A Pilot Randomized Controlled Trial of AI-Delivered vs. Human-delivered iCBT for Depression in Young Adults

A Pilot Randomized Controlled Trial of AI-Delivered vs. Human-delivered iCBT for Depression in Young Adults | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Pilot Randomized Controlled Trial of AI-Delivered vs. Human-delivered iCBT for Depression in Young Adults Yiyang Wu, Haoran Song, Chen Ye, Ruoyu Lin, Weihao Huang, You Wang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8269763/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 24 Feb, 2026 Read the published version in BMC Psychiatry → Version 1 posted 12 You are reading this latest preprint version Abstract Background: This randomized controlled trial compared AI-driven and human peer counselor-delivered internet-based Cognitive Behavioral Therapy (iCBT) for depressive symptoms (primary outcome) in young adults. Addressing a gap in literature, we explored the comparative effectiveness and acceptability/perception of these two modalities of online intervention. Methods: Ninety young adults were randomized to AI-driven iCBT, human peer counselor iCBT (participant-blinded), or a waitlist control. Interventions consisted of eight hours over four weeks. Depressive symptoms (primary outcome), suicidal ideation, and self-efficacy were assessed at baseline, week two, and week four. Qualitative analysis explored participant perceptions. Results: Both iCBT interventions significantly reduced depressive symptoms ( p < .05). No significant difference was observed between intervention groups at week two. The AI group improved significantly from baseline to week two ( p =0.004) but showed no further significant reduction by week four (a plateau effect), while the human group demonstrated continued improvement. Qualitative analysis indicated that participants valued AI’s convenience and accessibility but expressed concerns regarding its emotional understanding and personalization. Conclusion: AI-iCBT shows promising short-term efficacy, comparable to human counseling up to week two, but its limitations in emotional perception and sustaining therapeutic momentum resulted in a plateau effect. Future AI development must focus on improving emotional interaction and personalized support. Trial registration : ChiCTR2400088423. Registered on 19 August 2024. AI depression suicide ideation internet-based cognitive behavioral therapy randomized controlled trial young adults user experience Figures Figure 1 1. Introduction Depression, characterized by persistent low mood, anhedonia, psychomotor retardation, and in severe cases, suicidal ideation [1], represents a substantial global health challenge. The self-reported prevalence of depressive symptoms has risen steadily between 2001 and 2020, reaching 34% [2], suggesting that over one-third of the global population may experience depressive episodes. Psychotherapy, particularly cognitive behavioral therapy (CBT), has demonstrated efficacy in treating various mental health conditions, including depression [3]. However, access to qualified therapists is often limited, and traditional CBT typically involves extended treatment durations and considerable costs. These limitations are compounded by patient-level barriers, such as stigma surrounding mental illness and reluctance to seek help, further restricting access to and the overall effectiveness of psychological car [4]. Although effective, traditional CBT for depression presents practical challenges, including lengthy treatment periods, complex therapeutic processes, and substantial costs, prompting the development of internet-based CBT (iCBT) [5]. iCBT delivers CBT modules through online platforms using text, videos, or interactive exercises, offering advantages in convenience, cost-effectiveness, and accessibility, and has demonstrated therapeutic benefits in various settings [6]. Randomized controlled trials (RCTs) have shown the effectiveness of iCBT in reducing depressive symptoms among adolescents through diverse formats, including online modules combined with family support and therapist guidance [7]. Moreover, research indicates that guided iCBT, incorporating therapist support, can be particularly beneficial for individuals with moderate to severe depression [8]. A large individual patient data network meta-analysis by Karyotaki et al. (2021) found that while both guided and unguided iCBT were effective compared to control conditions, guided iCBT yielded greater short-term benefits for individuals with higher baseline depression scores. This underscores the potential for tailoring iCBT interventions to individual needs and symptom severity. However, iCBT is not without limitations. Notably, the complexity of certain online programs can lead to reduced treatment adherence due to patient avoidance or resistance [7]. The rapid advancement of artificial intelligence (AI) presents a promising avenue for addressing the limitations of traditional iCBT. Early studies have explored the potential of AI-based interventions in providing mental health support. For instance, Fitzpatrick et al. (2017) demonstrated the effectiveness of a fully automated conversational agent (Woebot) in reducing depressive symptoms among young adults. Similarly, Green et al. (2020) developed and tested a mobile application, "Healthy Moms," designed to provide automated psychological support for perinatal depression in Kenya. These preliminary findings suggest the feasibility, acceptability, and potential efficacy of AI-driven interventions, particularly in resource-constrained settings where they can expand access to mental health services. These early systems often relied