Mental health applications increasingly serve as stand-alone interventions or adjuncts to clinical care, yet their capacity to support users experiencing acute psychological distress remains poorly characterized. This study introduces the Mental Health App Crisis Support Assessment Framework (MHACSAF), a structured instrument for evaluating crisis support implementation in mental health apps, and reports findings from its application to six commercial AI-powered products. MHACSAF is grounded in suicide prevention guidance from the World Health Organization, evidence-based safety planning interventions, and established principles of digital health evaluation and accessibility. The framework comprises an eligibility screening step followed by seven scored dimensions totaling 65 possible points: ease of access, coverage and prioritization, hotlines and emergency services, content clarity, technical accessibility, localization, and awareness. Three licensed clinical psychologists independently evaluated Wysa, Youper, Flourish, Earkick, Replika, and Ash using iOS platforms between December 2025 and January 2026. Inter-rater reliability was strong (Fleiss’ kappa = 0.87, 95% CI [0.71, 1.00]; ICC(2,1) = 0.94, 95% CI [0.83, 0.99]). Mean total scores ranged from 13.0 to 40.3 (M = 24.9, SD = 9.3); no application achieved ‘Good’ or ‘Excellent’ classification. Wysa performed best but still demonstrated gaps in accessibility, localization, and offline functionality. Technical accessibility for users with disabilities was nearly absent across products. Crisis resources were frequently buried behind conversational interfaces, and several apps delegated safety-critical information to external websites with broken or inaccessible links. These findings indicate that current AI mental health applications inadequately address user safety during psychological emergencies and suggest MHACSAF provides a reproducible methodology for benchmarking and improving crisis support implementations.
Performance of large language models in delivering accurate and comprehensible patient information on heart failure and cardiomyopathy
BackgroundLarge language models (LLMs) are increasingly used by patients seeking cardiovascular health information through digital platforms. However, their accuracy and suitability for providing guidance on