Key Details
The paper tracks new hiring demand, not job losses, using online vacancies posted between November 2021 and April 2026.
Finding | What the Paper Reports |
|---|---|
Evidence base | About 700 million job postings across 110 economies |
Hiring in AI-exposed occupations | Postings fell by about 6% on average relative to less-exposed occupations in the preferred model; the estimated gap approached 10% late in the study period |
AI capabilities | The relative hiring decline largely stabilised despite further improvements in frontier AI |
Different kinds of exposed work | Hiring fell more in occupations whose tasks are easier to substitute with AI than in those where AI can complement human work |
Changing skills | Demand for 51 emerging AI skills rose rapidly, particularly in AI-exposed occupations |
Hiring Weakened Most Where AI Could Substitute for Tasks
The paper, How Is Labor Demand Changing as AI Improves? New Global Evidence, compares hiring trends in more- and less-exposed occupations within the same economy, industry and month. The gap began to emerge after the November 2022 launch of ChatGPT and widened through 2023 and 2024. It did not keep widening at the same paceas newer AI systems became capable of more complex work.
Exposure is not the whole story. Occupations combining high AI exposure with less scope for human–AI collaboration experienced a larger estimated decline in postings than exposed occupations involving more human judgement, responsibility or communication. The researchers found no consistent evidence that junior vacancies were hit harder than senior ones, or of a change in advertised wages in the subset with wage data.
Employers Are Also Changing What They Hire For
Vacancies increasingly ask for skills such as using AI tools, building AI agents and connecting AI systems to an organisation’s own information. Firms advertising for these skills tended to show faster growth in job postings, including for AI-exposed occupations. That is an association, not proof that adopting AI caused those firms to hire more.
The result is a labour market moving in two directions at once: weaker demand for some existing tasks and growing demand for workers who can apply AI. The paper’s evidence suggests the hiring reduction has so far been the stronger of the two effects in exposed occupations.
Policy Relevance
The paper does not provide a separate estimate for India, and online vacancies chiefly capture formal-sector hiring. They do point to a useful distinction for Indian labour and skills policy: monitoring whether vacancies are disappearing in particular occupations is different from tracking how the skills required within those occupations are changing. Training tied to demonstrated employer demand may be more useful than treating every AI-exposed job as either destined to disappear or certain to grow.
Follow the Full Working Paper Here: How Is Labor Demand Changing as AI Improves? New Global Evidence