Time Saved by AI Is Offset by New Tasks for 8.4% of Workers, Danish Study Finds

A newly released Danish study examining the 2023–2024 period finds that generative AI, including tools like ChatGPT, has thus far had little to no measurable impact on overall wages or employment, despite rapid adoption in many workplaces. The findings come from a large-scale, early empirical analysis conducted by economists aiming to understand AI’s real-world effects on workers across multiple occupations. The study centers on whether AI chatbots reshaped earnings, hours worked, and task structures, and it reveals a nuanced picture: adoption is widespread and fast, yet the aggregate economic effects on pay and employment remain modest at best during the observed window. This piece distills the study’s methods, its key results, the implications for workers and firms, and the important caveats that frame how we interpret these early signals.

Overview of the Danish AI study and its key findings

The Danish study investigates the labor market impact of generative AI models over a defined window spanning 2023 and 2024, with a deliberate focus on practical, real-world work environments rather than controlled laboratory settings. The researchers examine 11 occupations that are commonly viewed as potentially vulnerable to automation, including roles such as accountants, software developers, and customer support specialists. The data set comprises a substantial sample: data from roughly 25,000 workers and 7,000 workplaces spread across the Danish economy, providing a broad view of how AI adoption manifests across different firm sizes, sectors, and job families. The central aim is to determine whether AI chatbots and related tools have shifted earnings, influenced recorded hours, or altered the nature of tasks workers perform.

The core finding of the study is striking in its air of measured conservatism. Despite the rapid and widespread uptake of AI within most exposed occupations, the analysis concludes that AI chatbots have not produced a statistically significant impact on earnings or on the number of hours worked in any occupation during the study period. The researchers emphasize that the confidence intervals around the estimated effects exclude substantial average effects, effectively ruling out productivity gains or earnings boosts larger than a one-percent change in the average outcome. In practical terms, this suggests that, in the Danish labor market during 2023–2024, AI adoption did not translate into pronounced wage inflation or broad shifts in employment demand attributable to AI enhancements alone.

A notable feature of the study is its emphasis on the pace of adoption. The researchers underscore that the level of uptake of AI tools occurred very quickly: most workers in the exposed occupations had adopted the chatbots and related tools by the time of observation. Yet, when the researchers looked at economic outcomes—the key signs of value creation in the labor market—the signal remained weak. This juxtaposition between brisk usage and modest economic effects invites a deeper exploration of how AI is being integrated into work processes and what kinds of tasks are being salted into new workflows.

The study also addresses methodological considerations around observed effects. By focusing on a broad, population-based Danish sample and using robust statistical methods, the authors can place their conclusions within a defined confidence framework. The central takeaway is not that AI has no effect at all, but rather that the average effects on wages and hours over the observed period are small enough to be statistically indistinguishable from zero within the specified margins. This distinction is important: it signals that while AI is transforming certain micro-tasks and processes, the macroeconomic footprint in terms of pay and hours worked is, at least in the short run, limited in scope.

The scope and scope-limiting questions the study raises

In framing their conclusions, the researchers note that the data reflect an early phase of AI deployment. The Danish period of observation corresponds to a moment when many organizations were still integrating AI tools into daily routines, testing prompts, and refining workflows. It is possible that longer-run effects will emerge as AI becomes further embedded, as tasks are restructured, and as complementary technologies and organizational practices mature. The authors acknowledge that the Danish context provides a crucial but inherently limited snapshot, and that different labor markets—characterized by distinct wage structures, training ecosystems, and automation capabilities—could yield different outcomes.

The study’s geographic and sectoral focus is another important constraint. Denmark is a highly developed, highly educated economy with robust social and economic institutions. While these attributes contribute to the reliability of the findings within this context, they may dampen or exaggerate comparable effects in other settings, particularly in regions with different productivity baselines or labor market dynamics. In addition, the research centers on occupations flagged as potentially vulnerable to automation, but it does not parse every nuanced task within those jobs. The real-world mix of activities, including non-automatable cognitive tasks and highly collaborative roles, may dilute measurable AI-driven productivity gains in aggregate statistics.

The researchers also discuss the possibility that some benefits of AI are not captured by traditional metrics of earnings and hours. For instance, improvements in quality, timeliness, or customer satisfaction, or shifts in the complexity and creativity of tasks, might not immediately reflect in wage and hours data but could influence long-run career trajectories, firm performance, or the allocation of work. The study explicitly notes that the absence of strong wage or hours effects does not imply that AI adoption is inconsequential; rather, it underscores the multifaceted nature of how AI interacts with work tasks, processes, and organizational structures.

