Which jobs are most at risk from AI?

The 0–10 AI-exposure score on this site measures how much of an occupation's day-to-day tasks generative AI can already perform or assist with. A score of 0 means the AI tools available today have little or no ability to handle the core activities of that job, while a score of 10 indicates that almost every task can be done or supported by current generative AI. The score is derived by analysing the detailed task lists of nearly 19,305 occupations across 42 countries, covering over 2.8 billion workers. The global employment-weighted average exposure score is 4.8, meaning that across all jobs and countries, roughly half of typical workplace tasks are already within reach of generative AI. It is crucial to understand that this score reflects task exposure, not job loss – it flags where AI is most likely to reshape how work is done, not that the job itself will disappear.

Exposure is emphatically not the same as automation or elimination. A high score means that many tasks in that occupation can be performed or enhanced by generative AI, but it does not imply that the role will become obsolete. In fact, the highest-exposure industries globally include Business, Finance & Legal (average 8.1) and IT & Digital (8.0), where AI can handle document drafting, data analysis, and code generation. The lowest-exposure industries are Trades & Construction (1.9) and Agriculture & Environment (3.7), where physical and manual tasks dominate. Yet even in low-scoring sectors, AI may assist with scheduling, reporting, or design. The score is a starting point for thinking about how work might change, not a prediction that any particular individual's job will vanish.

To explore the data, you will find a treemap on the site. In this visualization, each tile represents an occupation, and its area corresponds to the size of the workforce in that occupation globally or for a selected country. The colour of the tile indicates the AI-exposure score: green for low exposure, shading through yellow to red for high exposure. This means that large red tiles show occupations with many workers and high task exposure – for example, clerks or customer service representatives. Small green tiles represent niche roles with low exposure, such as craft trades. Scores are ranked on a single global percentile scale, so you can compare an occupation's exposure across different countries. Because the scale is consistent, a score of 6 means the same level of task exposure whether you look at the United Kingdom or Japan.

Average exposure varies by country, reflecting differences in economic structure. Among the 42 countries, Iceland has the highest employment-weighted average at 6.1, followed by Singapore and Luxembourg at 6.0. At the other end, India averages 3.9, with Thailand at 4.2 and Mexico at 4.5. These differences arise from the mix of industries: countries with larger shares of business services and IT tend to have higher averages, while those heavy in agriculture and construction score lower. However, every country includes occupations across the full 0–10 range. For instance, the United States, with 803 occupational categories covering over 146 million workers, has an average of 5.3, reflecting a diverse economy.

Remember that the AI-exposure score is a starting point, not a final verdict. It helps identify where generative AI is most likely to change tasks, prompting workers and employers to consider complementary skills – such as critical thinking, creativity, or emotional intelligence – that AI cannot easily replicate. No score can predict what will happen to any individual's job, because workplace outcomes depend on choices by organisations, regulations, and economic factors. Use these scores as a guide to understanding the landscape of AI and work, and as a tool to think about future skills and training needs.

The exposure of occupations to generative artificial intelligence varies substantially by industry, and this variation is much greater than the variation between countries. Across the 42 countries and over 19,000 occupations covered, the global employment-weighted average exposure score stands at 4.8 on a scale from 0 to 10. However, when industry averages are considered, the scores range from a high of 8.1 in Business, Finance and Legal to a low of 1.9 in Trades and Construction. This spread of more than six points dwarfs the roughly two-point spread between the highest- and lowest-exposure countries, which range from 6.1 in Iceland to 3.9 in India. The implication is clear: the nature of the work itself is a far stronger determinant of exposure than the country in which a worker is based.

The highest-exposure industries are those dominated by office-based, information-intensive tasks that involve processing, analysing, and generating written or numerical content. IT and Digital scores 8.0, and Business, Finance and Legal scores 8.1. In these sectors, a large share of occupations involve writing reports, coding software, reviewing legal documents, performing financial calculations, and managing data – all activities that generative AI, particularly large language models, can perform or assist with. The tasks are often routine cognitive work, even if they require expertise, because they follow identifiable patterns and rely on symbolic manipulation. This makes them highly amenable to automation or augmentation by AI systems that can produce plausible text, code, or analyses. Consequently, workers in these fields are likely to see their tasks reshaped by generative AI more than those in most other industries.

Education and Community, at 6.0, and Engineering and Infrastructure and Government and Public Sector, both at 5.9, are also well above the global average. While these industries include roles that involve physical presence – such as teaching, engineering site work, or public service delivery – a significant proportion of their workforce is engaged in administrative, planning, documentation, and communication tasks. Creating lesson plans, writing policy documents, drafting engineering specifications, and managing public records are all activities that involve structured information and can be supported by generative AI. Even Healthcare and Care, at 5.0, sits exactly at the global average, reflecting the dual nature of the sector: alongside hands-on patient care, there is a substantial amount of administrative work, medical record keeping, diagnosis support, and patient communication that can be augmented by AI tools.

At the lower end of the exposure range, industries characterised by manual, physical, or in-person work have much lower scores. Trades and Construction, at 1.9, stands out as the least exposed industry by a considerable margin. Occupations such as electricians, plumbers, carpenters, and construction labourers require physical dexterity, on-site problem-solving, and the ability to manipulate materials in unpredictable environments – capabilities that current generative AI systems do not possess. Agriculture and Environment follows at 3.7, where tasks like planting, harvesting, animal husbandry, and environmental monitoring are similarly physical and location-specific. Transport, Logistics and Mining, at 4.3, includes some routine cognitive tasks such as route planning or logistics coordination, but many roles, like truck driving or mining extraction, involve operating machinery and responding to real-world conditions. Creative, Media and Personal Services, at 4.5, is also below average; while some creative tasks like copywriting or image generation can be automated, many roles require interpersonal interaction, artistic judgement, or physical services such as hairdressing or massage.

