AI job loss predictions and how far to trust them

01Three different things get called a prediction

Before comparing numbers, check which of these each one is.

• Exposure: the share of work AI could help with or do. It says nothing about whether jobs disappear.
• Displacement: jobs actually lost or not filled because of AI.
• Net change: jobs lost minus jobs created, which is the figure that decides employment overall.

02The main published estimates, in one line each

These are the figures most often quoted, each with the method behind it.

• Goldman Sachs Research, April 2023. Generative AI could expose the equivalent of 300 million full-time jobs to automation worldwide. Method: the task content of more than 900 occupations. The authors add that most jobs are only partly exposed and more likely to be complemented than replaced.
• International Monetary Fund staff, January 2024. Nearly 40 percent of jobs worldwide are exposed, about 60 percent in advanced economies. About half of affected jobs there may gain productivity. In the other half AI could take over key tasks, which could lower wages or hiring.
• World Economic Forum, January 2025. A survey of more than 1,000 large employers. They expect structural change to create jobs equal to 14 percent of today's employment and displace 8 percent by 2030, a net gain of 7 percent. AI is one driver among several.
• International Labour Organization, May 2025. One in four workers is in an occupation with some exposure. Method: task-level data, expert input and AI predictions across nearly 30,000 tasks. Most jobs are expected to be transformed rather than made redundant.
• US Bureau of Labor Statistics, August 2026. Total employment is projected to grow 3.5 percent from 2025 to 2035. Method: historical trends, assuming technology advances at past speeds. BLS says AI is moving fast and its uncertainty is very high.

03Why one question gave 47 percent and 9 percent

Earlier automation studies show how method changes the answer. A 2013 Oxford paper treated whole occupations as automatable and estimated that about 47 percent of US employment was at high risk over an unspecified period, perhaps a decade or two.

In 2016 researchers at a German economic institute, writing for the OECD, looked at the tasks inside each job and found that about 9 percent of jobs across 21 OECD countries were automatable. They argued that occupation-based studies overstate risk, because high-risk occupations still contain many tasks that are hard to automate, and that a share of jobs at risk must not be read as expected job losses. Both papers pre-date generative AI, so they illustrate method, not today's tools.

04What has been observed so far

Observed data is firmer than forecasts, but it covers only the past. The Yale Budget Lab's tracker, updated on 14 October 2026, finds no clear AI-driven shift in the mix of jobs. A Stanford payroll study, revised on 12 August 2026, finds that workers aged 22 to 25 in AI-exposed occupations have employment 19 percent below their less-exposed peers, but the authors call this descriptive, not causal. The two look at different groups, so they do not necessarily contradict each other.

05A checklist for judging any new number

Use these five questions on any figure you meet, including the ones above.

• Is it exposure, displacement or net change?
• Who made it, and do they call it a forecast? Several sources above say their figures describe exposure or conditional projections, not predicted job losses.
• What is the method: a task model, an employer survey, past trends or observed data?
• When was it made? Tools change fast, so older estimates may describe older AI.
• Does it show a range or a stated uncertainty? One clean number hides doubt.