Citizen Sentiment and Public Value in AI-Mediated Public Services: A Case Study of a Ministry Chatbot

Governments increasingly deploy artificial intelligence (AI)-driven chatbots to improve accessibility and efficiency in citizen service delivery. Yet empirical evidence remains limited regarding how citizens affectively experience these systems and how such experiences can inform the evaluation of AI-mediated public services. This study examines Israel's Ministry of Labor chatbot through a quantitative case study of 1,846 citizen-chatbot interactions collected between May and November 2024. Using public value and legitimacy as a conceptual lens, the study applies sentiment analysis with a Hebrew-language transformer model (HeBERT) and topic classification through a hybrid semi-supervised approach. Results show that most citizen enquiries were neutral (57.45%) or negative (39.24%), while positive enquiries were rare (3.31%). Chatbot responses were overwhelmingly neutral (90.24%), indicating a consistent institutional tone with limited emotional expressiveness. Topic analysis identified reserve military service as the dominant enquiry category (78.6%), followed by dismissal, general bureaucracy, and parenthood and leave. Although descriptive differences in sentiment across topics were observed, the association was not statistically significant. The findings underscore the importance of examining emotional and contextual dimensions of citizen-chatbot interaction alongside operational service metrics. More broadly, the study demonstrates how interaction log data can offer a scalable empirical lens on citizens' affective experiences with government chatbots and provides a replicable approach for analysing citizenfacing AI in public service contexts.