on pre-defined rules and limited natural language processing capabilities. However, recent breakthroughs in deep learning, particularly the development of large language models (LLMs), have revolutionized the potential of AI in mental health. LLMs, such as OpenAI's ChatGPT, are trained on massive datasets of text and code, enabling them to exhibit sophisticated natural language understanding, generation, and dialogue capabilities [11]. This capacity allows for more natural and engaging user interactions, potentially overcoming the challenges of program complexity and limited personalization observed in earlier iCBT approaches. Integrating the core principles of CBT into LLM-driven systems, delivered through user-friendly iCBT platforms, thus offers a compelling opportunity to enhance the therapeutic experience and improve intervention outcomes. LLMs, as a prominent subset of AI, possess several key advantages that can mitigate the limitations of both traditional CBT and earlier forms of iCBT. First, LLMs can deliver CBT interventions with high fidelity and consistency, ensuring standardized, evidence-based care and minimizing variability inherent in human-delivered therapy. Second, LLMs offer highly personalized and interactive experiences, adapting to individual user needs, preferences, and real-time progress. This dynamic adaptation can enhance user engagement and adherence, a recognized challenge in iCBT [12]. Third, LLMs can provide immediate and readily available 24/7 support, overcoming geographical barriers and scheduling constraints that often limit access to traditional therapy [13]. Furthermore, the capacity of LLMs to process and analyze vast amounts of data enables continuous learning and refinement of the intervention, potentially leading to more effective and efficient therapeutic outcomes over time [14]. These advancements provide a robust theoretical foundation for exploring iCBT powered by advanced LLMs, such as ChatGPT, as a potentially effective and scalable intervention for depression. Based on these theoretical considerations and empirical findings, this randomized controlled trial aims to compare the effectiveness of AI-driven (ChatGPT-powered) iCBT and human peer counselor-delivered iCBT in reducing depressive symptoms and suicidal ideation among young adults. We hypothesize that AI-driven iCBT will demonstrate efficacy in reducing depressive symptoms and suicidal ideation, and we will explore its comparative effectiveness relative to human peer counselor-delivered iCBT. Furthermore, we will explore young adults' experiences with and perceptions of both intervention modalities to understand the potential advantages and disadvantages of AI-delivered iCBT compared to a human-delivered approach. 2. Material and Methods 2.1 Study Design and ethical approval A randomized controlled trial (August 20–October 30, 2024) compared AI-driven and peer counselor-delivered internet-based cognitive behavioral therapy (iCBT) for reducing depressive symptoms in young adults. Three parallel groups were used (Fig. 1): AI intervention, peer counselor intervention, and a no-intervention control. Participants received four weeks of tailored online intervention. The study was approved by the Biomedical Ethics Committee of Southern Medical University, China ([2023] No.53, September 2023) and registered with the Chinese Clinical Trial Registry (ChiCTR2400088423 on 19 August 2024) at https://www.chictr.org.cn/about.html. All participants provided informed consent and could withdraw at any time without affecting other participants. All data were anonymized, encrypted, and access-controlled. The reporting of this randomized controlled trial adheres to the Consolidated Standards of Reporting Trials (CONSORT) guidelines. 2.2 Participants and recruitment Participants were recruited online using a poster titled "AI Psychotherapy." Eligibility was assessed via an online screening questionnaire collecting basic demographics (e.g., age), willingness to complete follow-up and permit data export, and medical history (including severe physical/organic brain disease, substance abuse, recent severe depressive episodes/suicide attempts, and previous mental disorder diagnoses). Eligible participants completed pre-interview assessments. All collected data were anonymized, securely stored, and used solely for this study. Inclusion criteria were: (1) Patient Health Questionnaire-9 (PHQ-9) score between 5 and 20 (indicating mild to moderate depression); (2) aged 17–30 years; (3) proficient in using smartphones or tablets and communicating in Chinese; and (4) agreed to complete follow-up questionnaires and permit the export of therapy chat records. Exclusion criteria were: (1) presence of severe physical or organic brain diseases; (2) history of alcohol or substance abuse; (3) a severe depressive episode or suicide attempt within one month prior to the study; (4) current receipt of other psychological or medication treatments; or (5) a previous diagnosis of a severe mental disorder (e.g., schizophrenia, bipolar disorder) likely requiring different or more intensive interventions. 