Early signs of AI-driven adjustments in Canada’s labour market

AI has the potential to reshape the labour market by automating some tasks for many workers. Even though evidence of a significant impact on the labour market remains limited, AI appears to be making it harder for some Canadians to find work in occupations that are most exposed to it.

Discussions about artificial intelligence (AI) and jobs often focus on whether this technology will replace human workers. These concerns are understandable given how fast AI capabilities have been improving.

But in reality, the effects of AI on jobs and the labour market are more nuanced.

AI is unlikely to cause most jobs to disappear entirely. Instead, it’s more likely to reshape how workers perform their jobs, automating some tasks and simplifying others while still leaving a portion of tasks largely unaffected. The amount of change may hinge on a job’s exposure to AI, which is based on the tasks typically performed in each role.

Our analysis of how AI is affecting jobs suggests that AI has not yet led to broad changes in the overall structure of the labour market. But it is affecting specific tasks within some occupations, and this may be influencing hiring needs. Our analysis shows that job seekers may be finding it more difficult than it was in 2019 to secure employment in occupations that are now the most exposed to AI.

These results are only signals. They’re not definitive about what is happening or will happen, because the future of AI is highly uncertain.

How we measure the exposure of Canadian jobs to artificial intelligence

Every job can be broken down into a series of tasks that workers do. A job’s exposure to AI, therefore, reflects the extent to which AI—based on its current capabilities—could affect those tasks.

This is what is captured in the Generated Index of Occupational Exposure (GENOE). One of the reasons we use GENOE in our analysis is that it incorporates ethical and regulatory considerations. For example, the work of judges is technically highly exposed to AI, but relying solely on AI for legal rulings would be considered unethical—especially if AI reproduces any biases that may exist in the data it is trained on.

We adapt GENOE to Canada’s National Occupational Classification system to produce AI exposure scores based on domestic data. Our analysis begins with 2015, when advances in AI capabilities and adoption started to accelerate.

The results of our analysis reveal that the average AI‑exposure score in Canada in 2025 was 0.29. This suggests that close to one‑third of jobs may undergo substantial changes due to AI integration, given today’s AI capabilities. Exposure is highest in occupations like receptionists and accountants, where most tasks are routine, codifiable and consist of information processing (Table 1). Exposure is lowest in occupations that rely more heavily on judgment, physical interactions or highly specialized human skills—for example, jobs in health care and skilled trades.

Note: AI is artificial intelligence. The occupations listed here are based on Canada’s National Occupational Classification system. We convert estimates of AI exposure for US occupations to estimates for equivalent Canadian occupations. US estimates are from M. Benítez‑Rueda and E. Parrado, “Mirror, Mirror on the Wall: Which Jobs Will AI Replace After All? A New Index of Occupational Exposure,” Inter‑American Development Bank Working Paper Series No. 1624 (2024).
Sources: Inter‑American Development Bank, Employment and Social Development Canada, Statistics Canada and Bank of Canada calculations

Unemployed workers from highly exposed occupations are increasingly struggling to find a job

Current exposure scores capture one aspect of how AI relates to the labour market. Another aspect to examine is how labour market outcomes have evolved over time across workers with different levels of exposure to AI.

For this second part of the story, we use individual‑level data from Statistics Canada’s Labour Force Survey. Rather than focusing on overall employment levels—which often adjust slowly—we zero in on three measures:

• the probability of being unemployed
• the probability that an unemployed person finds a job (i.e., the job finding rate)
• the probability that an employed person becomes unemployed (i.e., the job separation rate)

These measures combine unemployment with underlying flows into and out of employment, and they can respond quickly to changes in labour market conditions. Tracking these measures over time can show us differences in outcomes between more‑ and less‑exposed occupations.

Workers in jobs with high AI exposure already faced elevated unemployment risk in 2015–19. For this risk, the estimated gap between those in jobs with full AI exposure and those in jobs with zero AI exposure was 1.9 percentage points (Table 2). This is an upper‑bound estimate because no workers are observed at these extremes. By 2025, the estimated gap had risen to 2.8 percentage points—almost 1 percentage point greater than in the earlier period. This means that, since 2015–19, workers in more‑exposed occupations have become increasingly more likely to be unemployed relative to those in less‑exposed occupations.

Note: The estimated difference is calculated using a regression of each of the three labour market metrics on occupation‑level AI exposure. We convert estimates of AI exposure for US occupations to estimates for equivalent Canadian occupations. US estimates are from M. Benítez‑Rueda and E. Parrado, “Mirror, Mirror on the Wall: Which Jobs Will AI Replace After All? A New Index of Occupational Exposure,” Inter‑American Development Bank Working Paper Series No. 1624 (2024).
Sources: Inter‑American Development Bank, Employment and Social Development Canada, Statistics Canada and Bank of Canada calculations

What caused most of the increase of this estimated risk? The reason appears to be a decline over time in job finding rates among highly exposed workers relative to those with low AI exposure. This decline reveals that unemployed workers have had more difficulty finding employment in highly exposed jobs. In contrast, the rates at which workers have lost or left a job—the job separation rate—have remained relatively equal between more‑ and less‑exposed occupations.

Together, these results suggest that the post‑2019 rise in the unemployment risk for highly exposed occupations shows up mainly in hiring difficulties rather than more‑frequent layoffs or resignations.

Young workers may face higher risks related to artificial intelligence

Data from the Labour Force Survey also allow us to identify how AI may be affecting workers of different age groups. What we see is that, relative to older workers, a larger share of young workers tend to be in occupations with high AI exposure (Chart 1).

Several occupations that employ many young workers have moderate to high AI exposure, including customer service and sales support. Overall, this suggests that young workers may face higher AI‑related labour market risks than older workers.

What these findings do and do not mean

Our results overall should be seen only as early signals. They do not show that AI alone has led to differences in employment outcomes across occupations since 2019. Many factors affected the labour market over this period, including the COVID‑19 pandemic, a surge in immigration and a reconfiguration of global trade.

What our results do show are patterns about how the labour market may be adjusting to structural changes triggered by AI. These patterns suggest that any adjustments may show up first through slower hiring (a lower job finding rate) rather than more people leaving jobs (a higher job separation rate). The breadth and persistence of these adjustments will likely depend on:

• the pace of AI adoption
• improvements to AI models
• how quickly workers adjust

Find out more