In Japan, as in other countries, the use of AI appears to improve job performance and various aspects of the working environment. A wide range of workers – including those in SMEs, persons with disabilities, individuals balancing work with caregiving responsibilities for children or long-term care, and those in non-urban areas – report positive effects of AI. However, Japanese AI users tend to perceive these improvements more modestly than in other countries. In addition, some groups of workers – such as older workers and non-regular employees – are less likely to benefit. Japan must strengthen its efforts to promote initiatives that are associated with better outcomes for workers, such as training and worker consultations regarding the use of AI, while enhancing the inclusiveness of these efforts. It is also important for Japan to promote greater diversity in career pathways, including in job-based specialist roles characterised by a high degree of autonomy and a greater likelihood of benefiting from AI.
Artificial Intelligence and the Labour Market in Japan
Abstract
This chapter examines how AI affects performance at work and job quality in Japan, highlighting distinctive features and challenges faced by current AI users through international comparisons and in-depth national analysis, and by focusing on the following key indicators: job performance, enjoyment of the job, mental health, physical health, and fairness in management. The key findings are:
The use of AI is associated with greater work engagement, opportunities to learn new things on the job, opportunities for personal growth through work, greater flexibility at work (e.g. teleworking), as well as more annual paid leave days. In contrast, it may contribute to a reduction in total monthly overtime hours.
In Japan, the proportion of AI users reporting improvements in these five indicators after using AI exceeds those reporting deterioration. However, the magnitude of this net positive effect is consistently smaller than in other countries, suggesting that the positive impact of AI on job performance and working conditions may be more limited in Japan. One possible explanation is that Japan may lag behind other countries in implementing initiatives that are associated with better outcomes for workers when AI is adopted – such as company-provided training or worker consultations regarding the use of AI technologies – a topic further examined in Chapter 4. In addition, certain cultural features of Japanese workplaces – such as the tendency to form hierarchical organisational structures and the prevalence of generalist workers without specialised expertise, may limit the benefits of AI use. Promoting awareness of the Job-Based Personnel Management Guidelines, which include practical examples to support companies in advancing job-based human resources, could help build an environment in which workers can more readily benefit from AI by enhancing the diversity of career pathways, including those leading to job-based specialist roles that are more likely to benefit from AI.
A wide range of workers appear to benefit from using AI in the workplace, including: AI users working in SMEs, those with disabilities, those balancing work with childcare and/or long-term care responsibilities, and those residing in non-urban areas. Econometric analysis further reveals that Managers and Professionals – who also tend to have higher rates of AI use – are 6 to 7 p.p. more likely than Clerical support workers to report that their job performance “Improved a lot” after using AI. However, some AI users may face barriers in accessing the AI benefits. Middle‑aged and older AI users, those in non-regular employment, and those working as clerical support workers are less likely to report improvements in job performance and working conditions. Age‑related disparities appear to be greater in Japan than in other countries. In the manufacturing sector, when averaging the differences across five indicators in the proportion of AI users reporting improvements, the gap between younger and older age groups is 20.1 p.p. in Japan, compared to an average of 6.7 p.p. in the other seven OECD countries surveyed. These findings suggest that Japan needs to do more to ensure the benefits of AI are more equally distributed.
This chapter also examines the impact of AI on wages from two different perspectives: employees’ perceptions of how AI will affect future wages, and changes in actual average gross wages before and after using AI. The key findings are:
In Japan, although workers are less pessimistic about the impact of AI on future wages in their current sectors than in other surveyed countries, they are still more likely to predict a wage decrease in the next ten years rather than an increase – similar to worker expectations in other countries. Even among AI users, there is no clear consensus that AI will lead to wage increases.
Despite such negative expectations, the proportion of AI users in Japan reporting an increase in actual wages after using AI exceeds the proportion reporting a decrease. While a wide range of workers appear to benefit from such wage increases, some workers – such as female workers, middle‑aged and older workers, and non-regular employees – appear to be left out. Econometric analysis indicates that male AI users are 2.4 p.p. more likely than female AI users to report that their actual wages “Increased a lot” after using AI, suggesting an AI gender wage gap.
This chapter also explores how AI affects work processes. The key findings are:
Japanese AI users are less likely to report changes in the pace at which they perform their tasks and the control they have over task sequences than their counterparts in other countries, and when they do report changes the perceived improvements are more limited. These findings may reflect Japan’s unique employment practices and associated work styles, which tend to foster hierarchical organisational structures and may reduce the benefits of using AI at work.
Japanese AI users say AI helps them make faster and better decisions, similar to what AI users in other countries say.
There are initiatives that could be taken that are associated with better outcomes for workers:
Providing company training and financial support to help employees work effectively with AI, as well as encouraging employees self-initiated learning. According to econometric analysis, Japanese AI users who received both types of training are 17.0 p.p. more likely to report that their job performance “Improved a little” and 19.7 p.p. more likely to report it “Improved a lot” after using AI, compared to those who received neither form of training.
Promoting worker consultation on the use of new technologies in the workplace. Japanese AI users who were consulted by their employers regarding the introduction of new technologies are 7.1 p.p. more likely to report that their job performance “Improved a little” and 17.8 p.p. more likely to report it “Improved a lot” after using AI, compared to those who were not consulted.
Establishing internal rules or guidelines to ensure employees use GEAI appropriately in their work and maintaining ongoing communication between workers and their employers following its implementation. Among AI users, those working for companies that have already established internal rules or guidelines reported an average improvement of 61.0% across the five indicators of job performance and working conditions after using AI, compared to 39.8% for those whose companies have not yet established such guidelines.
Through better communication and the use of third-party risk assessments, companies could gain trust from their employees that they would only use safe and trustworthy AI in the workplace. When AI users do not trust their company, the impact of AI on job performance and working conditions tends to be more limited. On average, 57.3% of AI users who lack such trust report no impact from AI across all five indicators mentioned above, significantly more than who do have such trust (30.9%). Similarly, AI users who lack trust are less likely to perceive improvements in their job performance and working conditions resulting from AI. These findings also indicate the importance of supporting companies in addressing a broader range of AI-related risks – such as transparency, explainability, and accountability – in order to achieve improvements in job performance and working conditions.
According to the OECD job quality framework, the quality of jobs can be assessed by looking at the level and distribution of earnings, employment security, and the quality of the work environment, which includes aspects such as work intensity, performing physically demanding tasks, and worker autonomy. The introduction of AI in the workplace will affect many of these outcomes.
2.1.1. Quality of the work environment
The automation of tedious and repetitive tasks by AI could improve job enjoyment and allow workers to focus on more complex and interesting tasks. AI could also improve physical safety by automating dangerous tasks and improving monitoring systems and safety procedure controls. However, there are concerns around increased work intensity and stress related to AI use in the workplace. Ultimately, the effect of AI on the work environment depends on how thoughtfully and strategically it is integrated into workplace practices. Companies that use AI to support and empower their employees are more likely to see positive outcomes, while those that deploy AI in ways that increase pressure and reduce autonomy are likely to encounter challenges.
Workers who use AI are more than four times as likely to report that AI has improved working conditions rather than worsened them, across all indicators considered such as job performance, enjoyment, physical health, mental health, and fairness in management. Most AI users reported that AI had assisted them with decision making, which workers appeared to appreciate (Lane, Williams and Broecke, 2023[1]). Similarly, managers using algorithmic management tools – technologies, including AI, to fully or partially automate tasks traditionally carried out by human managers – report improvements in the quality of their decision making and in their own job satisfaction. Beyond benefiting firms, improved decision making by managers may also benefit workers through greater consistency and objectivity of decisions (Milanez, Lemmens and Ruggiu, 2025[2]).
Similar findings about a positive impact of AI on working conditions emerged from more qualitative research, with case study interviewees often reporting fewer tedious and repetitive tasks as a result of AI adoption, allowing workers to spend time on more stimulating and interesting tasks, thus increasing job enjoyment. For example, chatbots can deal with a high percentage of basic and repetitive customer service queries, freeing up agents to deal with more complex questions (Moore, 2019[3]). Workers’ physical safety was improved following AI implementation as the automation of processes allowed dangerous machines to run within enclosures or behind barriers. Visual inspection tools used to ensure the quality of products reduce workers’ eye fatigue. Mental health improvements were noted as well. For example, in one company, an AI system monitored the stock of materials along an assembly line and automatically ordered replenishments when needed, relieving workers from this responsibility, and reducing stress. Another example is a predictive maintenance AI technology used in a drug manufacturing company to identify early signs of degradation in seals, which was welcomed for removing the pressure of making these decisions. The implementation of an AI-based tool in manufacturing helped solve maintenance issues along a production line, aiding in the discovery of root causes and suggesting solutions, thus allowing maintenance workers to solve problems quickly and reducing stress (Milanez, 2023[4]).
However, some reports indicated that AI could also lead to a deterioration in job quality. Case study evidence showed some instances of increased work intensity due to higher performance targets or complexity induced by AI. An AI developer acknowledged this issue and chose to retain 10% of easy tasks, which could have been automated, to provide workers with necessary mental breaks (Milanez, 2023[4]). Survey evidence shows that three‑quarters of AI users said that AI had increased the pace at which they perform their tasks. While this could indicate increased worker productivity, it might also reflect increased work intensity. In the finance sector 49% of workers and in manufacturing sector 39% reported that their company’s AI systems collected data on them as individuals or on their performance at work. Of these employees, 62% in the finance sector and 56% in manufacturing sector experienced increased pressure to perform due to this data collection (Lane, Williams and Broecke, 2023[1]). Furthermore, one survey of workers in Japan suggests that the reorganisation of tasks in the wake of AI adoption contributes both to greater job satisfaction and increased stress. The authors suggest that AI allows workers to concentrate on more complex tasks that can only be performed by humans. These more complex tasks may intensify work-related stress but may also provide a greater sense of satisfaction once accomplished (Yamamoto, 2019[5]).
Some of this deterioration in working conditions may be linked to the increased of use of algorithmic management, which consists of using algorithms to either support or automate management decisions – such as deciding workers’ schedule, assigning tasks to workers, monitoring, and supervising the quality of the work. Once an algorithm assigns tasks and sets deadlines for their completion, workers might have to accelerate their pace to keep up with the speed dictated by the algorithm. This pressure can lead to stress and anxiety and, in the most extreme case, burnout if the deadlines set by the machine cannot be achieved. In the case of the gig economy, the high performance targets imposed by these algorithms can also lead workers to take unnecessary physical risks such as jumping traffic lights or speeding (Todolí-Signes, 2021[6]). Algorithmic management can also reduce workers’ autonomy and sense of control over their work. For example, warehouse workers are often equipped with wearable devices that dictate what to collect, where to find it, and how long they have to retrieve each item, but also guide their movements through vibrations to be more efficient. If it takes longer than suggested, the worker receives a warning notice. Depriving workers of the ability to make even minor decisions, or how to move their own limbs, reduces autonomy and can remove an important sense of dignity and humanity from work (Moore, 2018[7]; Brione, 2020[8]).
