Publication

Abstract : Citizen participation in international policymaking faces persistent structural barriers rooted in scale, linguistic diversity, and institutional distance between citizens and governance bodies. While AI tools offer promising opportunities for large-scale consultation, existing research tends to evaluate them at the level of platforms rather than across the sequential stages of a governance process, and rarely grounds evaluative criteria in collaborative governance theory. This paper addresses that gap by developing an analytical framework based on Emerson et al.’s integrative model of collaborative governance. It conceptualises AI as a stage-specific enabling condition that can strengthen or weaken shared motivation across four stages of principled engagement: initiation, discovery, deliberation, and determination. The framework is illustrated empirically using the Civil 20 India 2023 International Grassroots Survey ( N  = 5,152 participants from 64 countries in 24 languages), a large-scale consultation in a multilingual and multicultural global civil society setting. At the initiation stage, subgroup analysis of response length and non-response patterns reveals substantial variation in participation quality. Global South participants produce a median of 6 words compared to 9 in the Global North, with the sharpest compound disadvantage among Global South primary-educated participants, 74.2% of whom produce the answers too short to carry substantive policy-relevant content. At the discovery stage, keyness analysis reveals systematic frame divergence between subgroups that topic prevalence metrics alone obscure: groups that appear to share a concern often articulate it through fundamentally different terms. These findings provide diagnostic support for the initiation and discovery propositions and show how aggregate metrics can conceal the inequalities that AI-assisted design must address. The deliberation and determination propositions are developed theoretically, identifying the preservation of minority frames and traceability of citizen input into policy documents as key design conditions for AI-assisted participation.