To Prompt or Not to Prompt: AI Chatbot Use in Norwegian Public Administration

Data for this study will be collected in December 2025. Which factors determine whether public administrators embrace AI chatbots in their daily work? Public administration is experiencing rapid diffusion of large language model (LLM) chatbots into routine and decision-support workflows, yet systematic evidence on who actually uses these tools and under what organizational conditions remains scarce. This paper situates the inquiry within the broader digitalization of government and addresses the gap between governments as regulators of AI and governments as users. Understanding user characteristics is essential for assessing public value, discretion, and implementation risks. Building on scholarship in innovation diffusion and public management information systems, the study examines individual and organizational determinants of professional chatbot use among Norwegian central government administrators. The research asks: (1) How frequently do public administrators use AI chatbots for professional tasks? (2) How do hierarchical rank, role routineness, tenure, sector, and administrative level relate to use? (3) To what extent do attitudes toward AI, departmental training quality, and available use cases condition adoption and intensity of use? The theoretical framework adapts Lee’s determinants of Public Management Information System (PMIS) use. It posits inverse associations between leadership rank or tenure and use, and positive associations for routine-intensive roles and sectors with higher AI exposure such as health and finance. The planned methodology leverages the Norwegian Panel of Public Administrators (Round 8), targeting approximately 1,400 to 1,500 administrators across ministries and directorates. The dependent variable is self-reported frequency of professional chatbot use; covariates include LLM attitudes, department training quality, use cases, rank, tenure, role, administrative level, sector, and controls such as age, gender, and education. Analyses will comprise descriptive statistics, marginal means, and linear regression, with ordinal and binary logistic specifications as robustness checks. By clarifying who uses chatbots, for what purposes, and under which institutional conditions, the paper contributes evidence to ongoing debates on AI’s role in the public sector. The findings aim to inform policy design, training strategies, and governance safeguards for AI-assisted administrative work.