A closely related concept is that of people analytics, also known as HR analytics, which is the use of individualised data related to employee performance, engagement, turnover, and other HR metrics to inform strategic decision making. It can be used to decide who to hire, but also who should receive a bonus, training or a promotion. Occupational safety and health risks of stress and anxiety arise if workers feel that decisions are being made based on numbers and data that they have neither access to nor control over (Moore, 2019[3]).
Ultimately, AI technologies can lead to different outcomes depending on the organisational context in which they are implemented, and the initiatives undertaken to maximise their positive impact on quality of the work environment. In hierarchical and bureaucratic organisations, AI adoption can create an “algorithmic cage” limiting worker autonomy and increasing resistance to its use. In contrast, organisations that allow room for professional judgement can foster an “algorithmic colleague”, where AI is used as a supportive tool to enhance human decision making (Meijer, Lorenz and Wessels, 2021[9]). To address employees’ concerns and facilitate effective AI integration, managers must provide resources and formal training before and during implementation (Makarius et al., 2020[10]). Building employees’ trust in AI, clarifying its role and impact on work, and promoting a clear understanding of the technology enhance business performance. This can be attributed to improved emotional states at work, and a consequent increase in employee commitment and job satisfaction (Chowdhury et al., 2022[11]). According to an OECD survey, AI users who have received company-provided training or consultation on introducing new technology in workplace are even more likely to report positive outcomes of AI on working conditions, such as job satisfaction, physical health, mental health, and fairness in management (Lane, Williams and Broecke, 2023[1]). By integrating AI in ways that complement rather than replace human expertise, organisations can reduce anxiety, build trust, and fully harness AI’s benefits while minimising potential downsides (Bankins et al., 2024[12]).
2.1.2. Impact on wages
Empirical evidence shows no significant relationship so far between AI exposure and wage growth at the aggregate level. However, the effects vary across occupations and income groups, with workers employed in occupations with either high software prevalence or high-income benefiting more.
The impact of AI on wages is theoretically ambiguous. Automation could reduce the need for workers in certain tasks, lowering demand for those occupations and putting downward pressure on wages. Conversely, increases in productivity, leading to cost reductions and lower prices, could boost demand for goods and services, increasing labour demand and result in higher wages. Additionally, workers with skills in AI development and maintenance would be in higher demand, which could lead to higher wages for this subset of workers.
Empirical evidence at the aggregate level shows little evidence of a relationship between wage growth and AI exposure (Acemoglu et al., 2022[13]; Albanesi et al., 2023[14]). Similarly, in case studies carried out by the OECD in the finance and manufacturing sector of 8 OECD countries 84% of interviewees reported that the wages of workers most affected by AI remained unchanged and 15% reported increases (Milanez, 2023[4]). One possible explanation for this lack of relationship is that AI advances may not lead to substantial increases in productivity. Acemoglu (2024[15]) predicts that total factor productivity (TFP) gains over the next decade will be very modest. Furthermore, he anticipates a widening gap between capital and labour income, suggesting that the benefits of any productivity increases are likely to accrue more to capital owners than to workers. This implies that while AI may enhance productivity, the economic gains may not translate into significant wage increases for the labour force. These findings highlight the importance of social dialogue and collective bargaining in achieving a fair share of any productivity gains that might accrue.
However, some studies do find a positive relationship between AI and wage growth when focussing on specific groups of workers. Felten, Raj and Seaman (2019[16]) find that AI exposure correlates positively with wage growth, and that this result is largely driven by occupations requiring a high level of familiarity with software. The same study finds that the relationship between AI and wage growth is positive and statistically significant for high-income occupations but not for low- or middle‑income occupations. Fossen and Sorgne (2022[17]) find a positive relation between AI exposure and wage growth, but which is statistically significant only for individuals who received at least a high school degree. They argue that education enhances the ability to learn new information allowing highly educated workers to adapt better to new technologies. These workers are also more likely to possess skills that cannot easily be replaced by digital technologies such as creative and social intelligence, reasoning, and critical thinking skills.
With regards to the AI workforce, (Alekseeva et al., 2021[18]) document a dramatic increase in the demand for AI skills over 2010‑2019 in the U.S. economy across most industries and occupations. This is associated with a wage premium of 11% for job postings that require AI skills within the same firm and of 5% within the same job title. However, according to Green and Lamby (2023[19]), on average across countries in the sample, wages grew cumulatively by 7% for the AI workforce compared to 9% overall over the period 2017 to 2019, suggesting that supply of AI workers may be keeping up with the demand for them.
According to an OECD survey, in the finance and manufacturing sector, many workers expected AI to put downward pressure on wages, with twice as many workers expecting AI to decrease wages in their sector in the next 10 years as to increase them. However, AI users (particularly male workers with a university degree) or workers who have been consulted on introducing new technology in the workplace are more likely to expect wages in their sector to increase due to AI in the next 10 years, although the expected proportion of wage increases remains lower than that of decreases. Moreover, AI users who have received company-provided training are more likely to expect AI to increase wages in the sector in the next 10 years, with the expected proportion of increases surpassing that of decreases (Lane, Williams and Broecke, 2023[1]).
2.1.3. Equalisation of performance within occupations
AI technologies demonstrate potential to enhance productivity in certain tasks, particularly for less experienced and lower-performing workers, by capturing and disseminating the effective practices or tacit knowledge of top performers, thereby reducing inequalities in job performance. Consequently, although there is little evidence so far, AI may contribute to reducing wage inequalities within occupations. On the other hand, for complex tasks, these tools can sometimes be counterproductive, particularly for less experienced workers who have difficulty identifying errors in AI-generated outputs or who rely on them without thorough review.
When a job’s core skill involves making predictions based on data patterns, and AI enhances the accuracy of these predictions, AI could help close the productivity gap between high- and low-skilled workers. In the context of taxi drivers, an AI system assisting drivers with finding customers by suggesting routes along which the demand is predicted to be high narrowed the productivity gap between high- and low-skilled drivers by 14% (Kanazawa et al., 2022[20]).
Several other experiments and empirical studies focus specifically on GEAI, suggesting that GEAI tools can improve the performance of the least experienced or lowest performing workers the most. Peng et al. (2023[21]) found that GitHub Copilot, an AI tool that assists developers by providing code suggestions, helped programmers finish their task 55.8% faster, and developers with less programming experience and older programmers benefited the most. Noy and Zhang (2023[22]) examined the productivity effects of ChatGPT usage by college educated professionals in writing tasks. The results show that ChatGPT increased productivity: the average time taken to complete tasks decreased by 40% and output quality rose by 18%. Inequality among workers decreased, participants who scored lower in an earlier task benefited more from ChatGPT access. (Brynjolfsson, Li and Raymond (2023[23]) study the staggered introduction of a generative AI-based conversational assistant. The tool monitors customer chats and offers real-time response suggestions to agents, enhancing their efficiency while allowing them to maintain full control over the conversation and discretion to accept or ignore the suggestions. The authors find that access to the tool increases productivity, as measured by issues resolved per hour, by 14% on average, including a 34% improvement for novice and low-skilled workers but with minimal impact on experienced and highly skilled workers. Dell’Acqua et al. (2023[24]) conducted a field experiment to examine the performance implications of AI on realistic, complex, and knowledge‑intensive tasks. The authors find that for realistic consulting tasks within the frontier of AI capabilities, consultants using AI were significantly more productive, in terms of quantity and quality of the output, as well as time used. All consultants benefited, but those below the average performance threshold benefited more.
Analysing data across 19 OECD countries, Georgieff (2024[25]) finds some evidence that AI may be associated with lower wage inequality within occupations. While these findings are very preliminary and need to be interpreted with caution, they are at least consistent with the idea that AI can reduce productivity differentials between workers.
On the other hand, some researchers warn that LLMs can generate incorrect yet plausible results that are difficult to detect, especially for less experienced workers. LLMs has the tendency to produce wrong or faulty outputs (e.g. “hallucinations”) that, in the absence of expert knowledge, are difficult to detect and costly to rectify and might ultimately undermine the productivity of some workers. In cases where the underlying task is more complex and the output requires accuracy to be ultimately useful (e.g. software development), relying on GEAI might prolong task completion and decrease workers’ output quality (Dell’Acqua et al., 2023[24]). Professionals who had a negative performance when using AI tended to blindly adopt its output and interrogate it less. Kabir et al. (2024[26]) analysed ChatGPT’s response to 517 programming questions finding that 52% of ChatGPT answers contain incorrect information, however programmers in the user study failed to identify incorrect answers 39% of the time due to the comprehensive and well-articulated insights in ChatGPT answers. Kreitmeir and Raschky (2024[27]) explored the short-term effects of GEAI, focussing on ChatGPT, using data from over 36 000 software developers in Europe, using Italy’s sudden ban on ChatGPT as a natural experiment. The analysis showed that the ban disrupted both the quantity and quality of output, particularly among less experienced developers who relied on ChatGPT for assistance with complex tasks.
Dell’Acqua et al. (2023[24]) suggest that the capabilities of AI create a “jagged technological frontier” where some tasks are easily done by AI, while others, though seemingly similar in difficulty level, are outside the current capability of AI. To reduce job performance inequalities within occupations, it will be crucial to develop strategies that leverage AI’s strengths in appropriate fields by cultivating expertise on using AI in the workplace and effectively utilising their knowledge.
2.2.1. Distinctive features of the impact of AI on performance at work and job quality in Japan
More than half of employees in Japan expect the use of AI in the workplaces to increase over the next ten years (Figure 1.3). As AI adoption in the Japanese workplace grows, it is likely that even employees who are not currently AI users will be affected by its impact on their work. Therefore, understanding how current AI users are being affected, as well as identifying their characteristics and issues, will provide valuable insights for Japan to prepare for the future of work.
In this context, this section begins by examining five key indicators on job performance and working conditions: job performance, enjoyment, mental health, physical health, and fairness in management. Japanese AI users in both the finance and manufacturing sectors are more likely to report that AI had “No effect” on all these outcomes than those in other surveyed countries. When changes are reported, the D.I. (Improved − Worsened) scores for Japanese AI users are positive as in the other surveyed countries, however the magnitude of the positive difference is more modest than that observed elsewhere (Figure 2.1). The slightly higher proportion of Japanese AI users in the manufacturing sector reporting that AI “Worsened” job quality contributes to this result. However, the main reason behind the difference between the Japanese results and those for other countries, is that few Japanese AI users report that their outcomes “Improved” following the use of AI. Furthermore, across all industries, the average share of AI users in Japan reporting improvements across the five indicators of job performance and working conditions is only 35.8%.
There are several possible interpretations of these findings.
The first possible interpretation is that there are cultural differences in how individuals respond to surveys. In question-based surveys, Japanese respondents tend to avoid expressing positive evaluations, which may have led to the improvement effect of AI being rated lower than it actually is. Previous research has shown that Japan scores lower than other countries on indicators measuring positive states such as job satisfaction, self-efficacy, work engagement,1 and happiness (Lincoln, 1989[28]; Scholz et al., 2002[29]; Shimazu et al., 2010[30]). It has been pointed out that these results may be influenced by the fact that, while actively expressing positive emotions and attitudes is considered desirable in the West, there is a tendency in Japan to prioritise group harmony, with modesty and understatement being instilled as virtues from an early age (Shimazu et al., 2010[30]; Iwata et al., 1995[31]; Shimazu, 2022[32]).
