The noise around generative AI continues to grow louder. But beyond the headlines and hype cycles what’s really happening to jobs? Siân Harrington checks out ILO's latest research
A new working paper from the International Labour Organization (ILO), published in May 2025, delivers a much-needed dose of rigour. Drawing on nearly 30,000 real-world tasks, more than 50,000 human assessments and expert validation it presents a refined global index of occupational exposure to generative AI. The results bring clarity to what AI can and can’t do in the workplace right now.
A more precise view of exposure
The ILO’s 2025 study builds on its original 2023 framework, expanding the task database tenfold using Poland’s detailed occupational classification system. This allowed for far more granular analysis than the global ISCO-08 classification alone.
Crucially, the research combined algorithmic predictions from models like GPT-4o and Gemini with a national survey of 1,640 people and a global panel of experts. Their combined input shaped an AI assistant capable of predicting task-level automation exposure across job classifications. The result is a highly detailed, internationally relevant exposure map.
One in four jobs touched by GenAI but transformation, not replacement, is the story
According to the updated global estimates 24% of all employment worldwide sits within occupations now considered exposed to GenAI to some degree. Just 3.3% of jobs fall into the highest exposure category – those where most tasks could feasibly be handled by GenAI without human input.
Yet even within that top category, the study found almost no jobs made up entirely of automatable tasks. Human involvement remains essential across most roles, confirming that the more likely outcome of GenAI adoption is job transformation rather than job elimination.
Clerical work still tops the list but professionals are catching up
As in the 2023 index clerical support roles remain the most exposed. Data entry clerks, payroll clerks and typists continue to show the highest automation potential. However, the 2025 index notes a downward revision in many task-level scores, based on practical insights from real users. For example, while tools like ChatGPT can assist with scheduling meetings or drafting emails these tasks often still require human oversight, especially in nuanced contexts.
Meanwhile, exposure is rising in certain digitised and technical roles. Financial analysts, multimedia developers, software professionals and database administrators have seen their scores increase, thanks to GenAI’s rapidly expanding capabilities, including multimodal content generation and autonomous task handling.
Exposure is gendered and worse in wealthier economies
One of the most striking findings in the new data is the gender gap. Globally, 4.7% of women are in jobs with the highest level of GenAI exposure, compared to 2.4% of men. In high-income countries this gap widens dramatically: 9.6% of women’s jobs fall into the highest exposure category, against just 3.5% for men.
This reflects the gendered distribution of clerical and customer-facing roles, which are both more exposed and more common among female workers. It also highlights how occupational segregation compounds digital risk – a dynamic that’s becoming more acute as GenAI expands its reach.
Location matters. So does income.
The report confirms that income level is a key factor in exposure. In low-income countries, just 11% of total employment is in GenAI-exposed occupations. In high-income countries, that number jumps to 34%.
This difference is shaped by infrastructure, digital access and workforce structure. But it also underscores the risk of a growing global divide in who is most impacted, and who benefits, from GenAI.
Caution, not panic: exposure is not impact
A critical distinction in the ILO’s analysis is that exposure does not equal automation. These are best-case, upper-limit scenarios – estimates of what could be done with GenAI, not what will be. Many factors stand in the way of adoption, including:
- Inadequate infrastructure or broadband access
- Low digital skills among workers or managers
- Cultural and institutional resistance to automation
- Economic barriers to scaling AI solutions
There’s also the simple matter of usefulness. Just because a task can be automated doesn’t mean it should be. In practice many organisations will adopt GenAI as a complement, not a substitute – using it to enhance efficiency rather than replace people outright.
What next? A call for human-centred AI adoption
The ILO report ends on a clear note: the transition to GenAI in the workplace must be managed and not left to chance. This means active investment in:
- Reskilling and adaptation support for workers in exposed roles
- Inclusive technology design that involves end users
- Social dialogue between employers, employees and policymakers
- Responsible adoption that enhances, rather than erodes, job quality
Workers know their jobs best. They should be at the table when decisions about AI adoption are made. As the report notes: “The key to unlocking the productivity benefits of GenAI lies not in the search for labour savings, but in the extent to which human expertise can be complemented by new technological capacities.”