Adoption dynamics: how and why AI spread across workplaces

The rapid spread of AI tools within the study’s scope stands out as a defining feature. The analysis documents that employer encouragement and investment played a meaningful role in accelerating adoption. Firms that invested in AI capabilities, and that integrated these tools into their workflows, saw higher rates of tool use by employees across the studied occupations. In other words, organizational commitments to digital transformation and AI infrastructure provided the enabling environment for workers to engage with AI chatbots more readily than in firms without such investments.

From a practical standpoint, the adoption pattern suggests that signaling and support from leadership—such as providing access to AI-enabled platforms, offering training resources, and establishing guidelines for how AI should be used in daily tasks—are important drivers of initial uptake. The study’s authors emphasize that the observed adoption is not merely a function of individual curiosity; it reflects a broader organizational alignment around AI-enabled processes and the perceived value of these tools in day-to-day work.

Across the 11 occupations examined, the timing of adoption appeared contemporaneous with the introduction of AI capabilities into firms’ operating models. Workers who engaged with AI tools reported a mix of benefits, including time savings and improved workflow efficiency. However, the magnitude of these perceived benefits did not translate into clear, measurable increases in earnings or reductions in hours worked at the macro level. The divergence between perceived productivity enhancements and observed earnings or hours highlights a crucial area for future inquiry: whether firms leverage time savings to reallocate tasks, compensate with other performance metrics, or pursue broader efficiency gains that do not immediately show up in wage data.

Sectoral and occupational nuances in adoption

While adoption was broadly widespread, the study uncovers variations across occupations. Some fields displayed more pronounced use of AI tools, while others showed more cautious uptake. The differences reflected not only the technical feasibility of implementing AI within specific tasks but also the relative reliance on human judgment, complex problem-solving, and interpersonal interactions that resist full automation. For example, roles centered on routine data processing and coding-oriented activities might have a clearer path to AI-driven automation, whereas positions involving nuanced human interaction and strategic decision-making may experience more incremental benefits.

The occupation-level analysis helps explain why aggregate effects on earnings and hours are modest. If AI displaces or accelerates work in some tasks, but those gains are offset by the emergence of new tasks that require human oversight, prompt engineering, or quality control, then net effects in wages and hours could remain muted. The study’s task-tracking approach—which records not just time saved but the appearance of new work—offers insight into this balancing act between efficiency gains and new responsibilities that emerge with AI usage.

Time savings, productivity gains, and the translation to earnings

One of the study’s central facts concerns time savings. On average, workers who used AI tools reported a modest improvement: about 2.8 percent of work hours saved, equating to roughly an hour per week for the average user. This level of time savings represents a meaningful productivity improvement on an individual basis, yet at the macro level it did not accumulate into noticeable raises in earnings or reductions in reported hours across all workers in the exposed occupations. The interpretation is that time saved in some tasks did not necessarily translate into higher wages or fewer hours worked when viewed across the entire workforce sample.

The study also highlights that AI did not uniformly compress work into shorter hours for all users. While some individuals benefited from efficiency gains, others faced the emergence of new tasks or responsibilities that consumed time, offsetting the potential time savings from automation. In particular, the research identifies that a subset of workers—8.4 percent—found themselves assigned new or altered tasks as a consequence of AI adoption. This phenomenon occurred even among some workers who did not directly use the AI tools themselves, suggesting organizational and workflow changes that ripple through teams and across roles.

To illustrate the complexity of the productivity story, consider the kinds of new tasks that appeared in response to AI integration. Some workers began spending time auditing AI outputs to ensure quality and accuracy, while others focused on developing and refining prompts to elicit better responses from the AI systems. Others, such as educators or instructors, reported spending time detecting whether students used AI-generated content in assignments, an activity that did not directly reduce their workload but altered how they allocated attention to student performance and integrity.

The discrepancy with other evidence on AI-driven productivity

The Danish study’s findings contrast with some other research that reports higher productivity gains from generative AI. For instance, a randomized controlled trial conducted in a separate context indicated that AI could raise worker productivity by a more substantial margin—about 15 percent on average. This discrepancy invites an explanation rooted in the diversity of tasks across real-world jobs versus the particular tasks often highlighted in experimental settings. In controlled experiments, researchers may select tasks that are highly amenable to AI automation, which can amplify measured productivity gains. In everyday work environments, however, many tasks are not fully automatable, and workers rely on a blend of machine-assisted and human-driven activities. The different environments, selection of tasks, and the presence of learning curves in real-world implementations can all contribute to divergent findings between experimental and observational studies.