2.3 Randomization and Blinding A prospective, participant-blinded, parallel-group randomized controlled trial (RCT) compared AI, peer counselor, and control interventions. Participants were randomly assigned (computer-generated list, stratified by baseline PHQ-9 score) to one of three groups. Within the peer counselor group, counselors were randomly assigned to participants. To maintain consistent perception of the therapeutic modality, all intervention participants were informed their intervention was AI-driven. In the peer counselor group, this was achieved by omitting mention of human involvement in recruitment and providing briefings about the “AI system,” focusing participants on treatment sessions; non-treatment inquiries were addressed post-study. To further assess the effectiveness of this blinding strategy in the peer counselor group, a debriefing question was included in the post-intervention interview. Participants were asked, “During your interactions with the support you received, did you ever suspect that you were communicating with a real person rather than an AI system?” The responses to this question revealed that none of the participants in the peer counselor group reported suspecting that they were interacting with a human counselor. Blinding of peer counselors was not feasible due to the nature of peer support. Assessors (including interviewers) were blinded to group assignments using unique participant codes. This extended to all study personnel, preventing interactions that could reveal group allocation. Both active intervention groups received iCBT interventions with identical frequency and duration, differing only in delivery mode, controlling for non-specific therapeutic effects. 2.4 Intervention 2.4.1 AI Group In this study, the AI intervention group received support through a therapeutic chatbot named "Xiao Zhi." This chatbot was deployed on QQ, China's most popular instant messaging platform, and was directly powered by the GPT-3.5 large language model (LLM) architecture, ensuring a focus on the core language model technology. Participants in the AI group accessed and interacted with Xiao Zhi through the publicly available QQ application on various operating systems, including Windows, macOS, Android, and iOS. Participants engaged with Xiao Zhi via text-based input within the QQ interface, which the chatbot processed using natural language understanding technology to generate relevant responses. Mirroring human-delivered CBT, Xiao Zhi guided participants to identify and differentiate their emotions, thoughts, reactions, and behaviors, encouraging the development of more adaptive thought patterns. All conversation records with Xiao Zhi will be encrypted and stored separately from participant identifying information. Only authorized researchers will have access to this data under strict confidentiality agreements and solely for research purposes. A carefully constructed prompt engineering framework was developed to establish Xiao Zhi's role as a CBT-based psychological counselor. The key components of this framework include precise role definition, specific behavioral instructions, illustrative dialogues, comprehensive CBT guide integration, and system-level prompts. Specific content is detailed in the supplementary material: Prompt Engineering Framework for AI Chatbot (See Supplementary Material 2 for full details). 2.4.2 Human Group We selected peer counseling as the human intervention model for this study due to its recognized benefits in school mental health services. Peer counseling involves semi-professionally trained and supervised students providing listening, support, and counseling to their peers using verbal and nonverbal communication methods [15] . Research highlights the positive impacts of peer counseling, including effective psychological support, a relaxed atmosphere, reduced stigma, increased participation, and enhanced communication among peers [16,17]. To ensure high-quality peer counseling, we implemented a rigorous training and selection process. We recruited 30 peer counselor candidates majoring in applied psychology and psychiatry. Candidates underwent three months of intensive CBT skills training followed by two months of practical experience. We also provided specialized training to familiarize candidates with AI communication methods and enhance their text-based online intervention skills. We used standardized assessment criteria to evaluate candidates' ability to provide iCBT, selecting the top 10 students who met the criteria as peer counselors for the study. Throughout the study, a licensed psychotherapist with over 10 years of CBT experience provided ongoing supervision to all peer counselors to ensure adherence to iCBT protocol. More detailed information on the peer counselor training and supervision process is provided in Supplementary Material 3: Peer Counselor Training and Supervision Manual. Counseling records will be stored in encrypted form on secure servers or using encrypted paper documents kept in secure locations. Only authorized researchers will have access. 2.4.3 Control Group Participants in the control group were placed on a waitlist for the iCBT intervention. This waitlist control design allowed us to compare the effectiveness of both iCBT delivery methods (AI-driven and peer-counselor-delivered) to a baseline condition, accounting for the potential impact of time and nonspecific factors (e.g., natural recovery, time passage, repeated assessment effects, and the attention received from participating in the study) on symptom change.To further address ethical considerations, participants were instructed to report any engagement in other forms of psychological treatment during the study period to the research staff. This self-reporting allowed us to monitor potential confounding influences on the study outcomes. Additionally, should any participant experience a significant worsening of symptoms or express an urgent need for intervention during the study, they would be provided with referrals to appropriate mental health services. 