From Matching Models to Recruiting Agents: A Review of AI Recruitment Systems

Summary

This systematized narrative review examines how AI recruitment has evolved from bilateral retrieval and ranked lists to neural person-job matching, large language model components, and tool-using recruiting agents. The authors organize 40 representative works with supporting industrial and legal sources using a purposive search and coding protocol updated through 23 July 2026, with targeted updates through 2 September 2026. They identify three linked shifts: from similarity to reciprocal suitability, from a standalone model to a compound workflow, and from offline prediction to evaluation tied to evidence and productivity. The review covers document understanding, retrieval, ranking, assessment, interviewing, sourcing, and handoff to people, distinguishing evidence at field, pair, list, case, trajectory, and outcome levels. It cautions that behavioral labels can mix exposure, preference, and qualification, while private or synthetic data restrict external validity and final scores can hide failures within a pipeline. In the coded set, privacy is not directly evaluated, and no study jointly assesses utility, fairness, privacy, and security. The authors stress that these findings describe the coded sample rather than the entire field. They propose mapping evaluation evidence to the strongest defensible claim and call for reciprocal, evidence-grounded, temporally controlled, selective, and auditable systems that preserve uncertainty and support contestable decisions under explicit cost and risk constraints.