Hospitality, Retail and Tourism, at 4.9, is the only low-exposure industry that sits close to the global average. This reflects a mixed task profile: some activities such as booking systems, inventory management, and customer service communications can be assisted by AI, while others like food preparation, cleaning, and face-to-face sales rely on manual skills and human interaction. The overall pattern underscores that the most exposed industries are those centred on the creation and manipulation of information in an office setting, while the least exposed are those grounded in physical labour, manual dexterity, or direct personal care. The gap between the highest industry score (8.1) and the lowest (1.9) is over six points, far wider than the roughly two-point gap between the most exposed country (Iceland, 6.1) and the least exposed (India, 3.9). This reinforces that the type of work matters far more than the national context in determining how generative AI may reshape a job's tasks.

It is important to note that exposure to generative AI does not directly imply job losses; rather, it indicates the proportion of tasks within an occupation that could potentially be performed or augmented by AI systems. In high-exposure industries, many roles may be transformed rather than eliminated, with AI handling routine cognitive tasks while workers focus on higher-level judgment, creativity, and interpersonal skills. In low-exposure industries, the physical and social nature of the work provides a natural buffer, though even there, administrative and planning subtasks may be affected. The wide industry variation shows that the impact of generative AI will be felt very differently across sectors, and that policies and training programmes should be tailored accordingly. The data from 42 countries, covering nearly 2.8 billion workers, provides a robust foundation for understanding these patterns.

If you work in an occupation with a high exposure score, such as in Business, Finance and Legal (global average 8.1 out of 10) or IT and Digital (8.0), it is natural to wonder what that number means for your day-to-day work. The first thing to understand is that exposure measures how many of the tasks that make up your occupation are likely to be automated or assisted by generative AI – it does not predict job loss. In practice, this means many of your routine, rule‑based activities may become faster or even automatic, but the role itself will probably shift rather than vanish. For example, a legal associate might use AI to draft standard contracts or summarise case law, but still need human judgment to interpret nuance, advise clients and argue in court. The key practical takeaway is to identify which parts of your work are most and least exposed, and to invest time in the parts that remain firmly in human hands.

Generative AI excels at tasks that involve pattern recognition, language generation, data synthesis and simple reasoning – for instance, writing first drafts of reports, translating documents, writing code for common functions, or analysing structured data. It struggles with tasks that require physical dexterity, genuine emotional understanding, complex multi‑step problem‑solving in unpredictable environments, ethical judgment, and accountability. A job in IT might involve both writing standard code (highly complementable) and negotiating system architecture with stakeholders (less easily automated). A finance role might involve preparing financial statements (highly complementable) but also advising on investment strategy under uncertainty (less so). The frontier is moving quickly, but the pattern is consistent: AI works best when the task is narrow, well‑defined and rich in data. Human skill is most valuable when ambiguity, context and human interaction are central.

What can you do to adapt? The most practical step is to strengthen the skills that AI currently lacks. These include critical thinking that goes beyond pattern matching, creativity in ill‑defined problems, emotional and social intelligence, negotiation, and the ability to take responsibility for outcomes. Also valuable is the ability to work with AI itself – knowing how to prompt effectively, how to verify outputs, and how to integrate them into broader workflows. Many employers already value a combination of domain expertise and digital literacy, and this will only increase. Additionally, since exposure scores are averages across entire occupations, your personal exposure may be lower if your role focuses on strategy, client relationships or innovation. The data suggest that occupations in Creative, Media and Personal Services have a lower average exposure (4.5), but cross‑sector roles that blend technical and interpersonal tasks often offer more resilience.

It is important to be candid about what the exposure scores do not tell you. First, they measure the exposure of tasks, not entire jobs. A single occupation contains many tasks, some high‑exposure and some low; the overall score is a weighted average. Second, the data are built from research indices that analyse which tasks are susceptible to AI, and those indices have been mapped onto official occupation classifications such as ISCO or national SOC codes. This mapping is itself an approximation, and some of it relies on model‑assisted techniques – meaning there is inherent uncertainty. Third, the scores reflect technical potential, not real‑world adoption. How quickly employers adopt these tools depends on costs, regulation, organisational culture, and the availability of alternatives. The data also cannot predict new job creation, which has historically accompanied technological shifts. Finally, the global averages disguise large variation: the average exposure for the United Kingdom is 5.6, but that covers 379 occupations and 30.9 million workers – local labour market conditions vary enormously.

None of this means you should ignore the scores. They offer a grounded, research‑based starting point for thinking about how your work might change. But they are a guide, not a prophecy. The most resilient workers across history have been those who combine strong core skills with the ability to learn new tools and adapt to shifting contexts. Whether you are in a high‑exposure field like IT or a lower‑exposure one like Trades and Construction (global average 1.9), the fundamental advice is similar: stay curious, keep upgrading your human‑centred skills, and view AI as a tool to be mastered rather than a threat to be feared. The data we present – covering 42 countries, 19,305 occupations, and over 2.8 billion workers – is the most comprehensive picture available, but it is a picture of tasks, not of people. Your career is shaped by far more than an exposure score.