Percentage of AI users
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?”
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
The second possible interpretation is that differences in the organisational contexts in which AI technologies. Meijer, Lorenz and Wessels (2021[9]) are implemented between Japan and other countries may have influenced the results. pointed out that, in hierarchical and bureaucratic organisations, AI adoption can create an “algorithmic cage” limiting worker autonomy. In Japan, unlike in Western countries, the mainstream employment management system is characterised by the concept of hiring workers for the company itself, rather than for specific positions with defined job descriptions, and is based on practices such as mass recruitment of new graduates, long-term employment, seniority-based promotions and wages, and job rotation (Hamaguchi, 2009[33]). In many cases, employees in Japan are regularly assigned to new positions under directives from the HR department, gaining expertise from their supervisors and senior colleagues while experiencing a work style that emphasises teamwork and a meticulous approval process involving supervisors and various stakeholders. This unique work culture in Japan may contribute to the establishment of hierarchical organisational structures. As a result, the approval process often requires various considering opinions from supervisors and stakeholders. This may make it difficult for Japanese AI users to perceive improvements in their ability to control their work pace and tasks. Furthermore, in a hierarchical organisational structure, most Japanese employees are developed as generalists with broad knowledge across various fields rather than as specialists with expertise in specific areas. As a result, employees who lack sufficient expertise in AI may be more likely to perceive the introduction of AI in the workplace as something imposed upon them. These Japanese employees may tend to underestimate the value of AI systems, feel a loss of autonomy, or experience increased work intensity. These structural or cultural characteristics may have contributed to the lower reported positive effects of AI in Japan.
The third possible interpretation is that Japan’s lag in implementing various initiatives that are associated with better worker outcomes when adoption AI (e.g. company training or financial support for training to work with AI, worker consultation on the use of AI technologies in the workplace, and building employee trust that their company will introduce only safe and trustworthy AI technologies in the workplace) may have influenced the results by limiting the positive effects of AI.
The fourth possible interpretation is that the time gap between the previous OECD survey and the JILPT survey may have impacted the results due to differences in the AI technologies examined. Specifically, the JILPT survey was conducted after the global spread of GEAI, while the previous OECD survey was conducted before. Therefore, the JILPT survey includes responses from GEAI users, which could be less positive.2 However, Japanese GEAI users are more likely to report that AI improves their job performance and working conditions compared to AI users excluding GEAI3 (Annex Figure 2.A.1), so this cannot explain the results.
All of these factors are likely to contribute to a greater or lesser extent to the lower perceived benefits of AI in the workplace in Japan, but it is difficult to control for them. This chapter aims to identify which groups of Japanese AI users are less likely to benefit from the positive effects of AI, while also highlighting underlying issues in Japan’s work styles and organisational structures (explanation 2 above) by examining the impact of AI on employee discretion and changes in work pace. The possibility that Japan is lagging behind in the implementation of various initiatives associated with better worker outcomes when adopting AI (explanation 4 above) will be examined in greater detail in Chapter 4.
Gender and the impact of AI on performance at work and job quality
The 2022 OECD surveys revealed that male AI users were more likely than female AI users to report that AI improved their job performance and working conditions. However, this trend is less apparent in Japan (Figure 2.2). In the finance and insurance sector, male AI users tend to provide more positive evaluations than female AI users on several indicators. In the manufacturing sector, however, female AI users tend to be more positive than male AI users. Furthermore, when comparing the average proportion of AI users reporting improvements across the five indicators at all-industry level in Japan, no significant difference is observed between male and female users (male AI users: 43.0%, female AI users: 43.1%). A generalised ordered logit model, controlling for several individual attributes, also indicates that there are no statistically significant gender differences in the reported effects of AI use on job performance in Japan (Annex Table 2.A.1).
Percentage of AI users, by gender
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Education and the impact of AI on performance at work and job quality
The 2022 OECD surveys revealed that AI users with a university degree were more likely to report that AI improved their job performance and working conditions than those without a university degree. However, this trend is not as apparent in Japan as in the other countries and there are differences by sector (Figure 2.3). In the finance and insurance sector, AI users with a university degree tend to provide more positive evaluations than those without a university degree, while, in the manufacturing sector, AI users without a university degree tend to report more positive evaluations than those with a university degree. That said, when comparing the average proportion of AI users reporting improvements across the five indicators at the all-industry level in Japan, no significant difference is observed between those with and without a university degree (AI users with a university degree: 43.2%, those without a university degree: 42.6%). However, a generalised ordered logit model controlling for individual attributes shows that, compared to Japanese AI users without a university degree, those with a degree are 3.0 p.p. less likely to report that AI “Worsened it a little” and 6.3 p.p. more likely to say it “Improved it a little” in terms of job performance after using AI (Annex Table 2.A.1).4
Percentage of AI users, by education
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI. In Japan, the figure for “University degree” is the sum of four‑year university and graduate school.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Age and the impact of AI on performance at work and job quality
The 2022 OECD survey revealed that younger AI users were more likely than older AI users to report improvements in their job performance and working conditions after using AI. A similar trend is observed in Japan. However, the proportion of middle‑aged and older AI users reporting improvements in job performance and working conditions after using AI is noticeably lower in Japan than in other countries. Japan therefore exhibits a larger age gap in the evaluation of improvements in job quality due to AI (Figure 2.4). Specifically, in the manufacturing sector, the average difference across the five indicators.
Percentage of AI users, by age
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
between the proportion of AI users aged 16‑34/15‑34 who report improvements and the proportion of AI users aged 50 and over who do so, is 6.7 p.p. in the other seven OECD countries, compared to Japan 20.1 p.p. in Japan. When comparing the average proportion of AI users reporting improvements across the five indicators at the all-industry level in Japan, a substantial difference is observed across age groups (aged 15‑34: 50.8%, 35‑49: 41.4%, 50 and over: 33.9%). Furthermore, according to the results of a generalised ordered logit model controlling for several individual attributes, when using AI users in Japan aged 15‑34 as the reference group, AI users aged 35‑49 and those aged 50 and over in Japan are less likely to report that AI “Worsened it a lot”, “Worsened it a little”, or “Improved it a little”, and are instead more likely to report “No effect” (Annex Table 2.A.1).5
Employment status and the impact of AI on performance at work and job quality
Japanese AI users in non-regular employment are less likely to report that AI improves their job performance and working conditions compared to those in regular employment (.
Figure 2.5). When comparing the average proportion of AI users reporting improvements across the five indicators at the all-industry level in Japan, non-regular workers report a lower rate (37.6%) than regular workers (44.0%). Furthermore, according to the results of a generalised ordered logit model controlling for several individual attributes, non-regular AI users in Japan are 5.5 p.p. less likely than regular employees to report that AI “Improved it a lot” and 7.0 p.p. more likely to report “No effect” in relation to their job performance after using AI (Annex Table 2.A.1).6
Percentage of AI users, by employment status
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Company size and the impact of AI on performance at work and job quality
Japanese AI users in SMEs are more likely to report that AI improves their job performance and working conditions than those in large companies (Figure 2.6). When comparing the average proportion of AI users reporting improvements across the five indicators at the all-industry level in Japan, those working in SMEs report a higher rate (47.4%) than those working in companies with 10 000 or more employees (35.1%). Furthermore, according to the results of a generalised ordered logit model that controls for several individual attributes, AI users working in SMEs are less likely to report that AI had “No effect” on their job performance after using AI and more likely to report it “Improved it a lot”, compared to those working in company with 10 000 or more employees (Annex Table 2.A.2). In addition, an OECD survey has revealed that GEAI helps to improve employee performance and compensate for worker shortages and the skill gap among SMEs in Japan and in other countries (Box 2.1). That said, even workers in large companies are positive about the impact of AI on job performance and working conditions.
Percentage of AI users, by company size
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who said that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
There are a number of possible explanations for the positive evaluations observed among SMEs.
First, SMEs in Japan tend to experience greater labour shortages and face more difficulties in recruiting workers than large companies. As a result, employees in SMEs often find it difficult to specialise in certain tasks within the company and are required to handle a broader range of multiple tasks (Inoue, 2019[34]). In practice, some employees in SMEs have started utilising RPA (Robotic Process Automation) and cloud-based tools to efficiently handle multiple tasks (Sakamoto,T. & Takayama,J., 2023[35]). Since AI can assist SME employees in performing multiple tasks more efficiently, it may have been perceived as an especially valuable tool for enhancing job quality.
At the end of 2024, the OECD conducted a survey on generative AI targeting 5 232 SMEs – i.e. enterprises with fewer than 250 employees – across seven countries (Austria, Canada, Germany, Ireland, Japan, Korea, and the United Kingdom). The individual within each company who had the most comprehensive understanding of the technologies used in the organisation was selected as the respondent. The results suggest that GEAI can support SMEs in a number of ways (OECD, 2025[36]).
The main way GEAI helps SMEs is by enhancing employee performance, more so than by enabling them to scale up, to compete with larger companies or to increase revenue, including in Japan. Moreover, a higher proportion of Japanese SMEs report that GEAI has helped in performing new tasks and offering new products or services, compared to those in other countries (Figure 2.7).
GEAI also helps compensate for labour and skill shortages in SMEs. Among SMEs using GEAI and experiencing a skill gap in the last two years, 39.1% say that GEAI helped compensate for the gap. Similarly, 25.2% of SMEs facing a worker shortage in the last two years say GEAI helped them compensate for this worker shortage. SMEs in Japan are more likely than those in other countries to report that generative AI has helped address skill gaps and labour shortages. This may reflect the fact that Japan reports the highest incidence of both labour and skill shortages among the countries surveyed, suggesting that GEAI is being used as one of the tools to help tackle these challenges (Figure 2.8).
Percentage of SMEs that report generative AI helped for each benefit, by country
Note: SMEs using generative AI were asked: “Thinking about the use of generative AI within your company, has it helped your company… increase revenue / perform tasks that could not be performed before / offer new products and services / save money / improve employee performance / compete with larger companies?”
Source: OECD (2025[36]), Generative AI and the SME Workforce: New Survey Evidence, https://doi.org/10.1787/2d08b99d-en.
Percentage reporting that generative AI helped compensate for worker shortage/skill gap, among SMEs using generative AI and experiencing shortage/gap
Note: SMEs using generative AI were asked: “Did generative AI help your company compensate for this worker shortage (the lack of skills or experience)?”
Source: OECD (2025[36]), Generative AI and the SME Workforce: New Survey Evidence, https://doi.org/10.1787/2d08b99d-en.
