From Matching Models to Recruiting Agents: A Systematized Narrative Review of AI Recruitment Systems, Evaluation, and Governance
A systematic review extended the automated objects of artificial intelligence recruitment from resume matching and ranking lists to multi-stage workflows, covering evidence retrieval, candidate comparison, and action execution. The study was based on targeted search and coding protocols as of September 23, 2026, updated to September 29, integrating 40 representative articles and related legal industry sources. The review analyzed three coupled transformations: from similarity to mutual suitability, from single models to composite workflows, and from offline predictions to evidence-based and productivity-oriented evaluations. The study distinguished the levels of evidence for various stages such as document understanding, retrieval, ranking, evaluation, interviewing, recruitment, and manual handover, and identified persistent gaps including confusion in behavioral label exposure and preferences, limitations on private data affecting external validity, final scores masking process failures, and lack of direct privacy assessment. To address these issues, the researchers proposed a step-by-step mapping from evaluating evidence to making the strongest defensible claims, and developed a mutually inclusive, evidence-based, time-controlled, selective, and auditable system agenda.