2.4.4 CBT Principles and Guidelines In this study, both the Xiao Zhi chatbot and peer counselors strictly adhered to core Cognitive Behavioral Therapy (CBT) principles in their interventions, forming the theoretical foundation of our approach. These principles encompass cognitive restructuring (helping individuals identify and change irrational thoughts), behavioral activation (encouraging engagement in activities to improve mood and behavior), problem-solving (teaching effective methods and skills), and emotional regulation (helping individuals identify, understand, and manage emotions). Working together, these principles assist participants in coping with psychological distress from cognitive, behavioral, and emotional perspectives. Rather than adopting a single published guideline, a tailored set of CBT intervention guidelines was developed for this research, based on established CBT theory and rich clinical practice experience, and extensively referencing authoritative works [18, 19]. To ensure professionalism and applicability to online interventions, two psychology experts with doctoral degrees, advanced professional titles, extensive CBT clinical and research experience, specialized training, and relevant professional certifications participated in the guideline development process. Their combined expertise ensured the guidelines adhered to the theoretical framework of CBT and fully considered the unique characteristics of online delivery (e.g., clear/concise language, timely/effective responses, mitigating misunderstandings), thereby ensuring the intervention's safety and effectiveness. 2.5 Instruments Patient Health Questionnaire-9(PHQ-9) The PHQ-9 is a widely used, standardized screening tool designed to assess the severity of depressive symptoms [20] . It consists of nine items, each rated on a 4-point scale (0 = not at all, 1 = several days, 2 = more than half the days, 3 = nearly every day) based on the frequency of symptoms experienced over the past two weeks. The total score ranges from 0 to 27 and is used to categorize depression severity: none (0-4), mild (5-9), moderate (10-14), and severe (15 or greater). The PHQ-9 has demonstrated good internal consistency (α = 0.86-0.89) and criterion validity [20]. Multiple studies have validated the reliability and validity of the PHQ-9 in Chinese populations. For example, a systematic review by Yin et al. (2022) indicates that the Chinese version of the PHQ-9 demonstrates good psychometric properties across various populations, including acceptable internal consistency, good discriminant validity, and criterion validity. Beck Scale for Suicide Ideation-Chinese Version(BSI-CV) We used the Chinese version of the Beck Scale for Suicide Ideation (BSI-CV) to assess participants' current suicide ideation intensity. This 19-item scale uses a 3-point Likert response format (0-2) for each item, resulting in a total score ranging from 0 to 38. Higher BSI-CV scores indicate higher levels of current suicide ideation. The BSI-CV is a commonly used tool for suicide risk assessment and can identify individuals at elevated risk of suicide, aiding clinicians in risk assessment and intervention [22] . Liu et al. (2023) demonstrated good reliability and validity of the BSI-CV in Chinese adults. General Self-Efficacy Scale (GSES) This study used the General Self-Efficacy Scale (GSES) to assess participants' self-efficacy. The GSES is a widely used 10-item scale designed to assess individuals' general beliefs in their ability to cope with a broad range of life challenges [24] . Each item is rated on a 4-point scale ranging from 1 (not at all true) to 4 (exactly true). Sample items include: "I can manage to solve difficult problems if I try hard enough" and "If I am in trouble, I can usually think of a solution". The total score ranges from 10 to 40, with higher scores indicating higher levels of self-efficacy. The GSES has been widely used globally and validated in various cultural contexts, demonstrating good reliability and validity [25] . Self-compiled Participant Expectation Questionnaire (PEQ) To examine potential baseline differences in treatment expectations across groups, we developed a 22-item Participant Expectation Questionnaire. The questionnaire's design was informed by the bifactor model proposed by Sun Qiwu (citation), dividing the items into two dimensions: expectations of the counselor's professionalism and expectations of the participant's personal commitment. We also drew upon the Expectations About Counseling - Brief (EAC-B) scale (citation), adapting items to specifically measure participants' subjective expectations regarding AI-delivered psychological counseling. The first 20 items use a 5-point Likert scale (e.g., strongly disagree to strongly agree), while the final two items employ a 10-point scale (e.g., not at all likely to extremely likely). Higher scores across all items indicate higher overall expectations of psychological counseling. Total scores are calculated by summing the responses to all 22 items, with a maximum possible score of 160. This questionnaire served as an operational measure to compare baseline treatment expectations across the three groups, aiming to control for potential confounding effects of pre-existing expectations on intervention outcomes. It was not used as an outcome variable to assess changes in expectations following the intervention. Semi-Structured Interviews Grounded in the principles of grounded theory [26] , we conducted semi-structured interviews to explore the underlying mechanisms driving differences in intervention effects. The interview guide was developed iteratively, starting with a comprehensive literature review and initial data collection, followed by refinement based on preliminary investigations and expert consultation. The interviews centered around three key areas of inquiry: participants' experiences and psychological changes during the intervention, perceived similarities and differences between AI and human intervention, and preferences for future intervention modalities and the reasons behind those preferences. 