AI mental health chatbot feels empathetic, but safety evidence still lags

A new Scope review finds that while AI-powered mental health chatbots can feel approachable, personal, and empathetic, important questions remain about safety, continued use, and real-world clinical value.

Research: Generative AI mental health chatbots: A scoping review of intervention design and user experience. Image credit: ZilverlightArt / Shutterstock

In a recent article published in a magazine npj digital medicineresearchers reviewed the user experience (UX) and generative artificial intelligence intervention design (Genai) Mental Health Chatbot.

Approximately 25% of people around the world experience mental health problems, but most, approximately 85%, do not receive appropriate treatment due to barriers such as stigma, cost, shortage of professionals, geographic distance, and structural inequalities. The increasing prevalence and treatment disparities of mental disorders are leading to innovative approaches to mental health care. In this context, digital tools have gained attention for enabling mental health interventions due to their scalability and convenience.

Digital mental health interventions provide treatment and support through chatbots, websites, mobile applications, wearables, and more. Conversational agents (chatbots) are applications that use machine learning and natural language processing algorithms to simulate human interaction. Traditional mental health chatbots use rule-based or search-based systems to deliver pre-scripted treatment content, but are limited in their ability to personalize support and recognize users’ needs.

Unlike traditional chatbots, a large language model (LLM)-based chatbots can simulate core aspects of therapeutic encounters, such as personalized suggestions and empathetic consideration. but, Genai The system may generate incorrect or inappropriate responses. Plus, unlimited conversational ability Genai systems make intervention design more critical and complex compared to rule-based systems.

the study

In this study, researchers reviewed the design and UX Outcomes of the intervention Genai Mental health chatbot. First, we performed a systematic literature search to identify studies on system design and deployment. Genai Mental health chatbot. Reviews, editorials, media articles, and commentary are excluded. A total of 21 studies conducted in 11 countries between 2023 and 2025 were selected.

The most studies came from China and the UK, followed by the US. Interventions ranged from early stage prototype evaluation to clinical trials and one real-world implementation study. Most studies included general or clinical adult populations, including older adults with dementia, while some included simulated users and college students. Sample sizes ranged from 5 to 527 participants. There was considerable heterogeneity in outcome measures across studies, and most interventions remained in early stages of development.

Characteristics of the intervention design

Chatbot interventions often targeted depression and anxiety and employed common treatment mechanisms such as mindfulness, emotional regulation, and cognitive restructuring. Some systems focused on mental health, stress and loneliness, emphasizing general support and preventive care rather than specific treatments. Others targeted eating disorders, post-traumatic stress disorder, and dementia.

Dementia-focused interventions aimed to address psychological aspects associated with dementia, such as caregiver burnout, psychological distress, and loneliness, rather than central neurological features. Furthermore, most interventions were based on principles of cognitive behavioral therapy, such as behavioral activation, psychoeducation, Socratic dialogue, acceptance and commitment therapy, cognitive restructuring, and mindfulness.

Interventions also varied in frequency, delivery, and duration, with most being short-term, ranging from 2 to 8 weeks. These were primarily deployed through web-based interfaces and mobile applications. Some tools were introduced through messaging or social media platforms. Approximately 67% of interventions were disembodied, text-based chatbots, while other interventions used voice, avatar-based, augmented reality, or other multimodal interactions to enhance engagement and realism.

GenAI Mental Health Chatbot UX

All but two studies evaluated at least one UX The majority use quantitative scales such as Likert scales. Some studies used qualitative feedback such as open-ended questions or semi-structured interviews. The most common outcomes were user satisfaction and acceptance. Participants described the intervention as convenient and accessible across studies, with generally moderate to high acceptability and high user satisfaction reported.

Half of the studies investigated usability using qualitative feedback, system usability scales, or Likert scales. In addition to interface design, interaction modes, and deployment platforms, we also observed differences in usability across studies. Users preferred free-flowing chat interfaces and customizable features over predefined options. Additionally, some interventions were unclear in scope or limited in functionality, leaving users unsure about the chatbot’s functionality.

Only some studies reported objective usage and engagement metrics, such as session frequency, interaction duration, retention rates over time, and task completion. Repeated measures designs often exhibit attrition patterns over time. For example, uptake rates were initially high during a multi-week intervention but declined over time. Additionally, most chatbots have personalization capabilities, reflecting their ability to adapt conversations and conversations. UX Interface with previous interactions and emotional states.

The most common personalization strategy was emotion detection with adaptive interactions, which allows users to receive customized interactions and empathetic reflections. Studies did not consistently measure perceived impact as an independent indicator. It was usually embedded in qualitative or broader feedback. UX evaluation. For interventions with the most detailed data, the most commonly perceived benefits were improved clarity and awareness.

However, this review also found that personalization and empathy do not always lead to stronger clinical outcomes or sustained use. Some users reported repetitive, generic, or out-of-context responses, while others raised concerns about over-reliance on chatbots, reduced human contact, data privacy, and the system’s ability to respond appropriately in times of crisis. Some studies have also found that inaccurate or clinically inconsistent output is associated with decreased trust and loss of engagement.

The authors noted that some design features are not proven to produce results, but rather are associated with improvements. UX result. These include a familiar deployment platform, richer interaction modalities, integration into existing care pathways, personalization, domain knowledge foundation, structured delivery, proactive outreach, and co-design with experts and end users. However, the predominance of early-stage studies and limited direct comparative analysis prevented firm conclusions about which functions were directly improved. UX.

conclusion

Collectively, Genai Chatbots have significant potential to provide tailored, empathetic mental health support and demonstrate strong acceptance. However, by standardizing UX Evaluation, intervention design based on user needs and preferences, and maintaining engagement remain major challenges. Addressing these requires co-design with experts and users. UX Metrics in long-term studies, transparent reporting standards, independent evaluation, clear reporting of model design and training data, and stronger attention to the limits of safety, equity, and crisis response.