Which jobs are most at risk from AI?

The core measure on this site is a 0 to 10 AI-exposure score for each occupation. This score reflects the degree to which generative AI can already perform or assist with the specific day-to-day tasks that make up a given job. It is based on a detailed analysis of task descriptions from official occupational databases, matched against the current capabilities of generative AI systems. A score of 0 means no current tasks can be meaningfully performed or assisted by generative AI, while a score of 10 indicates that essentially all of the occupation's tasks can be done or significantly aided by these tools. It is crucial to understand that this is a measure of task exposure, not a prediction of job loss or automation. A high score does not mean the job will disappear; rather, it flags where AI is most likely to reshape the work involved, often by taking over routine or pattern-based tasks, potentially changing the skills that are most valued. A low score does not guarantee safety from change, but suggests that the core tasks are less susceptible to current AI capabilities. Exposure is about the potential for task transformation, not the inevitable replacement of workers.

Even the most advanced AI still struggles with tasks requiring physical dexterity, complex social interaction, nuanced judgment, or deep domain expertise in contexts with high variability. Therefore, occupations that are heavy on manual labour, face-to-face care, or high-stakes improvisation tend to score lower. Conversely, jobs that involve processing structured information, generating text or images, analysing data, or following clear procedural steps tend to score higher. This does not make those jobs 'at risk' in a simple sense. Instead, it suggests that workers in high-exposure roles may need to complement their existing skills with abilities that AI cannot easily replicate, such as strategic thinking, emotional intelligence, or creative problem-solving beyond pattern matching. The exposure score is a starting point for conversations about skills and training, not a verdict on any individual's employment future.

The data covers 13 countries, with a combined total of 6,675 occupations and over half a billion workers. To make it meaningful across such a large and diverse set, all scores are ranked on a single global percentile scale. This means that a score of, say, 6.0 means the same level of task exposure whether you look at an occupation in Germany or in Spain. The global employment-weighted average exposure across all countries is 5.2, but this hides significant variation. For example, Germany has the highest average at 5.8, while Japan and Spain share the lowest at 4.8. Each country's average gives a sense of how its occupational structure and AI task alignment compare with the global benchmark. This international comparability is essential for understanding which industries and roles are likely to be most reshaped by generative AI, regardless of national differences.

When you explore the data visually, you will encounter an interactive treemap. Each tile represents an occupation, and you can interpret it using two key dimensions. First, the area of each tile is proportional to the size of its workforce: larger tiles mean more people work in that occupation, reflecting its importance in the labour market. Second, the colour of the tile indicates the occupation's AI-exposure score, using a gradient from green (low exposure) to red (high exposure). This allows you to quickly see which large occupational groups are either relatively shielded from or heavily exposed to AI influence. For instance, trades and construction show the lowest industry average at 2.3, while IT and digital is the highest at 8.7. The treemap is not a map of job losses; it is a map of where task transformation is most likely to occur, and how many workers might be affected.

The highest-exposure industries globally include IT and digital (8.7), business, finance and legal (8.2), and creative, media and personal services (6.2). These are areas where generative AI can already draft reports, write code, design visuals, analyse contracts, or create marketing copy. At the other end, the lowest-exposure industries are healthcare and care (4.5), transport, logistics and mining (4.3), hospitality, retail and tourism (4.3), agriculture and environment (3.2), and trades and construction (2.3). These involve hands-on work, personal interaction, or unpredictable physical environments that current AI handles poorly. However, even within a low-exposure industry, specific occupations can have higher scores: for example, medical transcriptionists score high on exposure, while surgeons do not. Always look at the individual occupation score rather than just the industry average.

Finally, it is essential to reiterate that the AI-exposure score is a tool for awareness, not a prediction about your own job. An occupation's score tells you about the potential for AI to alter tasks, but it cannot capture the unique context of your workplace, your specific role, or your adaptability. Two people in the same occupation may be affected very differently depending on their employer, their specialisation, and how they combine AI tools with their human skills. The score is a starting point for thinking about complementary skills, such as data interpretation, ethical oversight, and human-centred design, which become more valuable as AI takes on routine tasks. Use this site to explore where your occupation sits, but treat the score as a guide for reflection and upskilling, not as a deterministic forecast.