2.6 Study Procedure Pilot Study Prior to the formal study, we conducted a pilot study with six participants (three in the AI intervention group and three in the human intervention group). Participants were informed that they were taking part in an experiment on AI psychological intervention. They then engaged in text-based counseling sessions and post-intervention interviews. The pilot study included an operational check to assess the effectiveness of the blinding procedures, specifically whether participants in the human intervention group believed their counselors were AI. Results confirmed that participants in the human intervention group did indeed perceive their counselors as AI. Formal Study Interested individuals first completed a screening questionnaire. Those meeting the inclusion criteria proceeded to the formal study, which consisted of a four-week intervention phase. Participants in the AI intervention group received iCBT from the chatbot "Xiao Zhi, " while those in the human intervention group received iCBT from peer counselors. Each participant engaged in two one-hour intervention sessions per week, totaling eight hours of intervention over the four weeks. We administered the PHQ-9, BSI-CV, and GSES scales at baseline, after two weeks, and again after four weeks to assess changes in depressive symptoms, suicidal ideation, and general self-efficacy. All three groups of participants received three measurements. Following the intervention, we conducted semi-structured interviews with all participants to explore their experiences and perceptions of the intervention process. Participants were asked to review their text-based counseling records prior to the interview. All interviews were audio-recorded for subsequent qualitative analysis(See Figure 1). 2.7 Debriefing Upon completion of the study, all participants attended a debriefing session. As a core ethical measure to address the deception inherent in the participant-blinding procedure, the research team fully explained the rationale for the study design, including the use of peer counselors posing as AI in the human intervention group. Participants had the opportunity to ask questions about the study design and share their feedback, thoughts, and feelings about their experiences. 2.8 Sample size and data analysis Sample size was calculated using G*Power 3.1 software, with a significance level (α) set at 0.05 and power (1-β) at 0.80. The effect size (f) was a key consideration. Previous research has shown that internet interventions (including iCBT) typically produce moderate to large effect sizes in improving depressive symptoms [8,20,27] . Furthermore, studies by Fitzpatrick et al. (2017) have also shown that AI-driven interventions can effectively provide mental health support. Based on these findings and referencing the typical range of effect sizes found in previous iCBT research, we hypothesized that AI interventions would have similar effects to human interventions, and ultimately set the effect size (f) at 0.30. Based on these parameters, a minimum of 75 participants were required. Considering a potential attrition rate of 10% and other unforeseen factors, we increased the recruitment target to 84 participants. We analyzed quantitative data using SPSS version 27. We compared baseline characteristics between the three groups using analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. To assess changes in depressive symptoms, suicidal ideation, and self-efficacy over time and between groups, we employed repeated measures ANOVA. We analyzed qualitative data from the semi-structured interviews using NVivo 12 Plus. We transcribed the audio recordings verbatim and employed an open coding approach to identify key themes and patterns. The initial codes were then categorized and integrated into axial codes, allowing us to compare and contrast the experiences and perceptions of participants in the AI and human intervention groups. 3. Results 3.1 Participant Characteristics and flow Of the 388 individuals who responded to recruitment materials, 90 met the eligibility criteria and were enrolled in the study. Of these, 87 participants completed the intervention, resulting in a dropout rate of 3. 3% (3 out of 90). The majority of participants were students (n = 79), with a small number employed (n = 5) or reporting other occupations (n = 3). Participants' ages ranged from 18 to 27 years (mean = 20. 46, standard deviation = 1. 94). 3.2 Baseline Analysis We conducted one-way ANOVAs to examine potential differences in baseline characteristics between the three study groups. As shown in Table 1, we found no statistically significant differences between groups in terms of age, gender ratio, depression level (PHQ-9), participant expectation (PEQ), suicide ideation (BSI-CV), or self-efficacy (GSES), indicating that the randomization process resulted in comparable baseline characteristics across the groups. These results support the validity of conducting subsequent analyses to compare the effects of the interventions (see Table 1). 