Generative AI: the first changes in the labour market are affecting young people (blog post 463)
A · Strong evidence Banque de France economists (with Bergeaud and Bertheau) use comprehensive French administrative hiring data and firm surveys, and state plainly what the data cannot separate.
Summary based on the full text·Original: about 6 minutes to read
In short
Banque de France blog post No. 463 (8 October 2026) by Marco G. Palladino, Antonin Bergeaud, Antoine Bertheau, Simon Bunel and Oscar Degos. Since late 2022, permanent-contract (CDI) hiring of young people in France has fallen more in occupations most exposed to generative AI. Terminations fell too, so turnover slowed. Firms reporting intensive AI use hired fewer young people, but their total employment did not diverge significantly. The first adjustment shows in hires and separations, not headcount.
Key numbers
- ~28% vs ~8%fall in permanent-contract hires of 21-29 year-olds, March 2026 vs November 2022, most vs least AI-exposed quintile
- ~20 percentage pointsgap between the two hiring trajectories for 21-29 year-olds
- ~23% vs ~6%fall in permanent-contract terminations for 21-29 year-olds, most vs least exposed quintile
- ~6,000 / ~600firms tracked in the survey-admin pairing, and firms reporting intensive AI use in early 2026
As reported by the source.
English·Web page·opens banque-france.fr
What it reports
- 21-29 year-olds: CDI hires in the most exposed quintile were about 28% below their November 2022 level in March 2026, against 8% in the least exposed; a gap of about 20 points. The gap narrows for 30-49 year-olds and almost disappears from age 50.
- Terminations for 21-29 year-olds: about 23% below November 2022 in the most exposed quintile vs 6% in the least. Hires fell more than terminations, so workforce renewal slowed.
- The authors' exposure index correlates highly with observed AI use by occupation from Google's AI & Economy ATLAS.
- Of about 6,000 firms matched to Banque de France business surveys, nearly 600 reported intensive AI use in early 2026. These firms were already hiring more in exposed occupations before ChatGPT.
- Youth hiring trends were less favourable in intensive-use firms, with no statistically significant divergence in total employment.
- Foreign evidence is mixed: a relative youth decline in the US, none detected in Norway, task reorganisation without hours or earnings effects in Denmark.
How they got it
A language model scored more than 1,500 occupations in France Travail's ROME directory on how far AI can perform or assist each skill, adjusting for tacit expertise, physical presence and human interaction. The index was joined to administrative workforce-movement data (DMMO/MMO, January 2019 to March 2026) and to firm surveys. Trends are seasonally adjusted and smoothed; firm comparisons use Poisson models with firm and sector-by-quarter effects.
Limits
- Monetary tightening, the post-Covid tech correction and telework hit the same occupations; the authors do not separate AI from these forces.
- AI use is measured only in early 2026, after the hiring gaps had opened; selection of adopting firms cannot be ruled out.
- The exposure index measures technical potential, scored by a language model.
- A short post; fuller results are deferred to a working paper.
Why it matters for companies
The adjustment runs through fewer junior hires and slower turnover, which headcount totals will not show. Firms tracking AI impact only through layoffs or total staff will miss it. The longer-run cost the authors flag is lost learning-by-doing if entry-level tasks are automated, which raises a pipeline question for any firm cutting junior intake in exposed roles.
Summary based on the full text. Read the original for the full argument and data.Read the original on banque-france.fr
- Tags
- Labour market