What do we actually know, worldwide, about how AI and robotics are changing work? Here we collect verified findings – from estimates of how many jobs are affected, through effects already measured, to forecasts that lie far apart. Checked on 29 October 2026.
‘Exposed’ means that some of a job’s tasks could technically be done or supported by AI. It is not a prediction of job losses but a statement about technical possibilities.
Estimates vary widely depending on the method. What they share is this: a large part of work is going to change. Whether it disappears is another question.
According to the International Labour Organization (ILO), about one in four workers worldwide is in an occupation with some degree of exposure to generative AI. The ILO distinguishes four levels of exposure. Its central message: most jobs are more likely to be transformed than made redundant.
The World Economic Forum surveyed more than 1,000 employers across 55 economies. They expect around 170 million new jobs and 92 million lost jobs by 2030 – a net gain of 78 million. They also expect 22% of jobs to change structurally, and 41% plan to cut staff because of AI automation. These are employers’ expectations, not measurements.
Source: WEF Future of Jobs Report, 2025
According to the consultancy McKinsey (McKinsey Global Institute), around 57% of working hours in the US could technically be automated with technology available today. At the same time, the study finds that more than 70% of the most sought-after skills remain relevant. Technically possible does not mean it will happen.
Source: McKinsey Global Institute, 2025
According to an analysis by the International Monetary Fund (Staff Discussion Note 2024/001, January 2024), almost 40 per cent of global employment is exposed to AI. In advanced economies the figure is about 60 per cent of jobs, in emerging markets 40 per cent and in low-income countries 26 per cent. Exposed does not mean replaced: some jobs could benefit from AI, while others could come under pressure.
Source: IMF (Kristalina Georgieva's blog on the Staff Discussion Note), 2024
Since ChatGPT launched at the end of 2022, the data have so far shown no widespread job losses caused by generative AI – neither in unemployment nor in wages or working hours.
That does not mean nothing is happening. Studies see shifts in how work is organised, and early signals among young people starting their careers. Most of the data come from the US and Denmark; whether wages tend to rise or stay flat remains open.
The Yale Budget Lab examined how the US occupational mix has changed since ChatGPT. The change is only about one percentage point greater than in the early years of the internet. The shares of workers with low, medium and high AI exposure have remained largely stable.
Source: Yale Budget Lab, 2025
US unemployment rose from 3.4% (April 2023) to 4.3% (March 2026). A follow-up study by the Yale Budget Lab, however, finds no measurable AI effects on employment shares or real wages. AI is therefore probably not (yet) the reason for the slowdown.
Source: Yale Budget Lab, 2026
A study using Danish survey and register data finds no measurable effects of generative AI on earnings or working hours. The authors rule out effects larger than 2% in the first two years after ChatGPT’s launch. What has changed is how work is organised, for instance through new tasks such as overseeing AI.
Source: Humlum & Vestergaard, NBER, 2025
A research team led by Chen and Evans finds no systematic effect of large language models on unemployment in the US. In highly exposed occupations, weekly earnings have even risen slightly. The study is a preprint that has not yet been peer-reviewed.
Source: Chen et al., arXiv, 2025
The clearest warning sign concerns young people starting out. In AI-exposed occupations in the US, 22- to 25-year-olds are less often employed than peers in other occupations – not because they are being dismissed, but because fewer are being hired.
The finding is disputed: the research team itself describes a correlation, not a proven cause-and-effect relationship. Other studies have so far found no broad effects, and comparable data for Europe are largely missing.
A team led by Stanford economist Erik Brynjolfsson analyses payroll data from the US provider ADP. As of August 2026, employment of 22- to 25-year-olds in AI-exposed occupations is 19% below the path of less exposed peers. The decline appears where AI replaces work, not where it complements it. This is a relative figure, not an absolute fall.
According to the Stanford AI Index 2026, the number of workers aged 22 to 25 in software development and customer service fell between 2021 and 2025, while older groups grew. In a company survey, 32% of firms expect AI to reduce their headcount in the following year, 43% expect no change and 13% an increase.
Source: Stanford AI Index, 2026
According to the OECD Employment Outlook 2026 (July 2026), the OECD-wide unemployment rate stood at 4.9 per cent in May 2026, with employment at a record 670 million. Unemployment among young people, however, has risen, among graduates and in a few countries also among young people without a degree.
According to the report, evidence that AI is behind this is so far limited. The OECD names the economic cycle and longer-term shifts in the skills employers demand as more significant drivers.
At the level of individual tasks, productivity gains from AI are well documented. Whether and when they will show up clearly across the whole economy is still open.
It is striking who benefits: in one widely cited study, newcomers gain the most. At the same time, AI is increasingly used to complete tasks on its own rather than merely to help.
A study of more than 5,000 customer service workers shows that, with an AI assistant, they resolved 14% more cases per hour on average. New and less skilled workers improved by 34%, experienced ones hardly at all. The data date from 2023.
