engagement and governance. However, the design and impact of these systems vary significantly across political and cultural contexts. This study addresses that gap by comparing six deployments to determine how participatory design practices influence citizen trust, usability and system performance. By situating LLM chatbots within broader debates on democratic legitimacy and data-driven urban governance, the research highlights the tension between rapid, centrally orchestrated rollouts and slower, deliberative co-creation. A mixed-methods strategy underpins the analysis. We combine 5.8 million interaction logs, average response-time metrics and task-completion rates with 1,142 resident surveys and 22 hours of co-design workshop transcripts. Cross case synthesis reveals two distinct patterns. Chinese cities achieve superior technical metrics (median task-resolution of 92 percent and sub-second average response times), through top-down integration of retrieval-augmented generation and learning-to-rank pipelines but involve citizens chiefly in post-deployment feedback. U.S. counterparts embed workshops and public betas early in the lifecycle, raising mean trust and satisfaction scores by 15–20 percentage points, albeit at the cost of longer development schedules and narrower initial scopes. Building on these insights, we propose a Participatory Chatbot Design (PCD) framework that couples risk-gated technical sprints with structured stakeholder engagement, performance dashboards and procurement clauses mandating citizen input. The framework reconciles the efficiency of centralized systems with the legitimacy of inclusive governance, offering actionable guidance for city managers and vendors.