Predictions that generative artificial intelligence would quickly remove large numbers of jobs have not been reflected in broad employment data to date. Available research instead points to a more uneven pattern: employment remains broadly stable in many exposed occupations, while job tasks, hiring expectations, and opportunities for younger workers are changing.
What the Data Shows So Far
A recent analysis discussed by The Guardian found no evidence of major economy-wide displacement since the public release of ChatGPT in 2022. The unemployment rate for the 20 percent of workers assessed as most exposed to AI rose by 0.77 percentage points. For the 20 percent least exposed, the increase was slightly larger, at 0.85 percentage points. These figures indicate that workers in highly exposed occupations have not, so far, experienced a larger aggregate unemployment increase than workers in less exposed occupations.
The findings do not establish that AI has had no effect on employment. They show that a large-scale employment decline has not yet appeared in the available data. Economists cited in the report note that technological adoption often requires firms to alter systems, job roles, and recruitment practices before measurable labour market effects become visible.
The International Labour Organization reaches a similar conclusion at the global level. Its 2025 study reports that one in four workers is employed in an occupation with some exposure to generative AI. However, 3.3 percent of global employment is classified in the highest exposure category. The study identifies job transformation, rather than the elimination of whole occupations, as the most likely overall effect because most jobs involve tasks that still require human input.
Where Effects are Emerging
While aggregate figures remain relatively stable, research indicates that the impact may differ according to age, experience, and occupation. A Stanford Digital Economy Lab working paper examined high frequency administrative payroll data in the United States. It found a 16 percent relative decline in employment among workers aged 22 to 25 in the occupations most exposed to AI after the technology became widely adopted. Employment among more experienced workers in the same occupations, as well as workers in less exposed fields, remained stable or continued to grow.
The paper reports that employment declines were concentrated in occupations where AI is more likely to automate tasks rather than support workers in completing them. It also finds that the adjustment occurred mainly through employment rather than changes in pay. The researchers describe the findings as early evidence consistent with a disproportionate impact on entry level workers, while recognising that labour market outcomes can be shaped by several factors.
This distinction is relevant because entry level work often includes tasks through which new employees develop professional experience. If employers reduce recruitment for roles involving routine information processing, writing, coding, or customer support, the structure of career entry may change even when overall employment totals do not decline.
Skills and Job Design
The available evidence also indicates that employers are placing greater emphasis on AI related capabilities. A 2026 ZipRecruiter survey, cited by The Guardian, found that 74 percent of employers viewed AI skills as either a strong advantage or a requirement for some positions. Thirteen percent required these skills across their organisations, while half expected candidates to have practical or advanced AI capability when beginning work.
The reported shift concerns more than explicit references to AI in job advertisements. Employers may also raise expectations for speed, output quality, and independent problem solving. At the same time, firms have both expanded and reduced staffing within functions such as technology, customer support, and business operations, indicating that workforce needs are still being defined.
What Remains Uncertain
Research on AI and employment remains limited by the recent adoption of the technology and by measurement constraints. Government employment statistics do not isolate the effect of a specific technology, while private sector data can provide more timely evidence but may not represent the entire labour market.
Current research therefore does not support a conclusion that AI has already caused mass job destruction. It does, however, document changes in task automation, employer skill requirements, and employment prospects within particular groups and occupations.
