Mapping the Ecosystem of Large Language Models’ Implementation in the Public Sector: A Systematic Evidence Map

This study examines the implementation of Large Language Models (LLMs) in the public sector through a systematic literature review. Despite growing scholarly attention, existing knowledge remains fragmented, lacking a comprehensive map of empirical evidence. Following PRISMA guidelines, 26 studies from academic databases were analysed using thematic analysis for qualitative synthesis and descriptive statistics to identify patterns across the literature. Four core themes emerged from the analysis: applications of LLMs across government functions; architectures enabling deployment; governance mechanisms shaping implementation; and challenges impeding adoption. Key research trends reveal that scholarly attention concentrates heavily on citizen-facing and policy domains, while operational areas remain underexplored. The field remains in an exploratory phase, with General GenAI/LLM studies dominating compared to enterprise platforms, multi-agent, and multimodal typologies. Internal organisational frameworks emerge as the predominant governance mechanism examined. Technical performance challenges motivate most scholarly inquiry, yet sovereignty and equity concerns appear least often as initial drivers yet recur as barriers in practice. Finally, architectural complexity inversely relates to adoption: basic architectures predominate while complex deployments remain limited, confirming that advanced configurations demand infrastructure, capacity, and resources concentrated in specialised research teams rather than widespread practice. As its central contribution, the study proposes a conceptual phased implementation framework accompanied by a comprehensive guideline tool designed to help public sector organisations navigate the complexities of LLM adoption. Two additional outputs are developed: a practitioner-oriented guide offering LLM implementation tools/checklists, and a research-oriented agenda outlining theoretical gaps for future inquiry.