How (un)Stable Are LLM Occupational Exposure Scores? Evidence from Multi-Model Replication -- by Michelle Yin, Hoa Vu, Claudia Persico
April 27, 2026
•NBER_Working_Papers•Strong impact[key points]
Key points
- •Large language models (LLMs) are used to assess occupational exposure to AI.
- •LLM-based exposure measures are highly variable across different models.
- •This instability affects empirical conclusions about the impact of AI on the labor market.