Large language models (LLMs) are increasingly discussed in digital government research, but existing studies remain fragmented across application opportunities, technical performance, and governance risks, with limited synthesis of how they support service provision within digital government service processes. Using the concept of interpretive support, this study examines: (1) what forms of interpretive support LLMs provide; (2) what task- and service-level effects are reported; and (3) what governance conditions shape responsible integration. Drawing on a socio-technical systems perspective, the study conducts a PRISMA-informed systematic review and thematic synthesis of 60 studies from the Web of Science Core Collection and Scopus. The findings show a shift from stand-alone question answering to broader forms of interpretive support, including rule explanation, service navigation, complaint interpretation, issue routing, and back-office knowledge structuring. The strongest evidence concerns efficiency, responsiveness, and accessibility, whereas claims about trust, accountability, and wider public value remain less well supported. The review concludes that public value is conditional rather than automatic, depending on reliability, legal boundaries, responsibility allocation, data security and privacy, and organizational capacity. Future research should examine whether these service-level gains persist in routine service environments and how governance mechanisms affect service quality, equity, and accountability.
Zhang et al. (Fri,) studied this question.