The rapid diffusion of large-scale generative systems into scholarly and industrial pipelines has fundamentally reshaped how investigators, practitioners, and institutions produce labelled data for downstream computational models. Where manual coding once dominated empirical social research, computational linguistics, and computer vision alike, the capacity of contemporary foundation models to classify, extract structured information, and interpret unstructured content has ignited a sweeping reconfiguration of annotation workflows across disciplines. This review advances a scholarly synthesis of the emerging paradigm in which labelling produced by generative systems is coupled with sustained, deliberate human judgment. It interrogates the conceptual foundations, methodological architectures, and socio-technical commitments that underlie this hybridisation, arguing that credible labelling now depends not on the replacement of the analyst but on the careful choreography of algorithmic suggestion and expert verification. Drawing on an interdisciplinary body of scholarship spanning computational linguistics, human–computer interaction, responsible artificial intelligence, and applied empirical research, the review traces the historical antecedents of supervised labelling, examines the capacities and limits of generative annotators, and articulates principles by which responsible collaboration between analysts and machines can be organised. It further considers the risks of unexamined reliance on generative outputs, including representational biases, reliability drift, and the erosion of analytic accountability, and proposes design heuristics that position the annotator as an interpretive partner rather than a passive verifier. The discussion closes with a programmatic outlook on governance, evaluation, and research priorities necessary to preserve methodological rigour as these systems mature. The contribution targets scholars, practitioners, and policy actors seeking practical frameworks that are epistemically defensible, organisationally scalable, and ethically sound across diverse real-world deployment settings worldwide.