Second, SMEs are less likely to establish a highly hierarchical decision making structure due to their limited number of employees. As a result, individual employees tend to take on relatively greater responsibilities (Inoue, 2019[34]). Additionally, decision making in SMEs is often heavily influenced by the president, who is also the owner of the company (Yamaguchi, 2012[37]; Inoue, 2019[34]). A relatively high proportion of employees in SMEs may have significant autonomy in their roles as a result of their broader responsibilities. This high degree of autonomy may lead employees to perceive AI assistance as more directly contributing to improvements in their job performance and working conditions.
On the other hand, the more positive perception of AI impacts reported by workers in SMEs are unlikely to be because SMEs are better at harnessing AI than larger companies. SMEs may have weaker initiatives to maximise the benefits of AI in improving job performance and working conditions. While SMEs are highly diverse, decision making in SMEs is often heavily influenced by the president, who is also the owner of the company. Additionally, in many cases, the president takes on a hands-on role on the front lines of daily operations (Kawakami, 2013[38]). As a result, the adoption of new technologies in the workplace is often determined at the sole discretion of the president, with little to no consultation with employees (Iwamoto, 2021[39]). Moreover, such a decision making process may contribute to employee concerns about whether the technology is safe and reliable. Additionally, from a financial perspective, SMEs generally lack the resources to invest in employee training to the same extent as larger companies.
It may instead be the case that, particularly in large companies with a greater number of employees, there are underlying factors that lead to more restrained evaluations of AI. Large companies, due to the involvement of multiple stakeholders, may adopt a more cautious approach toward issues such as data breaches and security risks associated with AI. As a result, some large companies may implement AI in a limited and more experimental manner – for example, by conducting pilot tests or restricting implementation to specific departments or tasks. Some AI users working in large companies may have evaluated AI based on its use at this preliminary stage, which may explain the more restrained assessments observed. Therefore, it would be prudent to interpret the finding that Japanese SMEs evaluate AI more positively than large companies with caution at this stage.
Disability and caregiving and the impact of AI on performance at work and job quality
Japanese AI users with disabilities and those engaging in childcare and/or long-term care are more likely to report that AI improves their job performance and working conditions (Figure 2.9) According to the results of a generalised ordered logit model that controls for several individual attributes, workers with disability are 18.2 p.p. less likely to report that AI had “No effect” on their job performance and 12.9 p.p. more likely to report that it “Improved it a lot” (Annex Table 2.A.3). Similarly, workers with caregiving responsibilities, such as childcare or long-term care, are less likely to report that AI had “No effect” on their job performance and more likely to report that it “Improved it a lot” (Annex Table 2.A.4) Workers with disabilities or caregiving responsibilities are assumed to have diverse working styles; however, due to the nature of their disabilities or the need to balance work and care, many of them are likely to face difficulties in working long hours. Therefore, they may be more conscious of the need to improve work efficiency, and it is possible that they have successfully adjusted their tasks in response to AI use, leading to improvements in job performance and working conditions.
Percentage of AI users, by disability status, childcare status, long-term care status
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI. Response by “I don’t know” was combined with figures for those who are not disabled or not involved in care as described.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Occupation and the impact of AI on performance at work and job quality
When looking across all five job quality indicators on average, the difference in the proportion of AI users reporting improvements between “Craft and related trades” (highest proportion reporting improvements) and “Clerical support workers” (lowest proportion) is 13.4 p.p. Japanese AI users in “Managers” and “Professionals” are slightly more likely to report improved job performance, while those in “Craft and related trades”, “Plant and machine operation, and assemblers”, and “Elementary occupations” are less likely to do so. On the other hand, those in “Craft and related trades” and “Plant and machine operation, and assemblers” are somewhat more likely to report improved physical health, while those in “Clerical support workers” are less likely to do so. AI users in “Clerical support workers” are as likely as those in other occupations to report improvements in job performance, but less likely to report improvements in working conditions (Figure 2.10). According to the results of a generalised ordered logit model that controls for several individual attributes, when using “Clerical support workers” as the reference group, “Managers” and “Professionals” are 6‑7 p.p. more likely to report that AI “Improved [their job performance] a lot”. However, groups such as “Elementary occupations” and “Craft and related trades workers”7 may be more likely to experience a decline in job quality after the use of AI (Annex Table 2.A.5).
Percentage of AI users, by occupation
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI. The values in the above figure represent the average of the five markers for each occupation shown in the figure below.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
The impact of AI on performance at work and job quality across regions
AI users in both rural and urban areas report that AI improves their job performance and working conditions. As highlighted in Chapter 1, there are considerable regional disparities in the proportion of Japanese employees using AI at work, with the South Kanto area – including Tokyo and Kanagawa – being the region with the highest use (Figure 1.8). However, in terms of AI’s impact on job quality, the South Kanto area does not stand out with particularly high ratings. The greatest improvements in job quality are reported by AI users living in Hokkaido/Tohoku, while the least improvement is found in North Kanto/Koshin – the difference being 14.9 p.p. between them. When looking at the outcomes individually, AI users in Hokkaido/Tohoku are more likely to report improvements in working conditions, while those in Kyushu/Okinawa are more likely to report improvements in job performance. By contrast, AI users in Kinki report a similar level of improvement in job performance to other regions, but a lower level of improvement in working conditions. Furthermore, AI users in North Kanto/Koshin are less likely to report improvements in both job performance and working conditions compared to those in other regions (Figure 2.11).
Percentage of AI users, by residential area
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI. The 47 prefectures of Japan are divided into 10 blocks based on Japan’s regional classification in the OECD regional database. The values in the above figure represent the average of the five markers for each occupation shown in the figure below.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
AI’s diverse contributions to better job quality
Analysis of the D.I. (Increased – Decreased) scores regarding changes before and after the use of AI for each indicator measuring job quality suggests that Japanese AI users report positive scores for work engagement, opportunities to learn new things on the job, opportunities for personal growth through work, work style flexibility (e.g. teleworking), and the number of annual paid leave days taken. In contrast, the score is negative for total monthly overtime hours (Figure 2.12). While the results suggest that work engagement among Japanese AI users may have improved after using AI, appropriate employment management is needed to ensure that the use of AI in the workplace does not lead to workaholism.
Higher work engagement has been shown to positively impact job performance, increase employee initiative, improve health, and reduce turnover (Bakker,A.B. & Demerouti,E., 2008[40]; Shimazu, 2022[32]; MHLW, 2019[41]). A work environment in which opportunities to learn new things through work are frequently provided may, in some cases, reflect an increase in job demands, such as a heavier workload or more complex tasks. If these increased job demands following the introduction of new technology such as AI are perceived as positive and challenging stressors (e.g. opportunities for personal growth through work), they are expected to enhance work engagement. In addition, sufficient holidays and breaks at work enhance work engagement (Shimazu et al., 2012[42]; MHLW, 2019[41]). Work style flexibility (e.g. teleworking) reduces the mental and physical burden associated with commuting. According to a survey conducted by the Japan Productivity Center, the proportion of Japanese employees (including managers) who reported that their physical or mental health improved after implementing telework exceeded those who reported it worsened by around 26.6 to 32.8 p.p.8 (Japan Productivity Center, 2023[43]). AI technology may automate some managerial tasks, reducing the need for managers to be physically present in the workplace and expanding flexible work options that better align with their lifestyles. Additionally, telework may improve work efficiency by minimising unnecessary meetings and allowing employees to focus more effectively on core tasks.
On the other hand, a positive correlation has been observed between work engagement and workaholism, suggesting that under certain conditions, individuals who are highly engaged in their work may be more prone to become workaholics. Furthermore, workaholism has been found to have a positive correlation with psychological and physical stress, while it has a negative correlation with job performance (Shimazu,A. & Schaufeli,W.B., 2009[44]; MHLW, 2019[41]). If the opportunities to learn new skills or grow personally through work become too overwhelming for employees following the implementation of new technology, they may be at a higher risk of developing workaholism. This, in turn, can lead to fewer paid leave days taken and increased overtime, potentially resulting in heightened mental stress and physical fatigue. Moreover, focussing on telework as an aspect of increased work flexibility, the proportion of Japanese employees (including managers) who reported that their working hours increased after implementing telework exceeded those who reported it decreased by around 8.3 to 15.7 p.p. (Japan Productivity Center, 2023[43]). Telework can eliminate the boundary between work and personal life. As a result, some employees may find it harder to disconnect from work, potentially leading to prolonged working hours. Additionally, telework allows employees to respond to sudden work requests even on their days off, raising concerns that it may further extend total working hours.
Percentage of AI users
Note: AI users were asked: “How has your own perception or evaluation of your work changed before and after the use of AI?” Work engagement is created using the results of responses related to enthusiasm for work, pride in work, vitality through work, and immersion in work. The number of AI user respondents is 1 854. The figure shows the proportion of AI users who said that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Nature of tasks and the impact of AI on performance at work and job quality
Figure 2.13 B to G shows the proportion of AI users who report improved job performance and working conditions, split by those who frequently do a certain task (i.e. allocate at least half their time to it during a typical working day) and those who do not engage in the that task at all during a typical working day. In addition, to facilitate discussion, Figure 2.13 A shows the average increase in the proportion of AI users reporting improvements across all five indicators at once. These findings suggest that the nature of the tasks Japanese AI users primarily perform may contribute to differences in their perceptions of AI’s impact on job performance and working conditions (Figure 2.13). AI users who allocate more time to management-related tasks during a typical working day tend to evaluate improvements in job quality more positively than those who primarily engage in other types of tasks. Similarly, AI users who spend more time on tasks in hazardous places also tend to evaluate improvements in job quality more positively, and in particular enhancements in working conditions and occupational safety.
Percentage of AI users, by tasks in typical working day
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” All employees were asked: “How much time do you spend on the following tasks in a typical working day? (Almost always during working hours; Over half of working hours; Almost half of working hours; Under half of working hours; Nothing at all)” The figure is based on employees who answered spending “over half” or “almost all” of their working hours for each task in a typical workday and shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI. “Physical tasks” include activities such as standing work, operating machinery or vehicles, and carrying loads. “Tasks in hazardous place” include activities at high places, in extremely hot or cold places, in places with a lot of machinery. The figure for A represents the average difference in the improvement effects on five job quality indicators, comparing AI users who don’t engage in a specific task with those who perform that task frequently.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Changes in tasks and the impact of AI on performance at work and job quality
Although a causal relationship cannot be clearly established, differences in how employees’ tasks change after using AI may be related to variations in their perceptions of improvements in job performance and working conditions (Figure 2.14). For instance, 5.4% of AI users in Japan report spending less time on routine, repetitive, and physical tasks and more time on complex tasks, and these users tend to be more likely to report improvements in job performance and working conditions. Similarly, 14.4% of AI users in Japan report spending less time on tasks other than routine and repetitive ones, and more time on routine and repetitive tasks. These users tend to be more likely to report improvements in job performance or working conditions. This may be the result of AI standardising tasks that previously required human judgement, transforming them into simpler procedural tasks, or taking over certain tasks performed in hazardous places or physical tasks. For example, in Japan’s retail sector, an increasing number of companies are implementing AI-assisted systems for shift scheduling. Previously, staff responsible for scheduling manually created shift schedules based on paper records, considering individual shift preferences and ensuring that staff with the necessary skills were assigned during peak hours. In some cases, this process required several hours of work each month for staff in charge. With the introduction of AI-assisted systems, employees can submit their shift preferences via smartphone, while AI quickly drafts optimal shift schedules by analysing past data, predicting peak hours, and considering employees’ skills. In this way, what was once a task requiring human judgement and tacit knowledge has been standardised and transformed into a routine task with AI support, enhancing job performance and reducing employees’ workload.
