AI-POWERED CHATBOTS IN CITIZEN SERVICE DELIVERY: EFFECTIVENESS, TRUST, AND INCLUSION

Abstract

Artificial intelligence-powered chatbots have emerged as one of the fastest-growing eGovernance innovations globally, with 43% of national government portals now deploying AI chatbot functionality as a primary citizen service interface. In India, chatbot deployments across central and state eGovernance platforms — including UMANG, DigiSeva, MADAD, and numerous state-level virtual assistants — represent significant investments in AI-mediated citizen-government interaction. Yet the dominant discourse around government chatbot adoption focuses almost exclusively on deployment metrics and cost efficiencies, while neglecting three dimensions of critical governance importance: the quality and accuracy of information delivered to citizens (effectiveness), the conditions under which citizens trust and rely on government chatbot responses (trust), and the differential impact of chatbot interfaces on digitally and socially marginalized citizen groups (inclusion). This paper addresses these gaps through a rigorous, multi-method empirical study drawing on primary data from 890 respondents across urban, peri-urban, and rural settings in five Indian states, supplemented by a systematic audit of query resolution quality across six government chatbot platforms and thematic analysis of 72 in-depth qualitative interviews. The study develops and validates an original Government Chatbot Quality Index (GCQI) — a composite instrument measuring technical performance, information accuracy, linguistic accessibility, emotional appropriateness, and inclusive design — and applies it across the six audited platforms. Key findings include: a mean query resolution rate of only 47.3% across sampled platforms; significant response quality disparities between English/standard Hindi queries and regional language queries (accuracy gap of 31.4 percentage points); elderly and rural users reporting substantially lower satisfaction (mean CSAT 2.8/5) than urban, educated users (4.1/5); and a critical finding that 62% of chatbot responses to sensitive welfare queries contained materially incomplete or inaccurate information that could adversely affect citizen decisions. The study identifies five structural failure modes in government chatbot design — resolution gap, linguistic exclusion, emotional blindness, bias in query interpretation, and accountability vacuum — and proposes the original CitizenFirst Chatbot Design Framework (C2DF) as a comprehensive design and governance architecture for equitable, trustworthy, and effective AI chatbot deployment in Indian eGovernance. Policy recommendations are directed at MeitY, NIC, the Ministry of Rural Development, and state IT departments.