3.3 Quantitative Analysis To examine the effects of the interventions on psychological well-being, repeated measures ANOVAs were conducted for each outcome variable (PHQ-9, BSI-CV, GSES). Time (baseline, second week, post-intervention) was treated as the within-subjects factor, and Group (AI, Human, Control) was the between-subjects factor. Greenhouse-Geisser corrections were applied when sphericity assumptions were violated. Bonferroni corrections were used for post-hoc pairwise comparisons. The results are presented in Tables 2 and 3. Primary outcome variable: Depression symptoms As shown in Table 2, there was a significant main effect of Time on depressive symptoms as measured by the PHQ-9, F = 28. 520, p <0. 001, partial η 2 = 0. 253. This indicates that depressive symptoms significantly decreased from baseline to both 2 weeks and 4 weeks across all groups. There was also a significant main effect of Group on depressive symptoms(Across three time points), F = 7. 088, p = 0. 001, partial η 2 = 0. 144, suggesting that the type of intervention (AI, Human, or Control) had a distinct impact on depressive symptoms. However, the significant Time x Group interaction effect, F = 5. 920, p <0. 001, partial η 2 = 0. 124, indicated that the rate of change in depressive symptoms over time differed between the intervention groups. To further investigate this interaction, we conducted a simple effects analysis to assess the effect of group at each time period. At baseline, there was no significant effect of Group on PHQ-9 scores, F = 1. 733, p = 0. 183, partial η 2 = 0. 040. Post hoc analyses with Bonferroni correction showed that there was no significant difference in depression levels at baseline (T0) among the three groups, suggesting that the data were comparable. At week two, there was a significant effect of Group on PHQ-9 scores, F = 7. 410, p = 0. 001, partial η 2 = 0. 150. Post hoc analyses with Bonferroni correction showed significant differences between the AI and control groups (d = -0. 86) and between the human and control groups ( d = -0. 76), however there were no significant differences between the AI and human groups ( d = -0. 12). At week four, there was a significant effect of Group on PHQ-9 scores, F = 10. 334, p < 0. 001, partial η 2 = 0. 197. Post hoc analyses with Bonferroni correction showed significant differences between the AI and control groups ( d = -0. 65) and between the human and control groups ( d = -1. 15), however there were no significant differences between the AI and human groups ( d = 0. 55)(See Table 3). Furthermore, the AI group showed a significant reduction in depression symptoms from baseline to the second week (T0 to T1, p < 0. 05), with no further significant change observed from the second week to post-intervention (T1 to T2). Conversely, the human intervention group demonstrated a continuous and significant downward trend in depression scores across all time points (T0, T1, and T2, p < 0. 05). Secondary outcome variables: Suicide ideation and self-efficacy As shown in Table 2, there was a significant main effect of Time on suicidal ideation as measured by the BCI-CV, F = 18. 830, p <0. 001, partial η 2 = 0. 183. This indicates that suicidal ideation significantly decreased from baseline to both 2 weeks and 4 weeks across all groups. There was also a significant main effect of Group on suicidal ideation(Across three time points), F = 4. 985, p = 0. 009, partial η 2 = 0. 106 , suggesting that the type of intervention (AI, Human, or Control) had a distinct impact on suicidal ideation. The significant Time x Group interaction effect, F = 3. 555, p =0. 012, partial η 2 = 0. 078, suggests that the rate of change in suicidal ideation differed among the intervention groups over time. Simple effects analyses were conducted to further examine this interaction. Unexpectedly, there was no significant effect of group on BCI-CV scores at baseline, week 2 and week 4. While no significant differences between groups were found, analysis of within-group changes highlighted important trends(See Table 3). For the AI group, there was a significant effect of Time on BCI-CV scores, F = 9. 797, p < 0. 001, partial η 2 = 0. 191, with a moderate decrease from baseline to 2 weeks ( d = 0. 48) and from baseline to 4 weeks ( d = 0. 74). The Human group also showed a significant effect of Time on BCI-CV scores, F = 7. 682, p =0. 001, partial η 2 = 0. 156, with a large reduction from baseline to 2 weeks ( d = 0. 46) and from baseline to 4 weeks ( d = 0. 71). In contrast, the Control group did not exhibit significant changes in BCI-CV scores over time, F =2. 324, p =0. 104, partial η 2 = 0. 053. There was a significant main effect of Time on self-efficacy as measured by the GSES, F = 29. 539, p < 0. 001, partial η 2 = 0. 260, indicating significant improvements in self-efficacy from baseline to both 2 weeks and 4 weeks across all groups. The Time x Group interaction effect was significant, F = 3. 670, p =0. 009, partial η 2 = 0. 080, indicated that the rate of change in depressive symptoms over time differed between the intervention groups. To further investigate this interaction, we conducted a simple effects analysis to assess the effect of group at each time period. At baseline, there was no significant main effect of Group on GSES scores, F = 1.217, p = 0.301, partial η 2 = 0.028, indicating that self-efficacy levels were comparable across the AI, Human, and Control groups. Similarly, at two weeks, the group effect was not significant, F = 0.533, p = 0.589, partial η 2 = 0.013. However, by week four, a significant main effect of Group emerged, F = 4.639, p = 0.012, partial η 2 = 0.099. Post hoc analyses with Bonferroni correction revealed a substantial increase in self-efficacy within the Human group compared to the Control group (d = 0.74). But there were no significant differences in self-efficacy between the Human and AI groups at any time point (See Table 3). Furthermore, we found that the AI group showed significant improvements in self-efficacy from baseline to post-intervention (T0 to T2, p < 0.001), specifically, the AI group showed significant improvements from baseline to two weeks post-intervention (T0 to T2, p < 0.001), but there was no significant difference between baseline and four weeks post-intervention. While the human intervention group also displayed an increase in self-efficacy from baseline to post-intervention, this change did not reach statistical significance. 