Governing in the Age of Generative AI: Rethinking Agenda Setting Theory in the AI Era

• In Event: AI, Power, and Political Narratives

Abstract

As generative artificial intelligence (GenAI) systems, such as ChatGPT and large language models, enter the policy arena, they enhance government capabilities and reshape foundational processes of public decision-making (Androniceanu, 2024; Salah et al., 2023). This paper examines how GenAI is transforming agenda setting, a critical yet under-theorized stage in the policy process literature, where issues are defined, problem frames are constructed, and political attention is mobilized. Drawing on Shenzhen’s pioneering application of GenAI in digital governance, the study engages directly with agenda-setting theory (McCombs & Shaw, 1993), assessing how GenAI alters the dynamics of problem recognition, policy stream formation, and issue prioritization. Agenda-setting theory traditionally conceptualizes the emergence of policy agendas as a product of three loosely coupled streams—problems, policies, and politics—that converge during windows of opportunity (Birkland, 2017). This framework's expert-driven discourse and elite framing are central in defining which issues are visible, legitimate, and actionable. However, GenAI’s ability to autonomously generate text, simulate stakeholder voices, and respond adaptively to prompts introduces a new actor in the discursive space of agenda setting. Through its large-scale language modeling, GenAI can produce policy memos, simulate public opinion, and even reframe issues in real time, potentially democratizing or disrupting traditional mechanisms of problem selection.

Using a mixed-methods approach, the study combines policy document analysis with a case study of Shenzhen’s integration of GenAI in public governance platforms, administrative planning, and automated policy briefing tools. These applications are analyzed to assess how GenAI technologies are deployed to identify emerging policy concerns, filter citizen input, and shape issue salience across bureaucratic hierarchies. The findings suggest that GenAI contributes to a recomposition of the agenda-setting process in three ways. First, it blurs the boundary between information provision and issue framing, as AI-generated texts influence how problems are defined and prioritized. Second, it accelerates the policy stream by enabling real-time synthesis of alternative proposals and stakeholder positions. Third, it introduces a new form of procedural opacity—algorithmic agenda-setting—where model training data, prompt design, or institutional delegation to AI systems may shape issue salience. These developments challenge the classical assumptions of agenda-setting theory, particularly the role of bounded rationality, elite framing, and the media-policy interface. GenAI functions as a discursive agent rather than acting solely as a tool for policymakers or interest groups, capable of reframing, amplifying, or de-emphasizing policy issues, sometimes in ways not fully anticipated by its users.

This paper contributes to contemporary policy theory by arguing that generative AI introduces a functional and epistemic shift in forming policy agendas. The Shenzhen case illustrates that as governments embrace AI to “listen at scale” or automate knowledge generation, the politics of attention, legitimacy, and problem framing must be re-theorized. In the algorithmic age, agenda setting is no longer just a political or media-driven process—it is increasingly a computationally mediated process.

Authors

• Tianle Liu, Harbin Institute of Technology (Shenzhen)
  Presenting Author

• Zhengsheng YU, Hong Kong University of Science and Technology (Guangzhou)
  Non-Presenting Co-Author

• Jing Lyu, Hong Kong University of Science and Technology
  Non-Presenting Co-Author