"Effectiveness of AI Chatbot in reducing Anxiety Levels in a Community Mental Health Center" (Preprint)

BACKGROUND Mental health in latin AMERICA´S countries does not have the importance that it it would have, because the profesionals , like psicologysts, psyquiatrics , nurses and others are not enough to provide the services that community needs. The WHO tells us “The global average number of mental health workers per 100,000 inhabitants has increased slightly, from nine workers in 2014 to 13 workers per 100,000 inhabitants in 2020.(4) In addition, mental health services do not reach a significant number in both developed and developing countries(5), thus raising the need to know if a chatbot installed on a smartphone of people with anxiety symptoms can reduce them or serve as support in psychological treatment, being a tool that can be, in the near future, integrated into the health system. The emergence of chatbots based on artificial intelligence has revolutionized the field of clinical psychology and psychotherapy, providing people with unprecedented access to professional assistance, overcoming the limitations of time and location (6) and above all due to its low cost. There are studies that show that the use of chatbots is promising to reduce the symptoms of depression, anxiety (6,7) through the use of CBT (cognitive behavioral therapy) achieving greater adherence to psychotherapy and therefore a better response to treatment. OBJECTIVE Objective: To determine the effectiveness of an artificial intelligence–based chatbot in reducing anxiety levels among adult users of the Centro de Salud Mental Comunitario Cayma, Arequipa, 2025. METHODS Methodology: A quasi-experimental study with a pretest–posttest design and a descriptive–relational level was conducted. The sample consisted of 30 users of both sexes, aged between 18 and 59 years, diagnosed with a mental health disorder and presenting anxiety symptoms. The “Florence” chatbot, developed with artificial intelligence and natural language processing, was implemented for six weeks. The Zung Self-Rating Anxiety Scale (SAS) was applied to assess anxiety levels before and after the intervention, and the System Usability Scale (SUS) was used to evaluate the perceived usability of the chatbot. Data were analyzed using descriptive and comparative statistics, focusing on anxiety reduction and system acceptability indicators. RESULTS Results: Before the intervention, 36.7% of users presented severe anxiety, 30% mild, 20% moderate, and only 13.3% normal levels. After six weeks of chatbot interaction, 53.3% of users achieved normal anxiety levels, while severe cases decreased to 16.7% and moderate to 13.3%. The overall reduction in anxiety symptoms reached 50%, showing a significant improvement in psychological well-being. Regarding usability, 46.7% of participants rated the system as acceptable, 13.3% as good, and another 13.3% as excellent, reflecting a positive overall perception of the chatbot as an emotional support tool. CONCLUSIONS Conclusion: The artificial intelligence chatbot “Florence” proved to be effective in reducing anxiety levels, demonstrating its potential as a complementary psychological intervention and emotional support resource within community mental health settings. Its accessibility and high usability make it a viable tool to enhance therapeutic support and optimize care for populations with limited access to mental health professionals. However, it should be considered a supplementary aid rather than a substitute for professional care, and further large-scale studies are recommended to validate and strengthen these findings. CLINICALTRIAL https://docs.google.com/document/d/1QkcTTspS5Y78R7iDlLYgHDLMq36mMp3m/edit