Someone Finally Asked Workers Directly: 3% Say AI Cost Them a Job

Almost everything written about AI and jobs is a forecast. Economists model it, AI companies publish reports about it, and headlines turn both into predictions. Jeffrey C. Dixon, a sociologist at the College of the Holy Cross, tried something less common: he asked workers. He commissioned YouGov to survey 1,250 employed US adults between 30 July and 4 August 2026, and the answers do not match either the doom or the hype.

About 3% said they had lost a job because of AI since 2023. Roughly 6% said they had landed a job that did not exist before AI, such as a data labeler or an AI compliance specialist. Around 9% said they had received a promotion or advancement they credited to AI, most often their own AI skills. For everything else, the “no” answers dominated: 95%, 90% and 88% respectively, with another 3% to 4% unsure in each case. Dixon describes himself as an AI pessimist and says he was surprised by how small the job loss number was. The findings line up with other work, including Yale’s Budget Lab tracker and research from the National Bureau of Economic Research, which have also struggled to find a large AI signal in labour market data so far. His paper is still in progress and has not been peer reviewed.

What’s actually going on here: the honest reading is not “AI does not affect work.” It is that measuring the effect is genuinely hard, and the survey shows why. Americans change jobs rarely to begin with, so small percentages need that context. Nobody can be certain whether AI or a budget cut ended their role. The people surveyed were all currently employed, which by definition leaves out anyone who lost a job and has not found another. And the survey ran in July 2026, when the official unemployment rate was 4.1% but the workforce was shrinking and AI was being named as a leading reason in layoff announcements. The most interesting number here is not the 3%. It is the 9% getting promoted, which suggests that so far AI is reshuffling who moves up more than it is deleting roles. Dixon adds a fair caveat to his own good news: agentic AI, meaning systems that carry out multi step tasks on their own, could make today’s AI skills less special fairly quickly.

What this means for you: if you have been quietly anxious about your job because of what you read, this is a useful piece of ground truth, and the ground is calmer than the coverage. If you are looking for a practical takeaway, the promotion number is the one to act on. Workers who learned to use these tools in their existing role are reporting more upside than workers who chased brand new AI job titles. That is a much lower risk move than retraining for a category that barely exists yet. And keep expectations grounded in both directions: one snapshot survey of employed people in one country cannot tell you what 2028 looks like.

Sources

• I surveyed workers to see if AI had caused job losses and was surprised by the findings (The Conversation)
• ChatGPT said you would lose your jobs right now, it is more like 3% of workers (Fortune)
• Tracking the impact of AI on the labor market (The Budget Lab at Yale)