3.4 Qualitative Analysis We conducted a qualitative analysis of the semi-structured interview transcripts using an open coding approach to identify key themes and patterns. From the initial codes, we developed three core categories representing the central themes emerging from the data: Intervention Process, Intervention Effects, and Intervention Prospects (See Table 4). Intervention Process: Advantages and Disadvantages Advantages: Participants in the AI intervention group highlighted several key advantages of AI-delivered iCBT, particularly its convenience and privacy. They appreciated the ability to access intervention regardless of time or location, saving both time and money. Many also felt that AI intervention alleviated privacy concerns, fostering a sense of safety and anonymity. This was especially true for those with social anxiety, who found the absence of a human counselor less intimidating, reducing stigma and encouraging self-disclosure. Participants in the human intervention group, on the other hand, valued the emotional interaction and nuanced understanding provided by their peer counselors, emphasizing the richer emotional experience. Disadvantages: Despite its advantages, AI-delivered iCBT also presented challenges. Participants noted difficulties the chatbot encountered in recognizing subtle emotional cues in language, limiting its ability to fully grasp emotional complexity and nuances. They felt that the AI, compared to human counselors, struggled to accurately perceive and validate their emotional dynamics, sometimes leading to formulaic responses or inappropriate questioning techniques that felt intrusive or "prying." Intervention Effects: Perceived Efficacy and Support While both the AI and human intervention groups acknowledged that the iCBT interventions provided practical suggestions and successfully initiated positive intervention outcomes, the qualitative data also highlighted the distinct difference in support depth. Both groups reported emotional support, but the human intervention was described as offering more profound validation and empathy, which is crucial for maintaining therapeutic motivation and sustained symptom improvement. Intervention Prospects: Future Preferences and Applications Participants expressed varying preferences for future modalities. While a significant number of participants were optimistic about the potential of AI-driven interventions, this preference stemmed primarily from its perceived advantages of personalized matching, cost-effectiveness, and a consistent process. Conversely, those who maintained a preference for human intervention highlighted the need for nuanced emotional understanding and flexibility—qualities they associated with human interaction, even when reflecting on the 'AI' support they received. 4. Discussion Current Findings This randomized controlled trial examined the comparative effectiveness of AI-delivered iCBT and human-delivered iCBT for reducing depressive symptoms and suicidal ideation among young adults. Importantly, our study sought to move beyond simply comparing efficacy by exploring users' experiences and perceptions of both intervention modalities. Our findings demonstrate that AI-delivered iCBT can significantly reduce depressive symptoms in young adults, aligning with previous research highlighting the therapeutic potential of AI in mental health interventions [28,29]. Moreover, we found that both AI and human-delivered iCBT led to significant reductions in suicidal ideation. These findings underscore the potential of both approaches in addressing these critical mental health concerns, particularly among young adults where suicide risk is a major concern. Crucially, the primary distinction between the intervention modalities emerged in the sustained therapeutic benefit. Our study revealed no significant differences in effectiveness between AI-delivered and human-therapist-delivered iCBT on any of the measured outcomes at week two. However, a key observation is that while the AI group showed significant improvement from baseline to week two in depression scores, this benefit did not continue from week two to week four, exhibiting a plateau effect. This stands in sharp contrast to the human intervention group, which demonstrated a continuous and significant downward trend in depression scores across all time points. This observation, even without resulting in a statistically significant difference between the two groups at week four, suggests that the human element possesses a unique, sustained advantage in maintaining therapeutic momentum and achieving more durable gains. Regarding self-efficacy, a significant time effect was observed, indicating a general trend of increasing scores from baseline to week two, followed by a decrease by week four in all groups. However, this trend was not differentially influenced by the intervention type, as no significant differences were found between the AI and Human groups at any assessment point. This observed pattern in self-efficacy, distinct from the plateau effect seen in depression scores, suggests that changes in self-efficacy may be attributable to factors other than