Humlum, one of the study’s lead authors, has suggested that this difference is not surprising. He posits that while controlled experiments may concentrate on tasks with strong AI fit, most actual jobs involve a broad spectrum of duties where AI cannot fully automate or replace human judgment. Additionally, organizations are still learning how to integrate AI into workflows; as they optimize usage, the observed productivity gains could evolve over time. The key takeaway is that the early Danish results do not prove AI cannot boost productivity, but they indicate that the early, broad-based integration yields modest average gains that are not automatically captured in earnings or total hours.

Economic impact: earnings, hours, and distributional effects

A crucial aspect of the study concerns the observed effects on earnings and hours. After examining the data across the 11 occupations and more than 25,000 workers, the researchers report that AI adoption did not generate measurable increases in average earnings or reductions in recorded hours, within the studied period. The results are framed with caution: the estimated effects are small and statistically indistinguishable from zero within the confidence bounds used in the analysis. This finding stands in contrast to the expectation that AI adoption might automatically translate into higher pay for workers who use AI tools effectively or into reduced workloads through automation.

The study also raises questions about the distribution of any observed gains. Even when some productivity improvements occurred, the benefits did not appear to accrue uniformly to workers. The evidence points to a broader distribution of gains, where only a small share of productivity improvements translated into higher earnings. For many workers, time saved did not translate into immediate wage increases, possibly due to how organizations structure pay, promotions, and performance evaluation in the presence of new AI-enabled workflows.

Where gains show up and where they don’t

Importantly, the study notes that the 3 to 7 percent range of productivity gains, when they do occur, translated into higher earnings only in a minority of cases. In other words, for most workers, any small reduction in time spent on tasks did not automatically result in more pay. This nuance points to a broader interpretation: AI-driven efficiency may contribute to firm-level productivity and profitability without creating a proportional, direct link to individual earnings in the near term. It also signals that wage dynamics are influenced by a constellation of factors beyond mere time savings, including negotiation dynamics, skill premiums, job tenure, and labor market demand for specific competencies.

Limitations, scope, and the broader context for AI’s labor-market role

The study’s authors are explicit about the limitations of their analysis and the caution needed when generalizing findings. First, the data reflect the early deployment phase of generative AI, a period during which firms were still learning to integrate AI into workflows. As AI technologies mature and as organizations develop more sophisticated methods for prompt design, process automation, and governance, the labor-market impact could evolve in ways not captured in this window. The observed effects may understate longer-term dynamics such as shifts in career paths, task diversification, and the creation of entirely new roles that leverage AI capabilities.

Second, the setting is Denmark, a country with a high level of human capital, robust labor-market institutions, and a well-developed digital infrastructure. These conditions may dampen or modify the magnitude of AI’s impact relative to countries with different educational profiles, wage structures, or industrial composition. Consequently, while the Danish experience offers valuable evidence, it does not automatically map onto every national context. Researchers and policymakers should be careful about overgeneralizing.

Third, the study concentrates on a specific subset of occupations identified as potentially vulnerable to automation. While this helps isolate the AI’s effects in high-exposure roles, it may miss broader, indirect effects that arise in other parts of the workforce or in cross-functional teams where AI tools influence collaborative work, brainstorming, and decision-making. The authors acknowledge that subsequent research should broaden the range of occupations and include diverse work environments to capture a more comprehensive view of AI’s labor-market implications.

Finally, the authors emphasize that AI’s long-term economic impact remains an open question. A single two-year window is not sufficient to capture all the potential lag effects or the cumulative impact of more integrated AI uses beyond chatbots. As AI models evolve and as organizations accumulate experience, the balance between time savings, task redefinition, and earnings effects could shift in ways that require ongoing analysis and updated assessments.

Implications for policy, business strategy, and worker preparation

Despite the modest short-run earnings and hours effects, the study offers several important implications for policy, business strategy, and how workers prepare for an AI-augmented economy. For policymakers, the findings suggest that broad expectations of immediate, large-scale wage gains from AI adoption should be tempered. Instead, policy discussions might focus on supporting retraining, facilitating task reallocation, and creating pathways for workers to move into roles that leverage AI-enhanced capabilities—while maintaining safeguards around wage progression and job security.