According to the Stanford AI Index, the share of organisations using AI rose from 55% in 2023 to 88% in 2025. Generative AI spread faster in three years than the internet or the PC once did. Use, however, does not yet mean that work is being replaced.
Source: Stanford AI Index, 2026
According to the AI company Anthropic, which analyses usage data from its chatbot Claude, the share of conversations in which the AI largely takes over tasks on its own rose from 27% (December 2024) to 39% (mid-2025). For the first time, automation outweighed assistance. These are the company’s own data about its own product.
Source: Anthropic Economic Index, 2025
There is broad agreement about the data so far: the effects on the economy as a whole are small. Experts disagree about how quickly AI will become more capable and how fast companies will adopt it.
The range is correspondingly wide – from a modest productivity gain to scenarios in which human labour loses much of its value.
MIT economist Daron Acemoglu estimates that AI will raise total factor productivity by no more than 0.66% over ten years; taking hard-to-learn tasks into account, by less than 0.53%. At the same time, AI could widen the gap between capital and labour income.
Source: Acemoglu, NBER, 2024
In its baseline scenario, the investment bank Goldman Sachs expects 6 to 7% of US workers to be displaced if AI is widely adopted (range 3 to 14%). According to the bank, these effects typically fade after about two years. With full adoption in advanced economies, it expects labour productivity to be around 15% higher.
Source: Goldman Sachs, 2025
Economists Anton Korinek and Donghyun Suh model scenarios up to artificial general intelligence (AGI). Whether wages rise or fall depends on the race between automation and capital accumulation. If work becomes fully automatable, wages can collapse, possibly even before that point. These are model scenarios, not predictions.
Source: Korinek & Suh, NBER, 2024
For robots there are hard figures, at least for industry. Annual installations of industrial robots have more than doubled in ten years and are concentrated in a few countries, above all China.
For humanoid robots, by contrast, there are mainly forecasts from investment banks. An independent statistic on actual deployments is still missing.
In 2024, 542,000 industrial robots were installed worldwide, more than twice as many as ten years earlier. The operational stock grew to around 4.66 million. With 295,000 installations, China accounted for 54%; Germany installed almost 27,000.
Source: IFR World Robotics, 2025
On global average, there are 132 robots per 10,000 manufacturing workers (2024 data). South Korea leads with 1,220, followed by Singapore with 818. Germany stands at 449, Japan at 446, the US at 307 and China at 166.
Source: IFR, 2026
The investment bank Goldman Sachs estimates the market for humanoid robots at 38 billion US dollars by 2035, with around 1.4 million units in 2035. It says manufacturing costs per unit have already fallen by about 40%. Such figures are scenarios for investors, not observed deployments.
Source: Goldman Sachs, 2024
In rich countries more jobs are exposed to AI, because more office and knowledge work is done there. Poorer countries are not protected by this, however: where reliable internet is lacking, productivity gains fail to materialise, while jobs that are already digitally connected can still be at risk.
Worries are unevenly distributed too. Fear of job loss is greatest in emerging economies in South-East Asia and lowest in the US and Japan.
A joint paper by the ILO and the World Bank (March 2026, 135 countries) shows that exposure is higher in advanced economies. In poorer countries, it is precisely the workers at risk of automation who are often already online, while those who could benefit from AI frequently lack reliable internet. The authors call this ‘disruption without dividend’.
Source: ILO & World Bank, 2026
For Latin America and the Caribbean, the ILO and the World Bank estimate that 26 to 38% of jobs are exposed to generative AI. 8 to 14% could become more productive, while 2 to 5% face full automation. Up to half of the productivity potential, around 17 million jobs, is blocked by a lack of digital access. According to the study, low digitalisation offers no protection against automation.
Source: ILO & World Bank, 2024
According to the AI company Anthropic, use of its chatbot Claude is closely linked to a country’s prosperity: 1% higher economic output per head goes along with around 0.7% more use. Per head, it is used most in Israel, Singapore, Australia, New Zealand and South Korea.
Source: Anthropic Economic Index, 2025
In a 30-country survey by the market research firm Ipsos (2025), an average of 35% expect AI is likely to replace their current job within the next five years. Values are highest in South-East Asia, led by Thailand, and lowest in the US and Japan. Notably, Americans are at the same time the most nervous about AI products.
Source: Ipsos AI Monitor, 2025
The World Bank's World Development Report 2026 ("The Promise of Artificial Intelligence", launched on 4 August 2026) estimates that 4.5 per cent of existing jobs in low- and middle-income countries are at risk of automation by generative AI, compared with 14.2 per cent in high-income countries, more than three times as many.
On the other hand, 16.2 per cent of jobs in developing economies could see their productivity meaningfully boosted by AI, only slightly below the 18.7 per cent in rich countries. According to the report, whether this happens depends on electricity, internet access, education and institutions.