Among the 13 countries studied, generative AI exposure varies noticeably, with Germany recording the highest employment-weighted average score of 5.8 on the zero-to-ten scale, closely followed by Australia at 5.7 and the United Kingdom at 5.6. Canada and New Zealand also sit above the global average of 5.2, with scores of 5.5 and 5.4. Ireland, the Netherlands, and the United States all tie at 5.3, while South Korea matches the global average at 5.2. At the lower end, France and Italy each score 4.9, and Japan and Spain are the lowest at 4.8. These differences may appear modest, but they reflect each country's distinctive mix of industries and occupations.

The variation across economies stems largely from the composition of their labour markets. Countries with higher average exposure tend to have larger shares of employment in industries that involve processing information, using digital tools, or performing routine cognitive tasks—activities where generative AI can directly assist or automate parts of a job. For instance, the IT and digital sector sits at the very top of the industry exposure range, with a global score of 8.7, followed by business, finance and legal at 8.2. These industries rely heavily on written communication, data analysis, coding, and documentation, all of which are highly susceptible to generative AI capabilities. Germany, Australia, and the United Kingdom, which rank highest in country-level exposure, all have substantial employment in these office- and information-heavy fields.

At the other end of the spectrum, the lowest-exposure industries are those grounded in physical, manual, or in-person work. Trades and construction has the lowest score at 2.3, followed by agriculture and environment at 3.2. Hospitality, retail and tourism, and transport, logistics and mining both score 4.3, while healthcare and care sits at 4.5. These sectors require hands-on manipulation of objects, direct patient or customer interaction, or operation of machinery in unpredictable physical environments—tasks that generative AI, as a software technology, cannot directly perform. The lower scores for Japan (4.8) and Spain (4.8) are consistent with their relatively larger employment shares in manufacturing, agriculture, tourism, and care work.

It is important to note that no country has an average exposure higher than 5.8 out of 10. Even in the most exposed economies, almost half of occupational tasks, on average, are not exposed to generative AI. The figures show a global spread of just one point between the highest and lowest countries, from 5.8 to 4.8, suggesting that the distribution of exposure is relatively compressed across developed economies. The industry scores, however, span a much wider range—from 8.7 down to 2.3—indicating that the type of work one does matters far more for exposure than the country in which one works.

The differences across countries can be understood as the weighted sum of these industry effects. A country like Germany, with a strong presence in IT, business services, and advanced manufacturing, inherits a higher aggregate exposure. Conversely, a country like Spain has larger shares of employment in hospitality, retail, agriculture, and construction, which pull its average downward. The data also show that the United States, despite its large technology sector, has an overall exposure score of 5.3—the same as Ireland, the Netherlands, and South Korea—because its vast and diverse economy also includes many low-exposure industries such as healthcare, transport, and hospitality. The same logic applies to Japan, where an aging society and reliance on manufacturing and care work produce a lower average.

The global employment-weighted average of 5.2, derived from over half a billion workers across 6,675 occupations, serves as a benchmark. France and Italy fall just below this mark, while Germany and Australia sit about half a point above. These differences, while real, are not dramatic. What the data reveal most clearly is that generative AI's potential impact on tasks is concentrated in a subset of industries—chiefly those centred on information processing and digital creation—and that national exposure levels are largely a reflection of how many workers are engaged in such work. Economies with a heavier tilt toward physical, manual, or in-person activities will naturally score lower, simply because generative AI, as currently conceived, cannot wield a hammer, tend a field, or provide bedside care.

If you work in an occupation that scores highly on our exposure index, the first thing to understand is that this score measures how strongly the tasks in your role overlap with what generative AI can already do, not how likely you are to lose your job. A high exposure score means many of your routine, language-based or data-processing tasks can be automated or augmented by current models. For example, drafting reports, summarising documents, translating text, generating code snippets, or analysing structured data are all areas where generative AI is already competent. But that competence is narrow. The same models struggle badly with tasks requiring physical dexterity, face-to-face interpersonal negotiation, nuanced empathy, creative originality that goes beyond pattern matching, and any decision-making that depends on real-time sensory input from the physical world. So if you are a solicitor, your high exposure score does not mean the end of your profession. It means that the parts of your job involving document review and contract drafting may be handled by an AI assistant, while client counselling, courtroom advocacy, and strategic legal judgement remain firmly in human hands. The practical takeaway is not to panic, but to identify which of your daily tasks are the most rule-based and language-heavy, and then deliberately shift your professional development toward the complementary skills that AI struggles to replicate: critical thinking, ethical judgement, complex communication, cross-domain problem solving, and the ability to manage and verify AI outputs.

