectoral organisational capital intensity, suggest that the positive effect of the share of prime-age workers rises with this intensity. One interpretation is that the presence of skilled management capable of implementing the organisational changes needed to accommodate AI adoption is key for reaping the adoption advantage provided by a higher share of more experienced workers.

Given the wide heterogeneity in sectoral exposure to AI (Eloundou et al. 2024, André and Schief 2026), it is reasonable to assume that the age–adoption nexus is moderated by exposure. Industry workers’ age interacts with other factors to influence AI adoption. These factors include workers’ experience, workers’ skills, and the propensity for workers’ tasks to be replaced or complemented by AI systems. 

Eloundou et al. (2024) classify tasks in each occupation according to different levels of AI exposure, and then aggregate occupations in each industry to derive industry exposure levels. In our study, we use their exposure level corresponding to a 50% improvement in task execution through AI, but our results are robust to their other exposure measures. Exposure varies greatly between industries: the proportion of jobs exposed to AI ranges from less than 8% in the construction to more than 35% in IT and other information services. 

Interacting age variables with sectoral exposure reveals that the inverted U-curve flattens substantially as industry AI exposure rises. The implication is that in industries at the high end of the AI exposure distribution, the negative association between high youth shares and adoption essentially disappears. For senior workers, the flattening is also significant in the same direction, though the interaction for seniors is less robust across alternative exposure indicators – such as sectoral digital intensity as measured by Smiderle et al. (2026) or Calvino et al. (2018) – suggesting that the exposure-moderation effect is cleaner for youth than for seniors.

Figure 1 shows this flattening of the inverted U-curve graphically by plotting the total marginal effects as a function of the Eloundou indicator for non-adoption (panel A), PC1 adoption (panel B), and extensive adoption (panel C), respectively, separately for young and senior workers. For instance, in panel B, at the mean exposure level (the red point on the graph), a marginal increase of 1 percentage point in the share of young workers is associated with a small decrease (about 0.012) in AI adoption. However, in sectors with high potential for task improvement through AI, this negative effect fades and even turns positive, exceeding the effect of senior workers.

Figure 1 Marginal effects of age on AI adoption outcomes versus sectoral AI exposure

A) Non-adoption

B) PC1 adoption

C) Extensive adoption

Note: Impact on the adoption (or non-adoption) of AI of a one-percentage-point increase in the proportion of young people or older people in employment, depending on the level of a job’s exposure to AI (as measured by Eloundou).

A natural concern is whether the young-worker effect reflects genuine age effects or is instead a proxy for job precariousness, since sectors with high youth employment (accommodation, retail) tend to have both high turnover and relatively low AI exposure. However, excluding the two most obviously precarious sectors (retail and accommodation and food service) from the full sample leaves the main age results essentially unchanged. The precariousness channel therefore cannot be ruled out but is at best a partial explanation. 

So, there is evidence of an inverted U-curve between age and AI adoption: where the share of mature workers (aged 25-49) is higher, AI adoption also tends to be higher. This is consistent with the idea that both inexperienced and elderly workers may not be at ease with the use of AI. However, AI exposure lessens the negative association between young age and adoption likely due to the skill-levelling effect of the new technology. 

Policy-wise, combining this insight with the importance for of overcoming mismatch adoption obstacles – mainly related to digital skills and managerial abilities – suggests that targeting public support to young and senior workers in less AI-exposed industries may be especially fruitful for AI diffusion.

Source : VOXeu