Three Years of ChatGPT, and Only 3% of U.S. Workers Have Lost a Job to AI

Four years after ChatGPT’s debut in November 2022 — through GPT-4, GPT-5, Claude, Gemini, and the entire agentic AI wave that followed — the most feared labor-market disruption in modern history still hasn’t shown up in American workers’ actual lives. That is the central finding of a new YouGov survey of 1,250 employed U.S. adults, fielded July 30 through August 4, 2026, and published this week by the sociologist who commissioned it.

The numbers are striking for how un-striking they are. Only about 3% of workers said they had lost a job due to AI since 2023. Roughly 6% landed a new job that did not exist before AI — think AI data labelers or AI compliance specialists. And approximately 9% said they have earned a promotion or advancement related to AI, such as a boost tied to their AI skills. Flip those numbers around and the real story emerges: roughly 95% of surveyed workers said no, they have not lost a job to AI; 90% said no, they haven’t landed a newly created AI job; and 88% said no AI-related promotion has come their way. Between 3% and 4% were “not sure” about each of the three.

For a technology that Goldman Sachs once estimated could automate away 300 million full-time jobs, and that polling consistently shows most Americans fear, the measured reality so far is a labor market that AI has barely dented — at least if you ask the workers themselves.

Who ran the survey, and why the details matter

The survey was commissioned by Jeffrey C. Dixon, a professor of sociology at the College of the Holy Cross in Massachusetts, and conducted online by YouGov with a sample of employed U.S. adults designed to reflect the nation’s working population by age, gender, race, education, and other characteristics. The fieldwork ran July 30–Aug. 4, 2026 — that is, after the official unemployment rate had fallen to 4.1%, but during a period when the workforce was shrinking and AI was reportedly a leading reason for many layoffs.

Dixon’s 25-question instrument went beyond the usual “will AI take your job” polling by asking workers to report their own lived experiences across three specific job changes since 2023: losing a job to AI automation; landing a job that didn’t previously exist before AI adoption; and receiving a promotion or advancement believed to be AI-related. It is the basis of a study in progress that has not yet been peer reviewed — a caveat worth holding onto — but it is one of the few U.S. data sources that measures AI’s labor-market impact from the worker’s side of the desk rather than from economic modeling or vendor-sponsored analysis.

The framing matters because the two most-cited flavors of AI jobs evidence have opposite biases. Economist projections extrapolate from task-exposure models and tend to produce dramatic headline numbers. Corporate case studies, often produced by AI companies themselves, tend to highlight efficiency gains and retraining success stories. Asking 1,250 randomly sampled workers whether AI actually changed their employment trajectory is a refreshingly direct approach — and it produced a result that surprised even the researcher.

The perception gap is the real story

Dixon describes himself as an “AI pessimist,” and by his own account the finding that so few workers reported AI-driven job losses raised his eyebrows. The surprise is sharpest when you set the survey against the broader polling landscape.

An Economist/YouGov poll from February 2026 found that most Americans say AI will reduce the number of jobs in the United States. A Quinnipiac poll from March 2026 put the figure at 70% of Americans who believe AI is likely to decrease job opportunities. Yet when workers are asked what has actually happened to them — not what will happen, not what could happen — the displacement rate sits at 3%, and the AI-job-creation rate (6%) is twice as high as the displacement rate. Add the 9% who report AI-linked promotions, and AI’s measured effect on individual careers currently skews more toward advancement than elimination.

That 3% is not zero, and it represents real people. But it is a very different number from the mass displacement narrative that dominates headlines, and it aligns with what other researchers in the U.S. and elsewhere have found: AI is not yet having widespread, measurable effects on the labor market as a whole.

The caveats are as informative as the headlines

The survey comes with honest limitations, several of which Dixon himself flags. Job changes of any kind are rare in the U.S. economy, so small percentages of AI-related job changes must be read against that quiet baseline. Workers cannot always know for certain whether AI deserves blame or credit for a career turn — a layoff attributed to “restructuring” may have AI underneath it, and a promotion attributed to “AI skills” might reflect broader trends. The sample is an opt-in online panel rather than a true random sample, so participants’ experiences may differ from those of the average worker. And crucially, none of the respondents were jobless — meaning people who lost a job to AI and stayed unemployed, or left the labor force entirely, are structurally absent from the data. Given that the workforce was shrinking during the survey window, that omission is not trivial.

There is also a timing argument. Agentic AI — systems that can act autonomously and automate multi-step work rather than just answer questions — only began reaching mainstream business deployment in earnest in 2025 and 2026. If the disruptive wave is coming, a survey fielded in mid-2026 may be taking attendance before the storm. Dixon notes that agentic AI may render some current AI expertise less valuable in the future, which cuts both ways: today’s 9% promotion beneficiaries could be tomorrow’s displaced if their skill premium erodes.

Why this matters now

The finding lands at a moment when AI’s labor-market story is being written by extremes. Tech executives promise productivity booms; doomer forecasts predict entry-level white-collar collapse; and a California AI unemployment tracker counts layoffs attributed to AI in real time. Against that noise, a modest survey with modest findings does something valuable: it anchors the debate to the actual reported experience of 1,250 working Americans at the three-year mark of the generative AI era.

The honest summary is that AI’s employment impact so far is neither the catastrophe pessimists predicted nor the jobs bonanza optimists promised. It is a slow, uneven remix: a small minority displaced, a slightly larger minority in genuinely new roles, and a meaningful slice using AI skills as a ladder. Whether that equilibrium holds as agentic systems spread through the economy is the question the next survey — and the next version of this one — will have to answer.