Percentage of AI users, by changes in time spent on tasks during a typical workday after using AI
Note: AI users were asked: “How has your time spent on tasks changed before and after the use of AI? Routine and repetitive tasks/Physical tasks/Tasks in hazardous place/Identifying problems and solving them using creativity/ Managing and motivating team members or subordinates/Analysing data and information to make decisions based on the results (Increased it a lot; Increased it a little; Decreased it a little; Decreased it a lot; No effect; I don't know)” “Physical tasks” include activities such as standing work, operating machinery or vehicles, and carrying loads. “Tasks in hazardous place” include activities at high places, in extremely hot or cold places, in places with a lot of machinery. AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
2.2.2. How the impact of AI on wages varies by worker characteristics
Expected impact of AI use on future wages
Although 40% of employees in Japan reported “Don’t know” when asked about the impact of AI on wages over the next ten years, the proportion expecting a “Decrease” exceeds the proportion expecting an “Increase”, similar to trends observed in other countries (Figure 2.15). These pessimistic predictions regarding the impact of AI on wages may stem from concerns that AI could alter employees’ tasks and skill requirements or even lead to job losses in the future. These findings highlight the need for Japan, like other countries, to closely monitor the impact of AI on wages in the future. That said, among AI users, a greater proportion tend to expect a wage increase over the next ten years, both in Japan and in other OECD countries. When examining this share by country, Japan ranks in the mid-range alongside the United Kingdom in the finance and insurance sector, and in the manufacturing sector, Japan records the second-highest share, following Ireland (Annex Figure 2.A.2).
Percentage of all employees
Note: All employees were asked: “Do you think that AI will have an impact on wages in your sector in the next 10 years? (Yes, AI will increase wages; Yes, AI will decrease wages; No, AI will not impact wages; Don't know)”
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Types of AI technologies and their impact on actual wages
Among Japanese AI users, the D.I. (Increase – Decrease) score for changes in gross wages (before deducting taxes and social security contributions) before and after AI use is 7.3 p.p., suggesting that wage increases are more prevalent than wage decreases. When analysed by type of technology, among Japanese GEAI users, the D.I. (Increase – Decrease) score for changes in wages before and after AI use is 10.6 p.p., whereas for non-GEAI users, it is −2.9 p.p. (Figure 2.16). Although the proportion of AI users reporting wage decreases is roughly the same for both technologies, GEAI users are far more likely to report wage increases (29.4% vs.15.0%). While a causal relationship cannot be determined, these differences in reported wage impacts may reflect GEAI users’ more positive assessment of job performance improvements after using AI (Annex Figure 2.A.1).
Percentage of AI users, by technical type
Note: AI users were asked: “How has your average gross wages (before deducting taxes and social security contributions) changed before and after the use of AI?”
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Worker characteristics and the impact of AI on actual wages
An analysis of the D.I. (Increase – Decrease) score for changes in wages before and after the use of AI indicates that male AI users, AI users with disabilities, and those with caregiving responsibilities tend to exhibit higher scores, compared to their respective counterparts. There is little variation in the D.I. score among AI users by educational background or annual income level in 2023. In contrast, middle‑aged and older AI users, as well as those in non-regular employment, tend to have lower D.I. scores than younger AI users and those in regular employment (Figure 2.17). According to the results of a generalised ordered logit model that controls for several individual attributes, male AI users are 2.4 p.p. more likely than female AI users to report “Increased it a lot” in their wages after using AI. However, the marginal effect is only significant at the 10% level, suggesting that the relationship is not particularly robust. According to the same econometric analysis, AI users with disabilities or caregiving responsibilities are significantly less likely to report “No effect” on wages and significantly more likely to report “Increased it a little” or “Increased it a lot” in their wages compared to their respective counterparts.9 In contrast, middle‑aged and older AI users, as well as those in non-regular employment, are less likely to report “Increased it a little” or “Increased it a lot” and more likely to report “No effect”, compared to their respective counterparts (Annex Table 2.A.6, Annex Table 2.A.7, Annex Table 2.A.8).10 These disparities are likely to reflect disparities in access to improvements in job performance resulting from AI use, as well as disparities in the way AI is implemented across job types – issues examined in Chapter 4.
Percentage of AI users, by gender, age, employment status, education, annual income, disability/caregiving status
Note: AI users were asked: “How has your average gross wages (before deducting taxes and social security contributions) changed before and after the use of AI?” The figure for “University degree” is the sum of four‑year university and graduate school. Respondents answered the income before taxes and social security contributions were deducted. “Low” is classified as Below JPY 2 000 000. “Middle” is classified as over JPY 2 000 000 and below JPY 8 000 000. “High” is classified as over JPY 8 000 000.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Company size and the impact of AI on actual wages
An analysis of the D.I. (Increase – Decrease) score for changes in wages before and after the use of AI reveals that AI users working in companies with more than 10 000 employees tend to show lower scores (2.9 p.p.) than, those working in companies with up to 300 workers (9.4 p.p.), suggesting larger wage benefits in SMEs (Figure 2.18).
Percentage of AI users, by company size
Note: AI users were asked: “How has your average gross wages (before deducting taxes and social security contributions) changed before and after the use of AI?”
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Occupation and the impact of AI on actual wages
An analysis of the D.I. (Increase – Decrease) scores for changes in wages before and after AI use reveals substantial variation across occupations among AI users. The D.I. (Increased – Decreased) score is relatively high among AI users working as “Plant and machine operators, and assemblers” (20.0 p.p.) and “Professionals” (14.1 p.p.). In contrast, AI users working as “Elementary occupations” or “Clerical support workers” show the D.I. (Increased – Decreased) that is either close to zero or negative, suggesting that wage decreases are more prevalent than wage increases (Figure 2.19). According to the results of a generalised ordered logit model that controls for several individual attributes, when using AI users working as “Clerical support workers” as the reference group, AI users in other occupations tend to be less likely to report “No effect” and more likely to report “Increased it a lot” in their wages after using AI. In particular, the probability of reporting “Increased it a lot” is 17.7 p.p. higher for those working as “Managers” and 13.0 p.p. higher for those working as “Professionals”. For groups such as “Elementary occupations” and “Craft and related trades workers”, the probability of reporting “Increased it a lot” increases by 8.7 p.p.; however, these groups also show a 5.4 p.p. increase in the probability of reporting “Decreased it a lot” (Annex Table 2.A.10). The results of the occupational analysis suggest that the wage benefits of AI may be influenced not only by improvements in job performance but also by the way in which task composition changes after using AI. For example, although AI users working as “Clerical support workers” report an improvement in job performance, the proportion of those experiencing a wage decrease after using AI exceeds that of those reporting a wage increase. In terms of task allocation before and after using AI, they spend less time, and may perceive the resulting improvement in work efficiency as an enhancement in job performance (Annex Figure 2.A.3). While this reduction in working hours may help improve mental health, it may also lead to a decrease in wages.
Percentage of AI users, by occupation
Note: AI users were asked: “How has your average gross wages (before deducting taxes and social security contributions) changed before and after the use of AI?”
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
The impact of AI on wage across regions
An analysis of the D.I. (Increase – Decrease) scores for changes in wages before and after AI use reveals regional variation among AI users. However, the scores are positive across all regions, suggesting that wage increases are more prevalent than wage decreases. The D.I. (Increase – Decrease) score is higher among AI users residing in “North Kanto/Koshin” (12.6 p.p.) and “Kyushu/Okinawa” (10.2 p.p.), while it is lower among those living in “Chugoku/Shikoku” (4.5 p.p.) and “Kinki” (1.7 p.p.) (Figure 2.20). AI may contribute to wage growth not only in urban centres but also in rural regions of Japan.
Percentage of AI users, by residential area
Note: AI users were asked: “How has your average gross wages (before deducting taxes and social security contributions) changed before and after the use of AI?”
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
2.2.3. The impact of AI on work processes by worker characteristics
The impact of AI on workers’ control over and pace of task performance
The share of AI users in Japan reporting that AI has “No effect” on the pace at which they perform their tasks and the control they have over the sequence in which they perform those tasks is higher than in the other surveyed countries. When AI users report such changes, the D.I. (Increased – Decreased) scores on both pace and control are positive in both the finance and insurance sector and the manufacturing sector, but remain lower than those observed in the same sectors in other countries (Figure 2.21). In other words, Japanese AI users appear less likely to be affected by AI in terms of the pace and control of their work, and even when such changes occur, the extent of change is more limited than in other countries. These findings may be one of the reasons why Japanese AI users report more moderate evaluations of the improvements in job quality after using AI compared to those in other countries surveyed (Figure 2.1).
As mentioned earlier, more moderate evaluations of the increases in the pace and control of work after using AI may be influenced by Japan’s unique employment practices and the associated work style. This may make it difficult for Japanese AI users to perceive improvements in their ability to control their work pace and tasks. The Cabinet Secretariat, the METI, and the MHLW, based on the “Grand Design and Action Plan for a New Form of Capitalism 2024 Revised Version “, which was approved by the Cabinet in June 2024, published the “Job-Based Personnel Management Guidelines” in August 2024. In the Basic Policy on Economic and Fiscal Management and Reform approved by the Cabinet in June 2025, the promotion of awareness regarding the “Job-Based Personnel Management Guidelines” is also emphasised. As mentioned earlier, while Japanese employees are typically developed as generalists, these guidelines may expand opportunities for cultivating specialist-type employees. Reskilling is encouraged by clarifying job scopes through job descriptions, thereby making it easier for employees to identify the skills and abilities they need to develop. Such a shift could help foster an environment where AI users can more easily benefit from AI. Specifically, Japanese employees with clearly defined job scopes through job descriptions may experience increased autonomy and control in the approval process for their work, as they have a clear understanding of their responsibilities and expected roles. If these employees use AI, they may find it easier to perceive benefits from using AI. Additionally, clarifying the skills and abilities employees need to develop can also facilitate reskilling for AI users, enabling them to work more effectively with AI and maximise its benefits.
Percentage of AI users
Note: AI users were asked: “How has AI changed how you work, in terms of the pace at which you perform your tasks?/the control you have over the sequence in which you perform your tasks?”