For businesses, the results highlight the importance of deliberate change-management practices when introducing AI tools. Firms may benefit from pairing technology deployment with careful workflow redesign, clear governance on AI use, and ongoing training that helps employees harness AI to complement, rather than merely replace, human labor. Because time savings exist but do not automatically translate into higher earnings, organizations should explore how to align compensation, promotions, and career development with performance measures that reflect AI-enabled collaboration, quality improvements, and the broader value added by human workers in AI-assisted work environments.

From a workforce-development perspective, the study underscores the value of upskilling in areas where AI complements human judgment and creativity. While mechanistic data work and routine processing may be affected by AI, roles that rely on nuanced analysis, interpretation, and interpersonal skills remain essential. Preparing workers to design, supervise, and refine AI-driven processes could yield more meaningful benefits over time, both in productivity and in earnings trajectories, particularly as AI systems become more capable and integrated into daily routines.

Practical guidance for organizations preparing for AI-augmented work

• Conduct occupation-specific assessments to identify where AI can most effectively support tasks without compromising quality.
• Invest in AI literacy and prompt-engineering training so employees can maximize tool effectiveness.
• Establish governance frameworks that address data privacy, ethical considerations, and output quality to maintain trust and reliability.
• Create incentive structures that recognize improvements in process efficiency, output quality, and collaboration enabled by AI, not just reductions in hours.
• Monitor long-term outcomes beyond immediate time savings, including job satisfaction, skill development, and career progression.

Future research directions and the evolving AI labor-market landscape

The Danish study represents an important stepping stone in understanding how generative AI interacts with real-world labor markets. It points to the need for ongoing, longitudinal research that can capture longer-term effects, lagged outcomes, and the potential emergence of new roles as AI capabilities mature. Future studies should consider expanding geographic scope to include diverse economic environments, widening the occupational scope to encompass a broader spectrum of tasks and industries, and employing richer measures that go beyond earnings and hours to include job satisfaction, skill accumulation, and organizational performance.

Researchers will likely explore several promising avenues. First, analyses that track workforce composition changes over longer horizons could reveal whether AI serves as a catalyst for job evolution—leading to new positions that combine domain expertise with AI oversight. Second, studies that examine the quality and impact of AI-generated outputs, including error rates, reliability, and the downstream effects on customers and clients, could provide a more nuanced view of AI’s value beyond time savings. Third, cross-country comparisons may illuminate how institutional features—training ecosystems, wage dynamics, and social safety nets—shape AI’s labor-market outcomes. Finally, as data availability improves, inquiries into heterogeneous effects—how AI affects workers with different skill levels, experience, or access to training—can help design targeted interventions that promote inclusive benefits from AI adoption.

The broader takeaway from this evolving landscape is that AI’s economic impact is likely to be gradual and context-dependent. Early results showing limited wage or employment effects do not foreclose meaningful long-run benefits, but they do encourage deliberate policy and corporate strategies that focus on thoughtful integration, skill development, and equitable distribution of gains. As AI technologies expand, their most significant effects may emerge through shifts in task design, new collaborations between humans and machines, and the creation of competencies that leverage AI to enhance human capabilities rather than simply substitute them.

Conclusion

The Danish study provides a rigorous, data-driven lens on the early interaction between generative AI tools and the labor market. It shows that adoption in multiple occupations occurred rapidly, yet the observed effects on earnings and hours were modest within the 2023–2024 window. It also reveals that AI can generate new tasks for a notable minority of workers, illustrating how efficiency gains can be offset by the emergence of additional responsibilities. Time savings exist, but the translation of those savings into higher pay is not straightforward, with only a small share of productivity gains appearing as earnings enhancements in the observed period.

These findings underscore that the economic story of generative AI is complex and evolving. They suggest a need for ongoing monitoring, more expansive research across geographies and occupations, and careful consideration of how to structure incentives, training, and task design to ensure that AI adoption yields tangible, equitable benefits for workers over time. As AI capabilities continue to mature and integrate into daily work, the lessons from this Danish snapshot point toward a broader understanding: technology can empower workers and improve processes, but the realization of sizable wage gains and employment shifts depends on deliberate, well-supported strategies at both organizational and policy levels, coupled with an openness to adapt as the AI-enabled workplace—and the skills that sustain it—continues to evolve.