Building complementary skills means learning how to work with these tools rather than against them. A financial analyst whose exposure score is high because much of the job involves spreadsheet manipulation and report writing can invest time in learning to prompt and interpret generative models, while also deepening their understanding of business strategy, regulatory contexts, and client relationships. A graphic designer whose exposure is driven by generative image tools can focus on art direction, brand strategy, and client liaison. A translator can concentrate on localisation, cultural nuance, and post-editing. Across all high-exposure roles, the most defensive skill is the ability to evaluate AI output critically. Generative models produce plausible but often inaccurate or biased content. A worker who can quickly spot errors, understand the model's limitations, and refine its output will be far more valuable than one who simply accepts what the machine produces. Similarly, skills in data literacy, project management, and interdisciplinary collaboration become more, not less, important when routine tasks are automated. The data show that even in the highest-exposure industry, IT and Digital, which scores 8.7 out of 10, there are still many tasks that remain firmly in human territory: system architecture design, ethical oversight, client negotiation, and creative innovation. The message is to double down on the human elements of your job, not to abandon the technical ones.

However, it is crucial to be candid about the limitations of these exposure scores. The index does not measure the risk of job displacement, nor does it predict how quickly employers will adopt generative AI, whether they will choose to automate or augment, or what new jobs might emerge. It is built from a systematic mapping of research about generative AI capabilities onto the detailed task descriptions in official occupational classifications. Those classifications, such as the Standard Occupational Classification used in the UK, were not designed with AI in mind, so the mapping is necessarily imperfect. Some of these mappings were assisted by machine learning models that read and interpret task descriptions, which can introduce their own errors. The scores are about tasks, not whole jobs, and many roles bundle together tasks with very different exposure levels. A single occupation can include both highly automatable tasks and ones that are completely resistant to automation. Therefore, a high score for an occupation tells you that a significant portion of its component tasks are exposed, but it does not tell you how those tasks are combined in practice or how employers will reconfigure work.

Additionally, the index is based on current generative AI capabilities as assessed by a limited set of research sources. It cannot account for future breakthroughs or for regulatory decisions that might constrain automation in specific sectors. It also cannot measure the speed of adoption, which depends on cost, organisational culture, legal barriers, and social acceptance. The data covers 13 countries, with a total of 6,675 occupations and over 500 million workers, but the global employment-weighted average exposure of 5.2 out of 10 conceals enormous variation between and within countries. For instance, Germany's average exposure score of 5.8 is a full point higher than Japan's 4.8, reflecting differences in industry mix and occupational structure. Even within a country, the variation between industries is dramatic: in the US, the average exposure in IT and Digital is 8.7, while in Trades and Construction it is 2.3. This means that a high exposure score should not be interpreted as a verdict on your personal career prospects, but rather as a signal to pay attention to how your specific tasks are changing. The best way to use this data is as a conversation starter, not a prophecy. Look at the task-level details for your occupation, identify the areas where generative AI is strongest, and then deliberately invest in the skills that complement those capabilities. The future of work will involve human-machine collaboration, not replacement, but only for those who adapt.

Finally, remember that the scores are snapshot estimates, not certainties. They are built from research indices that are themselves evolving as generative AI improves. The mapping onto official occupation classifications required simplification, and some judgments were made about which tasks are exposed, which can be contested. The data does not capture the creative ways in which workers and employers can redesign jobs to make the most of human strengths. New job creation is entirely outside the model's scope. A high-exposure occupation today may become a low-exposure one tomorrow if tasks are redefined or if regulation limits automation. Conversely, roles that currently seem safe might become more exposed as technology advances. The practical, non-alarmist stance is to treat these scores as a useful but incomplete guide. They give you a map of where the ground is shifting, but you are the one who decides which path to take. Use the data to start a conversation with your employer, your professional network, or your trade union about reskilling and job redesign. No index can tell you your future, but it can help you ask better questions about it.