Large Language Models in public policy research: methodological opportunities, normative standards, and potential challenges

Large language models (LLMs) have created new opportunities for public policy research, but their methodological value remains contested. This paper examines how LLMs may augment established research traditions in public policy research. It situates LLMs as an augmented extension of data-intensive research and analyzes their potential contributions to qualitative inquiry, quantitative analysis, and simulation-based research. Rather than treating LLMs as a completed methodological revolution or a fully independent research paradigm, the paper conceptualizes them as enabling tools whose value depends on validation, transparency, and ethical scrutiny. To address risks such as hallucination, algorithmic opacity, and value bias, the paper proposes an “Objective-Path-Boundary” framework as a domain-specific methodological governance framework for LLM-assisted public policy research. This framework defines reliability as the core objective, standardized protocols as the operational path, and value scrutiny as the ethical boundary. The paper also discusses unresolved challenges concerning knowledge production, researcher agency, academic institutions, and public governance. It contributes a conceptual map for the responsible use of LLMs in public policy research and identifies conditions under which their methodological potential may be realized.