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Impact of AI on workers’ decision making
The proportions of AI users in Japan reporting that AI does not assist with decision making are roughly the same as in other surveyed countries. However, while the proportions of AI users in Japan reporting that “AI assists with decision making” are positive in both the finance and insurance and manufacturing sectors, they are lower than those observed in the same sectors in other surveyed countries (Figure 2.22). There are three possible interpretations of this result. First, Japanese AI users tend to report “I don’t know” more frequently, possibly indicating a tendency to avoid making explicit judgements. Second, Japan may be less likely than other countries to introduce AI in the workplace with the specific objective of supporting decision making. Third, even if AI is introduced with the intention of supporting decision making, Japanese users may not perceive it as such, potentially due to structural features of Japan’s employment system and culturally embedded work practices, as mentioned earlier. Fourth, Japanese AI users tend to report “I don’t know” more frequently, possibly indicating a tendency to avoid making explicit judgements.
Percentage of AI users
Note: AI users were asked: “Thinking about your job, does AI assist you with decision-making? (Yes; No; Don’t know)”
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
The proportions of AI users in both the finance and insurance sector and the manufacturing sector in Japan who “Agree” that AI helps them make better or faster decisions are roughly the same as in the other surveyed countries. While Japanese AI users may be less likely than those in other countries to say that AI supports decision making, when focussing on those who do, their assessments of AI’s effectiveness in supporting better or faster decisions are comparable to those made by users in other surveyed countries (Figure 2.23). One possible interpretation is that, while AI supports users’ own decision making in a way similar to that in other countries, the evaluation of AI’s support may be weakened during the approval process in hierarchical organisations, where input from supervisors and various stakeholders is incorporated. Furthermore, the proportions of AI users in Japan who like that AI assists them with decision making are lower than in other countries. Among Japanese AI users, the proportion of those disagreeing that AI has improved decision making is slightly higher than in other surveyed countries, while the proportion who “Neither agree nor disagree” is noticeably higher.
Percentage of workers who are assisted by AI in decision making
Note: AI users who answered that they were assisted by AI in decision making were asked: “To what extent do you agree or disagree with the following statements? AI helps me make faster decisions / AI helps me make better decisions/ I like that AI assists me with decision-making”
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
2.3.1. Company training, financial support for training, self-learning
Company-provided training and AI-driven improvements in job quality
To begin with, this section compares the perceptions of AI’s impact on job quality between users who have received company-provided training or financial support for training to work with AI and those who have not. When comparing these groups of users, two notable characteristics emerge (Figure 2.24). First, across both the finance and insurance sector and the manufacturing sector, AI users who received company-provided training are more likely to report “Improved” across all five indicators of job performance and working conditions after using AI, compared to those who did not receive such training, a pattern that is also observed in other surveyed countries. Second, when comparing the proportion of AI users reporting “Improved” after using AI between those who received company-provided training and those who did not receive such training, the gap is wider in Japan across all five indicators than in other surveyed countries, in both the finance and insurance sector and the manufacturing sector. For example, in the manufacturing sector, the average difference across the five indicators is 37.3 p.p. in Japan, compared to 23.2 p.p. in the other surveyed countries. Although differences in the characteristics of AI users may also contribute to these outcomes, the results suggest that company-provided training may have a more substantial effect in Japan than in other countries. These findings point to an important role for the Japanese Government and companies in actively promoting training and financial support initiatives that enable AI users to work effectively with AI.
Percentage of AI users, by whether they received company training so that they can work with AI
Note: Employees were asked: “Has your company provided or funded training so that you can work with AI? (Yes/No/I don't know)”. AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
The most common type of training support provided by Japanese companies is “In-house training seminars” (44.2%), followed by “Training by supervisors or senior staff in daily tasks” (33.9%), and “Training seminars at external organisations” (32.2%). These suggest that Japanese companies are promoting not only OFF-JT (off-the‑job training) but also OJT (on-the‑job training) (Figure 2.25).
Percentage of employees whose company has provided or funded training so that they can work with AI
Note: Employees whose company has provided or funded training so that they can work with AI were asked: “Please answer all contents of training or financial assistance provided by your company” Respondents (N = 717) could select multiple answers.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
The OJT approach – “Training by supervisors or senior staff in daily tasks” – appears particularly effective in enhancing job performance for AI users. AI users who received this kind of training are 16.8 p.p. more likely to report an improvement in job performance than AI users who did not receive any training. AI users also report notable improvements in job performance resulting from OFF-JT (e.g. in‑house training seminars (+9.7 p.p.)) and financial support for self-learning (e.g. supports voluntary in-house study groups (+6.0 p.p.)) (Figure 2.26).11 Accordingly, Japanese companies should promote a balanced approach to training and support – encompassing OJT, OFF-JT, and self-learning initiatives – to maximise the benefits of AI in improving job performance.
Percentage of AI users whose company has provided or funded training so that they can work with AI, by training contents
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)?” Employees whose company has provided or funded training so that they can work with AI were asked: “Please answer all contents of training or financial assistance provided by your company” Respondents (N=717) could select multiple answers. The figure shows the proportion of AI users who answered that performance is improved (a lot or a little) by AI. The number of AI users whose company has provided or funded training so that they can work with AI is 597. Training contents with fewer than 100 samples were omitted.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Self-initiated learning and AI-driven improvements in job quality
In addition to company-provided training and financial support, self-initiated learning by employees is expected to enhance the positive impact of AI on job performance and working conditions. The proportion of AI users who engaged only in self-learning12 to work with AI and reported “Improved” after using AI is higher across all five indicators than among those who received neither received company training nor engaged in self-learning. In addition, AI users who engaged in both company-provided training and self-learning reported the highest rates of improvement after using AI across all five indicators, compared to the other three groups (Figure 2.27). According to the results of a generalised ordered logit model that controls for several individual attributes, when using Japanese AI users who received neither company-provided training nor engaged in self-learning as the reference group, those who received company-provided training and engaged in self-learning are 4.4 p.p. less likely to report that their job performance “Worsened it a little” after using AI, and are 17.0 p.p. more likely to report “Improved it a little” and 19.7 p.p. more likely to report “Improved it a lot” (Annex Table 2.A.10). These findings suggest that self-learning aimed at working with AI plays a significant role in enhancing job quality improvements brought about by AI. Furthermore, integrating company-provided training or financial support with employees’ self-directed learning may help to maximise these benefits.
Percentage of AI users, by whether they received company training or engaged to self-learning
Note: Employees were asked: “Has your company provided or funded training so that you can work with AI? (Yes/No/I don't know)”. Employees were asked: “In 2023, did you engage in reskilling or upskilling to work with AI? (Yes/No/I don't know)”. AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Combined training and AI-driven improvements in various other aspects of job quality
Compared to those who engaged in neither company-provided training nor self-learning, AI users who engaged in both were less likely to report “No effect / I don’t know” across a range of job quality indicators after using AI. When changes were reported, AI users who engaged in both company-provided training and self-learning reported higher positive D.I. (Increased − Decreased) scores than people who engaged in neither, in areas such as work engagement, opportunities to learn new things on the job, opportunities for personal growth through work, work style flexibility (e.g. teleworking), and the number of days of annual paid leave taken. The improvements observed in these indicators can be interpreted as reflecting enhanced outcomes for employees. However, these users also report higher positive D.I. scores in terms of total monthly overtime hours. In contrast, those who engaged in neither company-provided training nor self-learning report negative D.I. scores for overtime hours (Figure 2.28). These findings suggest that collaborative investment in human capital by both labour and management can amplify the positive effects of AI across multiple dimensions of job quality. At the same time, they highlight a potential risk: a substantial increase in opportunities for learning and growth at work – if it leads to excessively heightened work engagement – may inadvertently contribute to excessive working hours among employees.
Percentage of AI users
Note: Employees were asked: “Has your company provided or funded training so that you can work with AI? (Yes/No/I don't know)”. Employees were asked: “In 2023, did you engage in reskilling or upskilling to work with AI? (Yes/No/I don't know)”. AI users were asked: “How has your own perception or evaluation of your work changed before and after the use of AI?” Work engagement is created using the results of responses related to enthusiasm for work, pride in work, vitality through work, and immersion in work. The number of AI user answered that they received company training and engaged to self-learning is 403. The number of AI user answered that they didn’t receive company training nor engaged to self-learning is 1 016.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Combined training and AI-driven improvements in wages
Employees who make no investment in human capital to work with AI may find it difficult to access the benefits of wage increases, even if they use AI. Employees who engaged in neither company training or self-learning are more likely to report “No effect” of AI on wages, compared to users who engaged only in self-learning (55.2%), those who received only company-provided training (24.7%), and those who engaged in both (21.6%). Among AI users who experienced a change in wages after using AI, the D.I. (Increased − Decreased) score is highest for those who engaged in both company-provided training and self-learning (25.3 p.p.). The score declines in the following order: those who received only company-provided training (17.0 p.p.), those who engaged only in self-learning (7.5 p.p.), and those who did neither (‑1.8 p.p.). Notably, among AI users who received neither company-provided training nor engaged in self-learning, the proportion reporting a decrease in their wages exceeds the proportion reporting an increase (Figure 2.29).
Percentage of AI users, by whether they received company training or engaged in self-learning
Note: Employees were asked: “Has your company provided or funded training so that you can work with AI? (Yes/No/I don't know)”. Employees were asked: “In 2023, did you engage in reskilling or upskilling to work with AI? (Yes/No/I don't know)”. AI users were asked: “How has your average gross wages (before deducting taxes and social security contributions) changed before and after the use of AI?”
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
2.3.2. Worker consultation on the use of new technology
68.0% of AI users who say their employers consult them regarding the introduction of new technologies agree that “Worries about the introduction of new technologies have been alleviated”, 72.8% agree that “Effectiveness of the introduction of new technologies has been enhanced”, and 66.5% agree that “Skills needed in the future were identified” (Figure 2.30). Alleviating employees’ anxiety about introducing a new technology in the workplace through consultations can reduce the psychological distance between them and the technology, making it less likely for the new system to be perceived as something “forced” onto them. In other words, by relieving this anxiety, it becomes less probable that employees will fall into a situation where they undervalue the system. Furthermore, by incorporating input from frontline workers prior to implementation, consultations can facilitate smoother post-deployment operations and thereby increase the overall effectiveness of the newly introduced technology. Consultations may also enable employees to more concretely envision the skills and competencies the organisation will expect them to have in the future. These outcomes are likely to foster additional enhancements in job quality through the heightened impact of human capital investment. By operating through these mechanisms, employer consultations with workers regarding the introduction of new technologies may help enhance employees’ job performance and working conditions.
Percentage of AI users whose employers consult them
Note: Employees whose employers consult workers or worker representatives regarding the use of new technologies in the workplace were asked: “On the impacts of consulting workers or worker representatives about the use of new technologies in the workplace, to what extent do you agree or disagree with the following statements? Worries about the introduction of new technologies has been alleviated/ Effectiveness of the introduction of new technologies has been enhanced/ Skills needed in the future were identified”
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Across both the finance and insurance sector and the manufacturing sector in Japan, AI users who were consulted are more likely to report “Improved” across all five indicators of job performance and working conditions after using AI, compared to those who were not consulted, a pattern similar to that observed in the other surveyed countries (Figure 2.31). According to the results of a generalised ordered logit model that controls for several individual attributes, Japanese AI users who were consulted by their employers regarding the introduction of new technologies in the workplace are 1.8 p.p. less likely to report that their job performance “Worsened it a lot” after using AI, and 2.8 p.p. less likely to report “Worsened it a little”. In contrast, they are 7.1 p.p. more likely to report “Improved it a little” and 17.8 p.p. more likely to report “Improved it a lot” (Annex Table 2.A.11).
