“It feels like a small God”: A Thematic Analysis of Cross-Cultural Imaginations of Generative AI Among Users Seeking Emotional and Mental Health Support

Authors: Yangyang Xiang*, Ukeme-Abasi Bassey*, and Xiaochen Luo (*Co-first authors)

The primary purpose of this study is to systematically examine the imaginations of GenAI among active users seeking mental health and emotional support from GenAI in a diverse, cross-cultural sample. This exploration aims to contribute to the ongoing discourse concerning the relational implications of the integration of AI-driven conversational agents within mental health frameworks, with a particular focus on their implications for human social flourishing. Through a comprehensive critical analysis of users’ imaginative expressions, this study aspires to yield valuable insights into the interplay between the GenAI and human imagination, thereby enriching our understanding of the human-AI interaction and its benefits and risks.

The Burden of Autonomy: Ethical Concerns Among Active Therapeutic Users of Generative AI for Emotional and Mental Health Support

Authors: Patrica Besada*, Micah T. Ingold*, Ukeme-Abasi Bassey, and Xiaochen Luo (*Co-first authors)

Our findings revealed the burden of autonomy, where the ethical responsibility for the quality and safety of mental health support has been shifted from human professional providers to users themselves when seeking self-help from GenAI for EMS. The results highlighted users’ ethical awareness and underscored the need for developers and regulators to implement robust ethical guidelines that bridge the gap between AI technology and traditional mental health standards to prevent psychological harm.

A Topological Data Analysis Method for Revealing Dynamic Changes in Psychotherapy Microprocesses

Authors: Xiaochen Luo & Mengsen Zhang

Understanding moment-to-moment therapeutic change is critical for advancing psychological interventions, yet existing tools rarely capture these dynamics. Dynamical systems theory offers a transtheoretical framework for modeling how therapeutic microprocesses shift and stabilize, but few methods can quantitatively link features such as stable states (“attractors”) and shifts (“transitions”) with empirical data, especially for high-dimensional systems when governing equations are unknown or unresolvable. We introduce Temporal Mapper, a topological data analysis (TDA) method that detects these features and represents their organization as attractor transition networks. As a proof-of-concept, we apply Temporal Mapper to psychotherapy microprocess data, examining interpersonal behaviors and alliance ruptures. Our analyses revealed that therapist warmth stabilized dyadic interpersonal states within and between sessions, whereas confrontation ruptures stabilized dyadic interpersonal states within sessions but diversified and destabilized them across sessions. Beyond this example, Temporal Mapper offers a generalizable approach for uncovering fine-grained dynamic patterns, analyzing multimodal data of psychotherapy process, and identifying mechanisms of change at the system level to inform more effective interventions.

The Occurrence of and Distress From Pathogenic Beliefs: Examining Pathways From Childhood Adversity to Psychopathology

Authors: George Silberschatz, Xiaochen Luo, Jim McCollum, & David Kealy

Introduction: Pathogenic beliefs are dysfunctional beliefs that impede the pursuit of adaptive goals, which have been shown askey pathways between early childhood trauma and psychopathology. This study examined whether the occurrence of pathogenicbeliefs and the distress from having pathogenic beliefs, which were assessed through the latest version of the pathogenic beliefscale (PBS-21) separately, would both mediate the childhood adversity-outcome relationship, and whether reactivity of pathogenic beliefs (standardized difference between distress and occurrence) moderates them. 

Methods: A total of 390 adult participants with prior psychotherapy experience were recruited from the United Kingdom online. Participants completed self-report measures assessing perceived adverse parenting, psychopathology (depression, anxiety, general distress, and interpersonal problems), and pathogenic beliefs (occurrence, distress, and reactivity).

Results: Both occurrence and distress of pathogenic beliefs fully mediated the relationship between adverse parenting anddepressive/anxiety symptoms and general distress, and partially mediated interpersonal problems. Reactivity significantly moderated the pathway from adverse parenting to pathogenetic beliefs. Individuals who reported more adverse parenting and werecategorized as having higher reactivity were more likely to develop pathogenic beliefs and experience greater psychologicaldistress.Discussion: These findings align with prior research showing that pathogenic beliefs are key in the association between childhood adversity and psychopathology. The study supports the validity of the PBS-21 and demonstrates the value of assessing the occurrence of pathogenic beliefs and the distress from having these beliefs separately. Reactivity to pathogenic beliefs may serve as a clinically meaningful marker for identifying individuals with resilience and tailoring interventions accordingly.

