The review documents the evolution of AI in recruitment, moving from simple profile matching to complex workflows involving evidence retrieval, candidate comparison, and action support.
It covers 40 works, including neural person-job matching, large language model components, and recruiting agents that use tools. The review emphasizes the importance of evidence at various levels, such as document understanding, ranking, and outcome assessment.
Key issues include confounding behavioral labels, limited external validity due to private data, and the lack of comprehensive evaluation of utility, fairness, privacy, and security. The authors propose a staged mapping from evidence to defensible claims and suggest an agenda for more transparent, evidence-grounded systems.
Progress in AI recruitment should be measured by the ability to retrieve relevant evidence, maintain uncertainty, support contestable decisions, and improve outcomes within defined risk constraints.
Source: https://arxiv.org/abs/2609.04286