Furthermore, consistent with the results presented in Figure 2.28 and Figure 2.29, worker consultation can also be observed to enhance improvements in various other aspects of job quality – such as work engagement – and to additionally contribute to greater increases in wages following the use of AI.
Percentage of AI users, by whether employers consult them regarding the use of new technologies in the workplace
Note: All employees were asked: “In your experience, does your employer consult workers or worker representatives regarding the use of new technologies in the workplace?" AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who said that each of these outcomes were improved (a lot or a little) by AI.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
2.3.3. Internal rules or guidelines for the appropriate use of GEAI
Given the rapid pace of innovation in AI technologies, continued communication between employers and employees – even after AI has been introduced into the workplace – will help maximise the potential benefits of AI for job performance and working conditions. One useful tool to support such ongoing communication is the development of internal regulations and employee guidelines. While this issue will be discussed in more detail in Chapter 4, among 34.8% of GEAI users in Japan report that internal rules or guidelines have been established to support appropriate use of GEAI in their work. In addition, 12.4% of AI users report that their companies are currently in the process of preparing such rules or guidelines, whereas 52.8% report that no such measures have been developed or that they are unsure whether such measures exist.
Among AI users, those working for companies that have already established internal rules or guidelines reported an average of 61.0% improvement across the five indicators of job performance and working conditions after using AI, while the corresponding figure was 39.8% for those whose companies have not yet established such rules or guidelines (Figure 2.32). Although differences in the characteristics of GEAI users may also contribute to these outcomes, the results indicate that establishing internal rules or guidelines to support employees in using GEAI appropriately in their work has a substantial effect on job performance and working conditions.
Furthermore, consistent with the results presented in Figure 2.28 and Figure 2.29, the establishment of internal rules or guidelines for working with AI can also be observed to enhance improvements in various other aspects of job quality – such as work engagement – and to additionally contribute to greater increases in wages following the use of AI.
Percentage of GEAI users, by whether internal rules or guidelines for the appropriate use of GEAI at work have been established already
Note: All employees were asked: “Have internal rules or guidelines been established to ensure that employees use generative AI appropriately in their work? (Yes; No, but my company is now preparing ; No, and my company is not preparing; I don’t know)” AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
2.3.4. Trust in own company to only use AI that is safe and trustworthy
Whether employees trust that their company to use only safe and trustworthy AI technologies is likely to be a key determinant in ensuring the effective integration of AI in the workplace, and ultimately, in enabling improvements in employees’ job performance and working conditions. Moreover, insufficient trust in the company can cause friction in communication between labour and management, and even if internal rules or guidelines are established, employees may be less inclined to engage with them or seek to understand them. In some cases, this lack of trust may also reduce employees’ motivation to engage in self-learning to effectively use AI in their current work. While a more detailed discussion is provided in Chapter 4, 65.9% of AI users in Japan report that they trust their company to use only safe and trustworthy AI technologies. In contrast, 20.7% report that they do not trust their company in this regard, and 13.4% report that they are unsure. Based on this, the following section examines differences in perceived improvements in job quality between AI users who trust their company and those who do not (including those who are unsure).
Across both the finance and insurance sector and the manufacturing sector, AI users who trust their company are more likely to report “Improved” across all five indicators after using AI compared to those who do not, a pattern similar to that observed in other surveyed countries. In addition, when comparing the proportion of AI users reporting “Improved” after using AI between those who trust their company and those who do not, the gap for some indicators of job quality tends to be wider in Japan than in other surveyed countries. In the finance and insurance sector, the average difference across the five indicators is 32.7 p.p. in Japan, compared to 21.9 p.p. in the other surveyed countries. On the other hand, in the manufacturing sector, the average difference across the five indicators is 24.2 p.p. in Japan, compared to 25.5 p.p. in the other surveyed countries. (Figure 2.33).
When AI users do not trust their companies, the impact of AI on job performance and working conditions may be limited. AI users who do not trust their company are very unlikely to report that AI either improved or worsened job performance and working conditions, so that the DI (Improved − Worsened) scores across the five indicators are either negative or only marginally positive.13 In addition, an average of 57.3% of AI users who do not trust their company report “No effect” across the five indicators of job quality. Therefore, despite having invested in the introduction of AI technologies in the workplace, companies may find that the resulting improvements in employees’ job quality are limited. This highlights the importance of fostering employee trust in these technologies through various forms of communication and the use of external third-party risk assessment bodies, thereby maximising the impact of companies’ investments in workplace technologies.
Furthermore, consistent with the results presented in Figure 2.28 and Figure 2.29, the establishment of trust that the employer uses only safe and trustworthy AI can also be observed to enhance improvements in various other aspects of job quality – such as work engagement – and to additionally contribute to greater increases in wages following the use of AI.
Percentage of AI users, by whether they trust own company to only use AI that is safe and trustworthy
Note: Al users were asked: “To what extent would you trust your company to only use AI that is safe and trustworthy?" AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who answered that each of these outcomes were improved/increased (a lot or a little) by AI.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Percentage of AI users, by technical type
Note: AI users were asked: “How do you think AI has changed your own job performance (performance)/how much you enjoy your job (enjoyment)?/your physical health and safety in the workplace (physical health)?/your mental health and well-being in the workplace (mental health)?/how fairly your manager or supervisor treats you (fairness in management)?” The figure shows the proportion of AI users who said that each of these outcomes were improved (a lot or a little) by AI.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 809 observations (AI users). The reference category is regular employment. Controls include gender, age group, educational background, and occupational category. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 809 observations (AI users). Controls include gender, age group, educational background, employment status and occupational category. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 809 observations (AI users). The reference category is AI users who reported having no disabilities. Controls include gender, age group, and employment status. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 729 observations (AI users). Controls include gender, age group, educational background, and employment status. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 729 observations (AI users). Controls include gender, age group, educational background, employment status, and company size. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Percentage of AI users, by country
Note: All employees were asked: “Do you think that AI will have an impact on wages in your sector in the next 10 years? (Yes, AI will increase wages; Yes, AI will decrease wages; No, AI will not impact wages; Don't know)” The figure shows the proportion of AI users who answered that wages would increase due to AI in next 10 years.
Source: OECD worker survey on the impact of AI on the workplace (2022), JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 761 observations (AI users). Controls not listed in the table include educational background, company size, occupation, and average weekly working hours. The marginal effects of educational background were not statistically significant in any case. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 689 observations (AI users). The reference category is AI users who reported having no disabilities. Controls include gender, age group, educational background, and employment status. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 689 observations (AI users). Controls include gender, age group, educational background, employment status, and occupation. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 761 observations (AI users). Controls include gender, age group, educational background, employment status, company size, occupation, and average weekly working hours. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Percentage of AI users, by occupation and changes in time spent on tasks during a typical workday after using AI
Note: AI users were asked: “How has your time spent on tasks changed before and after the use of AI? Routine and repetitive tasks/Physical tasks/Tasks in hazardous place/Identifying problems and solving them using creativity/ Managing and motivating team members or subordinates/Analysing data and information to make decisions based on the results (Increased it a lot; Increased it a little; Decreased it a little; Decreased it a lot; No effect; I don't know)” “Physical tasks” include activities such as standing work, operating machinery or vehicles, and carrying loads. “Tasks in hazardous place” include activities at high places, in extremely hot or cold places, in places with a lot of machinery. The marker indicates changes in the time spent on tasks during a typical workday after using AI, represented by the D.I. (Increased – Decreased). The bar line shows the average proportion of respondents who reported “No effect” on working hours across six tasks.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 809 observations (AI users). Controls include gender, age group, employment status, and occupation. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
Note: Estimates are based on 1 809 observations (AI users). Controls include gender, age group, educational background, employment status, company size, and occupation. Due to a violation of the proportional odds assumption (Brant test), a generalised ordered logit model was employed. *** Significant at the 1% level, ** 5% level, * 10% level.
Source: JILPT worker survey on the impact of the introduction of AI into the workplace on working practices (2024).
References
[13] Acemoglu et al. (2022), “Artificial Intelligence and Jobs: Evidence from Online Vacancies”, Journal of Labor Economics, Vol. 40/S1, https://doi.org/10.1086/718327.
[15] Acemoglu, D. (2024), The Simple Macroeconomics of AI, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w32487.
[14] Albanesi et al. (2023), “New Technologies and Jobs in Europe”, http://www.nber.org/papers/w31357 (accessed on 21 June 2024).
[18] Alekseeva, L. et al. (2021), “The demand for AI skills in the labor market”, Labour Economics, Vol. 71, p. 102002, https://doi.org/10.1016/J.LABECO.2021.102002.
[40] Bakker,A.B. & Demerouti,E. (2008), “Towards a model of work engagement”, Career Development International, Vol. Vol. 13/No. 3, pp. 209-223, https://doi.org/10.1108/13620430810870476.
[12] Bankins, S. et al. (2024), “A multilevel review of artificial intelligence in organizations: Implications for organizational behavior research and practice”, Journal of Organizational Behavior, Vol. 45/2, pp. 159-182, https://doi.org/10.1002/JOB.2735.
[8] Brione, P. (2020), My boss the algorithm: Research An ethical look at algorithms in the workplace, ACAS.
[23] Brynjolfsson, E., D. Li and L. Raymond (2023), “Generative AI at Work”, National Bureau of Economic Research Working Paper Series, Vol. No. 31161, https://doi.org/10.3386/W31161.
[11] Chowdhury, S. et al. (2022), “AI-employee collaboration and business performance: Integrating knowledge-based view, socio-technical systems and organisational socialisation framework”, Journal of Business Research, Vol. 144, pp. 31-49, https://doi.org/10.1016/J.JBUSRES.2022.01.069.
[24] Dell’Acqua, F. et al. (2023), “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality”, SSRN Electronic Journal, https://doi.org/10.2139/SSRN.4573321.
[16] Felten, E., M. Raj and R. Seamans (2019), “The Occupational Impact of Artificial Intelligence: Labor, Skills, and Polarization”, SSRN Electronic Journal, https://doi.org/10.2139/SSRN.3368605.
[17] Fossen, F. and A. Sorgner (2022), “New digital technologies and heterogeneous wage and employment dynamics in the United States: Evidence from individual-level data”, Technological Forecasting and Social Change, Vol. 175, p. 121381, https://doi.org/10.1016/J.TECHFORE.2021.121381.
[25] Georgieff, A. (2024), “Artificial intelligence and wage inequality”, OECD Artificial Intelligence Papers, No. 13, OECD Publishing, Paris, https://doi.org/10.1787/bf98a45c-en.