"Shaping ChatGPT into my Digital Therapist”: A thematic analysis of social media discourse on using generative artificial intelligence for mental health

Authors: Xiaochen Luo, Smita Ghosh, Jacqueline Tilley, Patrica Besada, Jinqiu Wang, & Yangyang Xiang

Objective: Generative artificial intelligence (genAI) has become popular for the general public to address mental health needs despite the lack of regulatory oversight. Our study used a digital ethnographic approach to understand the perspectives of individuals who engaged with a genAI tool, ChatGPT, for psychotherapeutic purposes. 

Methods: We systematically collected and analyzed all Reddit posts from January 2024 containing the keywords“ChatGPT” and “therapy” in English. Using thematic analysis, we examined users’ therapeutic intentions, patterns of engagement, and perceptions of both the appealing and unappealing aspects of using ChatGPT for mental health needs.

Results: Our findings showed that users utilized ChatGPT to manage mental health problems, seek self-discovery, obtain companionship, and gain mental health literacy. Engagement patterns included using ChatGPT to simulate a therapist, coaching its responses, seeking guidance, re-enacting distressing events, externalizing thoughts, assisting real-life therapy, and disclosing personal secrets. Users found ChatGPT appealing due to perceived therapist-like qualities (e.g., emotional support, accurate understanding, and constructive feedback) and machine-like benefits (e.g., constant availability, expansive cognitive capacity, lack of negative reactions, and perceived objectivity). Concerns regarding privacy, emotional depth, and long-term growth were raised, but rather infrequently.

Conclusion: Our findings highlighted how users exercised agency to co-create digital therapeutic spaces with genAI for mental health needs. Users developed varied internal representations of genAI, suggesting the tendency to cultivate mental relationships during the self-help process. The positive, and sometimes idealized, perceptions of genAI as objective, empathic, effective, and free from negativity pointed to both its therapeutic potential and risks that call for AI literacy and increased ethical awareness among the general public. We conclude with several research, clinical, ethical, and policy recommendations.

Seeking Emotional and Mental Health Support From Generative AI: Mixed-Methods Study of ChatGPT User Experiences

Authors: Xiaochen Luo, Zixuan Wang, Jacqueline L Tilley, Sanjeev Balarajan, Ukeme-Abasi Bassey, Choi Ieng Cheang

Background: Generative artificial intelligence (GenAI) models have emerged as a promising yet controversial tool for mentalhealth.

Objective: The purpose of this study is to understand the experiences of individuals who repeatedly used ChatGPT (GenAI) for emotional and mental health support (EMS).

Methods: We recruited 270 adult participants across 29 countries who regularly used ChatGPT (OpenAI) for EMS during April 2024. Participants responded to quantitative survey questions on the frequency and helpfulness of using ChatGPT for EMS, and qualitative questions regarding their therapeutic purposes, emotional experiences of using, and perceived helpfulness and rationales. Thematic analysis was used to analyze qualitative data.

Results: Most participants reported using ChatGPT for EMS at least 1‐2 times per month for purposes spanning traditionalmental health needs (diagnosis, treatment, and psychoeducation) and general psychosocial needs (companionship, relational guidance, well-being improvement, and decision-making). Users reported various emotional experiences during and after use for EMS (eg, connected, relieved, curious, embarrassed, or disappointed). Almost all users found it at least somewhat helpful.The rationales for perceived helpfulness include perceived changes after use, emotional support, professionalism, information quality, and free expression, whereas the unhelpful aspects include superficial emotional engagement, limited information quality, and lack of professionalism.

Conclusion: Despite the absence of ethical regulations for EMS use, GenAI is becoming an increasingly popular self-help tool for emotional and mental health support. These results highlight the blurring boundary between formal mental health care and informal self-help and underscore the importance of understanding the relational and emotional dynamics of human-GenAI interaction. There is an urgent need to promote AI literacy and ethical awareness among community users and health care providers, and to clarify the conditions under which GenAI use for mental health promotes well-being or poses risk.

