JMIR Publications

ABSTRACT

Background:

Mental health needs have increased significantly while access to care remains limited, prompting interest in digital tools that can help bridge treatment gaps. Artificial Intelligent (AI) mental health chatbots are increasingly being used to help address the shortage of mental health providers by offering accessible, scalable support when traditional services are unavailable or overwhelmed. Existing research focuses largely on technical development and ethical issues, with far less attention to user experiences and adoption factors.

Objective:

This study provides a multifaceted analysis of user perceptions, experiences, and attitudes toward two widely used mental health AI chatbots—Woebot and Wysa. By analyzing user reviews, the study aimed to identify potential benefits, limitations, and areas for improvement that influence consumer adoption.

Methods:

Linear regression analysis revealed a decline in Woebot’s average star ratings over time, while Wysa ratings remained stable. However, due to the ordinal nature of star ratings and a low R² value, these findings warrant cautious interpretation, and future analyses may benefit from using ordinal regression models. Sentiment analysis—conducted using both AI-based and lexical-based tools—indicated overall positive sentiment, although sarcasm and linguistic ambiguity posed challenges. Thematic analysis, informed by previous work from [1], identified key adoption determinants such as performance expectancy, price value, trust, and perceived anthropomorphism.

Results:

Concerns about privacy, generic responses, emotional disconnect, and cost emerged as significant barriers. The study highlights the need for more consistent standards in chatbot evaluation, data privacy, and trust-building in AI healthcare tools. Suggestions for improving chatbot usability include integrating feedback loops, developing more nuanced sentiment analysis through models like BERT and LSTM, and expanding linguistic research on user-chatbot interactions. Furthermore, equitable access remains a priority, with policy discussions underway to support insurance reimbursement for AI-based mental health tools.

Conclusions:

Overall, the findings underscore the promise of mental health chatbots while emphasizing the necessity for continued research, ethical oversight, and interdisciplinary collaboration to ensure these tools are safe, effective, and accessible to all.