The range of estimates for how many jobs AI will displace is wide enough to be confusing: 14% per OECD task-level analysis, 47% per the original Oxford Martin School paper, and 300 million full-time equivalent jobs globally per Goldman Sachs. All three figures are credible. All three are measuring different things.
The key distinction is between jobs that are technically automatable and jobs that will actually be automated in a given timeframe — and between tasks AI can do and employers who choose to restructure around that capability. Understanding what each estimate actually counts is what makes the data usable.
What the research says
92M
jobs displaced by 2030
WEF Future of Jobs 2025
170M
new roles created by 2030
WEF Future of Jobs 2025
41%
of employers plan AI-driven headcount reductions
WEF 2025
55K
job cuts explicitly attributed to AI in 2024
Challenger, Gray and Christmas
The difference comes down to unit of analysis. Oxford's Frey and Osborne paper from 2013 rated entire occupations as automatable or not, leading to a high aggregate figure. The OECD reanalysed the data at the task level, finding that most occupations contain a mix of automatable and non-automatable tasks. Even in high-risk occupations, some tasks require physical dexterity, social judgment, or creative problem-solving that is not currently automatable. Using task-level data dropped the estimated share of highly automatable jobs to 14%.
Goldman Sachs's estimate of 300 million globally affected refers to full-time equivalent jobs where at least 50% of tasks are exposed to AI automation. This is a broader definition than the OECD's 'highly automatable' threshold and a different unit than the WEF's displacement projection. Goldman Sachs also projected that generative AI could raise global GDP by 7% over 10 years, suggesting a net positive economic outcome even as significant displacement occurs. The 300 million figure represents exposure, not inevitable unemployment.
None of them, taken alone. Aggregate percentages tell you about the labour market as a whole, not about your specific role. What matters for your own assessment is the task composition of your actual job, the AI adoption velocity in your specific industry, and how quickly employers in your sector are restructuring their operations around AI. An occupation-level risk percentage is a starting point, not an answer.
Yes, documented displacement has begun in specific categories. Challenger, Gray and Christmas tracked 55,000 job cuts explicitly attributed to AI in 2024 in the US alone. Klarna's announcement that AI handled the work of 700 agents, Duolingo reducing contractor reliance by 50%, and IBM's decision to pause hiring for 7,800 roles it expected AI to fill are documented examples. The current displacement rate is lower than the most pessimistic projections, but it is accelerating.
The WEF figure of 41% of employers planning AI-driven headcount reductions is consistent with the range of academic estimates because it measures employer intent rather than total job exposure. Even if only 14-47% of jobs are technically automatable, employers do not need to eliminate those roles entirely to reduce headcount. They can maintain output with fewer workers by using AI to increase the productivity of those who remain. That productivity effect is what the 41% figure captures.
Free Assessment
Free Fossil Score assessment. No account required.
Get My Fossil Score