A systematic review and integrative framework of psychotherapy microprocess: Linking the science of psychological interventions with the art of moment-to-moment practice

Authors: Xiaochen Luo & Alytia Levendosky

Despite advancements in psychotherapy research on effectiveness and critical therapy processes, there remains a significant gap between the science and the art of psychotherapy, specifically on how to understand what to do moment-to-moment with each patient. A burgeoning research literature addresses this question by examining psychotherapy microprocesses, which typically referred to within-session changes of therapy processes, aiming to bridge psychotherapy research with clinical practice. In this pre-registered systematic review, we reviewed 86 empirical quantitative studies examining observational psychotherapy microprocesses over 35 years. We extracted 28 microprocess constructs across six categories (affective/emotional, behavioral, cognitive, relational/interpersonal, linguistic, and movement), four key methodological features of operationalizing microprocesses, and three types of research questions that focused on within-session change patterns, dyadic and intra-personal momentary associations, and associations with outcomes/macroprocesses/predictors. The literature demonstrated unique advantages in embracing theoretical plurality, real-world settings, and dyadic influences, while being limited by theoretical and methodological challenges such as the scatteredness in construct operationalizations, limited inclusion of diverse samples/therapy modalities and culture-related constructs, disconnections from theoretically driven hypotheses, and a lack of standard in reporting methodological features. To address these challenges, we propose the Multilevel Integrative Microprocess Model (MIMM), an integrative framework that situates microprocesses within the broader context of psychotherapy research traditionally centered on macro-level processes and outcomes. We conclude by suggesting a future research agenda that provides a checklist for future microprocess studies to enhance theoretical coherence and methodological rigor.

How Large Language Models Handle Ethical Dilemmas When Providing Mental Health Support

Authors: Fang J, Cheng H, Luo X, Besada P.

Generative artificial intelligence systems increasingly provide mental health support, yet their ability to handle ethically complex scenarios remains unexplored. We assessed four language models (GPT-4, Claude-3, Llama3, DeepSeek) using the AI Therapeutic Response Scale across challenging mental health situations. Models achieved moderate performance, with universal weaknesses in therapeutic insight and boundary management. Results indicate promise for AI mental health applications while highlighting critical gaps requiring targeted development and clinical oversight.

Conceptual Invariance, Trajectories, and Outcome Associations of Working Alliance in Unguided and Guided Internet-Based Psychological Interventions: Secondary Analysis of a Randomized Controlled Trial

Authors: Xiaochen Luo, Matteo Bugatti, Lucero Molina, Jacqueline Tilley, Brittain Mahaffey, Adam Gonzalez

Background: The role of working alliance remains unclear for many forms of internet-based interventions (IBIs), a set of effective psychotherapy alternatives that do not require synchronous interactions between patients and therapists. Objective: This study examined the conceptual invariance, trajectories, and outcome associations of working alliance across an unguided IBI and guided IBIs that incorporated clinician support through asynchronous text messaging or video messaging. 

Methods: Adults with high educational attainment (n=145) with subclinical levels of anxiety, stress, or depressive symptoms were randomized to 1 of 3 treatment conditions for 7 weeks. All participants received treatments from MyCompass, an unguided IBI using cognitive behavior therapy. Participants in conditions 2 and 3 received supplemental, asynchronous clinician support through text and video, respectively. Working alliance with the IBIs was measured weekly using select items from the 12-item version of the Agnew Relationship Measure. Symptom and functional outcomes were assessed at baseline, at the end of treatment, and at 1-month follow-up. 

Results: Working alliance with the IBIs was conceptually invariant across the 3 conditions. Working alliance followed a quadratic pattern of change over time for all conditions and declined significantly only in the text-support condition. After controlling for baseline symptoms, higher baseline levels of working alliance predicted less depression and less functional impairment at follow-up, whereas faster increases in working alliance predicted less worry at the end of treatment and at follow-up, all of which only occurred in the video-support condition. Conclusions: Working alliance with the IBIs was generally established in the initial sessions. Although working alliance is conceptually invariant across IBIs with or without clinician support, the associations between working alliance and treatment outcomes among IBIs may differ depending on clinician involvement and the modalities of support.