Studies of AI and work measure tasks inside jobs, and they find that far more jobs are touched by AI than are replaced by it. Statistics Canada estimated that 60 percent of Canadian workers in 2021 held jobs highly exposed to AI, split almost evenly between jobs likely to be assisted and jobs with less room for assistance, and it warns that exposure does not mean job loss. A U.S. study of language models reached a similar picture: about 80 percent of the workforce could see at least a tenth of their tasks affected.
Key takeaways
- Exposure to AI describes how many of a job's tasks a system could do or assist, and it is not a forecast of layoffs.
- Statistics Canada found 29 percent of workers in highly exposed, highly complementary jobs, 31 percent in highly exposed jobs with lower complementarity, and 40 percent outside highly exposed jobs (May 2021 data).
- In March 2026, generative AI use at work was 53.8 percent in the first group, 45.9 percent in the second, and 14.2 percent in low-exposure jobs.
- A 2023 experiment with 453 professionals found that ChatGPT access cut writing task time by about 40 percent and raised rated quality by about 18 percent.
Why researchers count tasks instead of jobs
An occupation is a bundle of tasks. A nurse charts, assesses, and talks with families; a paralegal drafts, searches, and files. A system that handles one task well can change the shape of the job without removing it, which is why exposure studies score occupations by how closely their tasks match what AI systems can do. Two further ideas separate the studies from headline claims. Exposure can be paired with complementarity, meaning how far the remaining human tasks gain from assistance. And an exposure score is a snapshot of what systems could do at the time of the study, so it ages as the tools change.
Statistics Canada states the limit directly: relatively higher occupational exposure to AI does not necessarily imply a higher risk of job loss. Financial, legal, and institutional factors can delay replacement, and jobs may be reshaped instead.
Canada: the exposure split and who is using the tools
In its September 3, 2024 release on experimental AI exposure estimates, Statistics Canada used a complementarity-adjusted index applied to census occupations. The shares in May 2021 were similar to those in May 2016. Education mattered: 50 percent of workers with a bachelor's degree or higher were in highly exposed, highly complementary jobs, against 13 percent of workers with a high school diploma or less.
A later supplement to the agency's monthly employment survey, released July 30, 2026, asked workers directly about use. Overall, 35.9 percent had used generative AI tools at work in the previous 12 months, and the gap between occupational groups was wide.
| Occupation group (Statistics Canada) | Share of workers, May 2021 | Used generative AI at work, March 2026 |
|---|---|---|
| High exposure, high complementarity | 29 percent | 53.8 percent |
| High exposure, low complementarity | 31 percent | 45.9 percent |
| Low exposure | 40 percent (not highly exposed) | 14.2 percent |
The two columns come from different years and different surveys, so they describe composition and use separately and do not show cause and effect. The agency also notes that the index indicates potential for AI to transform job tasks and does not predict job loss.
United States: tasks that language models could touch
Researchers Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock published the paper "GPTs are GPTs" on arXiv in March 2023 and revised it in August 2023. Their abstract estimates that about 80 percent of the U.S. workforce could have at least 10 percent of work tasks affected by large language models, and about 19 percent of workers could have at least half of their tasks affected. The numbers are potential effects estimated from task descriptions, not observed changes in hiring or pay, and the models studied are from 2023.
What one controlled experiment measured
Exposure scores say little about what happens when someone actually uses a tool, so a controlled experiment fills part of the gap. Shakked Noy and Whitney Zhang, then at MIT, gave 453 college-educated professionals incentivized, occupation-specific writing tasks and randomly gave half of them access to ChatGPT. The study, published in Science in July 2023, reported that time taken fell by about 40 percent and rated output quality rose by about 18 percent. Performance gaps between workers narrowed, and exposed participants were about twice as likely to report using ChatGPT in their real jobs two weeks later. The tasks were short and online, so the result does not show how a full job changes over months.
A checklist for reading the next claim about AI and jobs
- Does the figure describe exposure, use, or actual job changes? These are three different measurements.
- What year is the data from, and which AI tools did the authors consider?
- Is the sample workers, employers, or occupations scored by researchers?
- Does the source say what exposure does not show, as Statistics Canada does?
- Does the headline number depend on one assumption, such as a threshold at the median?
FAQ
Does high AI exposure mean a job will disappear?
No. Statistics Canada says higher occupational exposure does not necessarily imply a higher risk of job loss. Exposure counts tasks that systems could do or assist.
Which Canadian workers use generative AI most?
In March 2026, workers in highly exposed, highly complementary occupations reported the highest use at 53.8 percent, while workers in low-exposure occupations reported 14.2 percent.
How many U.S. workers could be affected by language models?
The 2023 estimate by Eloundou and colleagues is about 80 percent of workers with at least 10 percent of tasks affected and about 19 percent with at least half. It is a potential, not a measured outcome.
Why do different studies give different numbers?
They define exposure differently, use different years, and look at different groups. Comparing a task-based index, an employer survey, and a worker survey side by side usually mixes incompatible measures.
Related reading
- How AI is changing society for the better
- The role of AI and automation across industries
- The AI revolution and its effect on industries
- AI in data science
More in AI Basics.
Last verified: October 2026
Sources
- Experimental estimates of AI occupational exposure in Canada, The Daily, Statistics Canada, September 3, 2024
- Use of generative artificial intelligence tools among Canadian workers, March 2026, The Daily, Statistics Canada, July 30, 2026
- GPTs are GPTs: An early look at the labor market impact potential of large language models, Eloundou, Manning, Mishkin, and Rock, arXiv, 2023
- Experimental evidence on the productivity effects of generative artificial intelligence, Noy and Zhang, Science, July 2023