This paper outlines a doctoral research proposal to develop a computational framework using large language models (LLMs) for the systematic analysis of public discourse in policy-making. A wide range of citizen feedback, offered through mechanisms such as public comment systems and amplified by generative artificial intelligence, is often not meaningfully incorporated into administrative decision-making, resulting in a lack of substantive policy reflection. This study proposes an integrated framework to evaluate public input along three key dimensions: “policy reflection,” “deliberative quality,” and “diversity and balance.” We aim to develop and validate a model that can automatically quantify these aspects using Japan’s Sixth Basic Environmental Plan as a case study. Initial findings from previous studies indicate that LLMs are particularly effective in evaluating the intuitive dimensions of deliberative quality. This research will culminate in a prototype for an online platform that fosters more transparent and verifiable information-based dialogue between citizens and policymakers.