AI Healthcare Chatbots as Information Infrastructure: A Large-Scale Study of User-Reported Breakdowns

This study investigates the real-world performance of AI-enabled healthcare chatbots by analyzing over 15,000 negative user reviews from 59 mobile applications across the Google Play and Apple App stores. By conceptualizing these chatbots as information infrastructures—systems that mediate access to health knowledge and support—the authors employ topic modeling and interpretive analysis to categorize recurring user complaints. This approach moves beyond clinical evaluations to understand how these tools function within the lived, everyday contexts of their users.

The analysis identifies three primary categories of failure that disrupt the user experience:

1. Access Barriers and Service Unreliability: Users report significant technical friction, including connectivity issues, app crashes, and failures to access promised health information or services.
2. User Experience and Interaction Quality: This category captures dissatisfaction with the chatbot's conversational capabilities, such as inaccurate responses, lack of clinical grounding, and the inability to handle complex or emotional health queries effectively.
3. Billing and Customer Support Issues: Many users express frustration with opaque subscription models, unexpected charges, and the lack of responsive human support when automated systems fail.

Furthermore, the study highlights that while explicit mentions of privacy, security, and data handling are relatively infrequent, they are strongly correlated with the most severe negative ratings. This suggests that data-related concerns are not merely routine critiques but are markers of heightened distrust in the platform's governance.

As AI chatbots become increasingly embedded in digital health ecosystems, they act as critical intermediaries for health management. This research demonstrates that these systems often fail to meet user expectations, creating friction that can undermine their potential as reliable health tools. By framing these failures as infrastructural breakdowns, the authors provide actionable insights for designers and policymakers to improve the usability, transparency, and trustworthiness of digital health systems, ensuring they better serve the needs of users in everyday life.