[19] Green, A. and L. Lamby (2023), “The supply, demand and characteristics of the AI workforce across OECD countries”, OECD Social, Employment and Migration Working Papers, No. 287, OECD Publishing, Paris, https://doi.org/10.1787/bb17314a-en.
[33] Hamaguchi, K. (2009), The new employment society - towards a restructuring of the human resource management system..
[34] Inoue, K. (2019), “The reality of small enterprises engaging in diverse work practices”, JFCRI Discussion Papers, Vol. 44, https://www.jfc.go.jp/n/findings/pdf/ronbun1908_01.pdf.
[39] Iwamoto, K. (2021), Difficulties in introducing DX to SMEs include factors that differ from those in larger companies, https://www.rieti.go.jp/users/iwamoto-koichi/serial/129.html (accessed on 10 February 2025).
[31] Iwata et al. (1995), “Japan–U.S. comparison of responses to depression scale items among adult workers.”, Psychiatry Research, Vol. 58(3), pp. 237–245, https://doi.org/10.1016/0165-1781(95)02734-E.
[43] Japan Productivity Center (2023), Survey on attitudes toward telework, https://www.jpc-net.jp/research/assets/pdf/appendix01_tele-work_report.pdf (accessed on 18 February 2025).
[26] Kabir, S. et al. (2024), “Is Stack Overflow Obsolete? An Empirical Study of the Characteristics of ChatGPT Answers to Stack Overflow Questions”, Conference on Human Factors in Computing Systems - Proceedings, https://doi.org/10.1145/3613904.3642596.
[20] Kanazawa, K. et al. (2022), “AI, Skill, and Productivity: The Case of Taxi Drivers”, http://www.iza.org (accessed on 22 August 2024).
[38] Kawakami, Y. (2013), “Managerial Approach to Study of SMEs : Specific Managerial Research”, Fukuoka University review of commercial sciences, Vol. 58(3), pp. 341-362, https://fukuoka-u.repo.nii.ac.jp/record/1671/files/C5803_0341.pdf.
[27] Kreitmeir, D. and P. Raschky (2024), “The Heterogeneous Productivity Effects of Generative AI”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.4745624.
[1] Lane, M., M. Williams and S. Broecke (2023), “The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers”, OECD Social, Employment and Migration Working Papers, No. 288, OECD Publishing, Paris, https://doi.org/10.1787/ea0a0fe1-en.
[28] Lincoln, J. (1989), “Employee Work Attitudes and Management Practice in the U.S. and Japan: Evidence from a Large Comparative Survey”, California Management Review, Vol. 32, pp. 89-106, https://doi.org/10.2307/41166736.
[10] Makarius, E. et al. (2020), “Rising with the machines: A sociotechnical framework for bringing artificial intelligence into the organization”, Journal of Business Research, Vol. 120, pp. 262-273, https://doi.org/10.1016/J.JBUSRES.2020.07.045.
[9] Meijer, A., L. Lorenz and M. Wessels (2021), “Algorithmization of Bureaucratic Organizations: Using a Practice Lens to Study How Context Shapes Predictive Policing Systems”, Public Administration Review, Vol. 81/5, pp. 837-846, https://doi.org/10.1111/PUAR.13391.
[41] MHLW (2019), White Paper on the Labour Economy, https://www.mhlw.go.jp/english/wp/l-economy/2019/summary.pdf (accessed on 18 February 2025).
[4] Milanez, A. (2023), “The impact of AI on the workplace: Evidence from OECD case studies of AI implementation”, OECD Social, Employment and Migration Working Papers, No. 289, OECD Publishing, Paris, https://doi.org/10.1787/2247ce58-en.
[2] Milanez, A., A. Lemmens and C. Ruggiu (2025), “Algorithmic management in the workplace: New evidence from an OECD employer survey”, OECD Artificial Intelligence Papers, No. 31, OECD Publishing, Paris, https://doi.org/10.1787/287c13c4-en.
[3] Moore, P. (2019), “OSH and the Future of Work: Benefits and Risks of Artificial Intelligence Tools in Workplaces”, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 11581 LNCS, pp. 292-315, https://doi.org/10.1007/978-3-030-22216-1_22.
[7] Moore, V. (2018), The Threat of Physical and Psychosocial Violence and Harassment in Digitalized Work, International Labour Office, Geneva.
[22] Noy, S. and W. Zhang (2023), “Experimental evidence on the productivity effects of generative artificial intelligence”, Science, Vol. 381/6654, pp. 187-192, https://doi.org/10.1126/SCIENCE.ADH2586/SUPPL_FILE/SCIENCE.ADH2586_MDAR_REPRODUCIBILITY_CHECKLIST.PDF.
[36] OECD (2025), Generative AI and the SME Workforce: New Survey Evidence, OECD Publishing, Paris, https://doi.org/10.1787/2d08b99d-en.
[21] Peng, S. et al. (2023), “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot”.
[35] Sakamoto,T. & Takayama,J. (2023), Mastering business tools and handling multiple tasks with a small team, https://www.works-i.com/research/project/futureofwork/automation04/detail008.html (accessed on 6 February 2025).
[29] Scholz et al. (2002), “Is general self-efficacy a universal construct? Psychometric findings from 25 countries”, European Journal of Psychological Assessment, Vol. 18(3), pp. 242–251, https://doi.org/10.1027/1015-5759.18.3.242.
[42] Shimazu et al. (2012), “Validation of the Japanese version of the recovery experience questionnaire”, Journal of Occupational Health, Vol. 54/3, pp. 196–205, https://doi.org/10.1539/joh.11-0220-oa.
[30] Shimazu et al. (2010), “Why Japanese workers show low work engagement: An item response theory analysis of the Utrecht Work Engagement scale”, BioPsychoSocial Medicine, https://doi.org/10.1186/1751-0759-4-17.
[44] Shimazu,A. & Schaufeli,W.B. (2009), “Is workaholism good or bad for employee well-being? The distinctiveness of workaholism and work engagement among Japanese employees”, Industrial Health, Vol. 47, pp. 495–502, https://doi.org/10.2486/indhealth.47.495.
[32] Shimazu, A. (2022), New edition Work Engagement - Positive Mental Health for a Vibrant Everyday Life, Roudo Chousakai.
[6] Todolí-Signes, A. (2021), “Making algorithms safe for workers: occupational risks associated with work managed by artificial intelligence”, European Review of Labour and Research, Vol. 27/4, pp. 433-452, https://doi.org/10.1177/10242589211035040.
[37] Yamaguchi, T. (2012), “The Influence of Proximity in Small Business Management”, Shogakuronkyu (The Kwansei Gakuin University), Vol. 59(3), pp. 71-91, https://kwansei.repo.nii.ac.jp/record/20395/files/59-3-4.pdf.
[5] Yamamoto, I. (2019), The impact of AI and information technologies on worker stress, https://cepr.org/voxeu/columns/impact-ai-and-information-technologies-worker-stress.
← 1. Work Engagement is a positive, fulfilling, work-related state of mind that is characterised by vigour, dedication, and absorption. Individuals with high work engagement feel proud and purposeful in their work, approach tasks with enthusiasm, and derive energy from their work, leading to a vibrant and fulfilling work experience. This definition is based on the Utrecht Work Engagement Scale (UWES), which assesses these three dimensions of work engagement.
← 2. (Dell’Acqua et al., 2023[24]) warn that LLMs may reduce the performance of less experienced workers when used for complex tasks.
← 3. One possible reason why GEAI appears to have a greater positive impact on job quality in Japan is that, rather than being applied to complex tasks such as coding or debugging, it is primarily used for more supportive tasks, such as drafting documents or texts (Annex Figure 1.A.2).
← 4. AI users with a university degree are also 3.3 p.p. less likely to report that AI “Improved it a lot” compared to those without a university degree.
← 5. When using AI users aged 15‑34 as the reference group, the generalised ordered logit model controlling for several individual attributes shows statistically significant results for those aged 35‑49 in the following categories: “Worsened it a lot” (marginal effect: ‑2.2%, z-value: ‑2.4), “Worsened it a little” (‑2.9%, ‑1.9), “No effect” (9.2%, 4.0), and “Improved it a lot” (‑5.6%, ‑3.0). Similarly, for AI users aged 50 and over, statistically significant results were found in: “Worsened it a lot” (‑2.3%, ‑2.1), “Worsened it a little” (‑7.9%, ‑4.0), “No effect” (19.2%, 6.82), and “Improved it a lot” (‑10.5%, ‑4.6).
← 6. AI users in non-regular employment are also 5.3 p.p. less likely to report that AI “Worsened it a lot” compared to those in regular employment.
← 7. Due to the small sample size, “Skilled agricultural, forestry and fishery workers”, “Craft and related trades workers”, “Plant and machine operators, and assemblers”, “Elementary occupations”, and “Armed forces occupations” were combined into a single group for estimation purposes.
← 8. It is important to note that around 50% of respondents stated that there was “No significant change” in their physical and mental health after implementing telework.
← 9. According to the econometric analysis, AI users with disabilities or caregiving responsibilities are significantly less likely to report “No effect” on their wages after using AI, compared to their respective counterparts. It is important to note, however, that some of this shift is associated with an increased likelihood of reporting “Decreased it a little”.
← 10. Regarding the impact of educational background on wage changes after using AI, no statistically significant effects were observed across any category. Similarly, when using middle‑income AI users as the reference group for 2023 annual income, no significant effects were found for high-income AI users in any response category. In contrast, for low-income AI users, compared to the middle‑income reference group, the likelihood of reporting “No effect” decreased by 15.1 p.p. (z-value = –4.92), while the likelihood of reporting “Decreased it a little” increased by 4.5 p.p. (z-value = 1.75), and “Increased it a little” increased by 9.5 p.p. (z-value = 3.06). These results suggest that among low-income AI users, the adoption of AI may contribute to widening wage disparities.
← 11. It is important to note an important limitation of this analysis: ideally, the analysis would be limited to AI users who received only one specific type of training or financial support. However, because many users have report having received multiple forms of training or financial support, narrowing the analysis in this way would lead to a substantially reduced sample size. Therefore, the findings of this analysis should be interpreted with some degree of caution, as AI users who received a particular form of training or financial support may have also benefited from other types of support.
← 12. Self-learning AI users are defined here as those who answered “Yes” to the question, “In 2023, did you engage in reskilling or upskilling to work with AI?” and “No” to the question, “Has your company provided or funded training to help you work with AI?”. Based on this definition, the number of self-learning AI users was 241. Additionally, based on the combination of responses to these questions, 1 854 AI users in Japan were classified into the following four categories: (1) those who received both company-provided training and engaged in self-learning, (2) those who received only company-provided training, (3) those who engaged only in self-learning, and (4) those who neither received company-provided training nor engaged in self-learning.
← 13. Across all industries in Japan, the D.I. (Improved − Worsened) scores for AI users who do not trust that their company uses only safe and trustworthy AI technologies are as follows: +31.0 p.p. for job performance, −1.3 p.p. for enjoyment, +3.3 p.p. for mental health, +3.6 p.p. for physical health, and +6.8 p.p. for fairness in management.