Abstract Public-sector organizations increasingly deploy artificial intelligence (AI) for forecasting, service delivery, and high-stakes decision-making, but adoption is constrained by accountability, fairness, and trust requirements that have no counterpart in private-sector deployment. Research on AI in the public sector has grown rapidly and fragmented across disciplines, making it difficult to translate scholarship into governance-relevant guidance. We propose a functional framework that organizes this fragmented literature by four practical functions of public governance ( creating public value , delivering public services , responsiveness to the public , and protecting state–society relations ) rather than by application domain, technology, governance risk, policy-cycle stage, or public value alone. The framework offers a translation layer that makes the field discoverable and comparable across the disciplines that study it in parallel. Combining a structured citation-based literature review with BERTopic modeling of 3,268 works retrieved from OpenAlex through May 2026, we map where scholarship concentrates and where it is thin. The literature has pivoted sharply in the post-2022 period, which we date heuristically to the public release of ChatGPT: domain-application work on creating public value has been overtaken by an expanding cluster on state–society relations (algorithmic fairness, ethics, regulation), while scholarship that explicitly situates AI within the policy cycle remains comparatively scarce.
Mintz et al. (Mon,) studied this question.