Research check · 19 October 2026
The AI-skill pay gap vanishes within the same firm
A Swedish matched-hire study finds that people hired for AI development or implementation skills earned about 5–6% more than recent hires in comparable occupations. That gap disappears when the comparison is within the same firm. It is not evidence that learning to use a chatbot raises your pay.
What the paper measured
Mark Hellsten links Swedish job advertisements to external hires and register earnings from 2014 through 2023. AI jobs are vacancies requiring skills to develop or implement AI, not ordinary use of generative AI tools. The pay analysis compares recent hires and adjusts for occupation and worker characteristics; adding firm effects removes the remaining average gap.
The paper also reports a much larger raw pay difference before controls. That difference is not a return to AI skills: AI hires are concentrated in better-paying jobs and firms. The author describes a distinct, more highly paid doctoral group, which should not be folded into the typical AI hire.
What it cannot answer
This is an observational comparison, not a training experiment. It excludes workers already employed in the firm and uses a restricted private-sector sample. Matching an advert to a hire can be imperfect, and the study does not test current chatbot skills, layoffs or the effect of taking a course. The 2014–2023 window also predates much of the latest workplace adoption.
A worker deciding what to learn needs evidence that training changes their own hiring or pay, not just evidence that AI specialists tend to work at higher-paying firms. The broader AI job-loss and hiring answer remains mixed and narrow.
One study, not two
The 16 October 2026 paper says it previously circulated as Returns to AI skills. We count this as one study lineage, not an independent replication. The institute's record and full PDF identify the version checked.