Original title: Biases-Informed Job Search Guidance: Characterization, Implications, and Targeting Support
Bruno Crépon, Aurélien Frot, Christophe Gaillac · August 2026
When people look for work, their guesses about finding a job are often biased. Researchers wanted to know what these biased expectations reveal and how to use them to guide job support. They studied survey and administrative data on French job seekers using computer models and machine learning. They found that biased overall guesses summarize mistakes about two key basics: how often job offers arrive and what wages are offered. These underlying errors vary across people but link together, making their effects worse. Correcting these mistakes helps pessimistic searchers find work. However, correcting errors can demotivate optimistic searchers and hurt them. Because of this, services must target help carefully rather than treating everyone the same way.
Imagine you are planning how long to study for a test. You have a guess about how hard the test will be, but that guess might be wrong. If you think it will be very hard, you study a lot and do well. If you think it will be easy, you might not study enough and fail. But if someone tells you the true difficulty, it could help the pessimist who overstudies, but it could also make the optimist who was already confident stop studying altogether. In this paper, job seekers' guesses about finding a job are like the test difficulty guess, and the researchers find that correcting those guesses helps some but hurts others, so they build a way to decide who should get the correction.
Employment services can now use job seekers' own expectations to decide who needs help and who might be harmed by it. This means resources go to those who actually benefit, and optimistic job seekers are not accidentally discouraged. The result is more effective reemployment support without wasting effort on counterproductive interventions.
The abstract doesn't explicitly state limitations. It does note that the informational content of biases was initially ambiguous, but the study resolves that. The findings are based on French job seekers and a specific modeling approach, so generalizability may be limited. The machine learning tool relies only on reemployment expectations, which may miss other relevant factors. The paper does not test real-world implementation of the targeting tool.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Biases-Informed Job Search Guidance: Characterization, Implications, and Targeting Support — Bruno Crépon, Aurélien Frot, Christophe Gaillac
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.