the specific intervention modality itself. Therefore, the current findings offer limited insight into the specific impact of AI-delivered iCBT on self-efficacy, warranting further investigation with more refined methodologies and longer follow-up periods to disentangle the observed trend from potential confounding factors. Delving deeper into the qualitative data, we uncovered several key themes that shed light on the nuances of user experiences with AI-delivered iCBT. As anticipated, participants appreciated the advantages of AI, such as temporal and spatial flexibility and privacy. These benefits echo prior findings, suggesting that individuals, particularly those with social anxiety or introversion, may feel more comfortable disclosing personal information to AI systems [30–32]. The anonymity provided by AI can foster a sense of safety and reduce the stigma often associated with seeking mental health care, potentially leading to more honest self-disclosure and deeper engagement with the therapeutic process. However, our qualitative findings also revealed limitations in AI-delivered iCBT, particularly regarding the chatbot's capacity for understanding and responding to complex human language and applying therapeutic techniques with the same flexibility as a human therapist. Participants noted instances where the AI struggled to fully grasp emotional nuances and provide tailored responses, resulting in a perceived lack of personalized support. This limitation aligns with previous studies highlighting challenges in AI's ability to process and interpret complex human communication [10,33]. Participants reported that the AI often failed to recognize subtle shifts in emotional tone and complex semantic layers, demonstrating a significant weakness in clinical intuition and judgment. Its responses were sometimes perceived as rigid, formulaic, and poorly timed, especially during deeper explorations involving core beliefs. This tendency towards mechanical application of techniques, rather than flexible, personalized support, led some participants to feel unheard or misunderstood. This resultant loss of engagement and therapeutic momentum is hypothesized to directly contribute to the observed plateau effect in the AI group's depression scores, contrasting sharply with the sustained improvement demonstrated by the human group. Interestingly, despite participants in the human group believing they were interacting with an AI, this group showed sustained improvement. This suggests that the human counselors' superior ability to interpret complex emotional context and provide nuanced, non-repetitive responses—which is a function of advanced cognitive and empathetic ability—rather than the explicit awareness of a human relationship, may be the key driver for long-term therapeutic gains. The current AI, while able to generate basic empathetic responses, hit a 'cognitive ceiling' when attempting to fully address deeper core beliefs. This superiority is rooted in the human capacity for nuanced interpretation (i.e., listening for the nuance or 'between-the-lines' meaning) and providing non-formulaic, adaptive engagement necessary for deep cognitive and emotional restructuring. Furthermore, the analysis of future preferences revealed a practical functional distinction: participants viewed AI as suitable for accessible, short-term support and initial symptom management, while human counselors were deemed essential for addressing complex, long-term psychological issues requiring genuine relational interaction. Future research and Limitations Despite these findings, our study has limitations that warrant consideration. First, our sample primarily comprised young adults, potentially limiting the generalizability of our findings to other age groups. Future research should encompass more diverse samples to determine if AI-delivered iCBT demonstrates comparable effectiveness across different demographics and mental health needs. Second, while we assessed emotional responses through self-report measures and qualitative interviews, our study design relied solely on text-based communication. This approach, while aligning with the nature of many digital mental health interventions, limits our understanding of how AI might perform in interventions incorporating audio or visual cues. Future studies incorporating multimodal communication channels could provide richer insights into the nuances of AI-driven empathy and emotional support. Third, our study focused on short-term outcomes. Exploring the long-term efficacy and sustainability of AI-delivered iCBT, compared to human-delivered approaches, is crucial for determining its potential as a scalable and sustainable mental health solution. Finally, our use of a waitlist control group has limitations. While it allowed us to assess the overall effectiveness of both interventions compared to no intervention, it may not have fully controlled for non-specific effects associated with engaging with an online platform and the expectation of receiving support. Future research could consider employing an active control group (e.g., an attention-matched control group receiving a non-therapeutic online application) to better isolate the specific therapeutic effects of the AI and human-delivered iCBT. Our study focused on a single AI chatbot utilizing a specific iCBT program. Future research should explore users' experiences with various AI models and therapeutic approaches to determine the generalizability of our findings and identify optimal combinations of AI technology and therapeutic techniques. 5. Conclusion In conclusion, our findings indicate that AI-delivered iCBT can be a promising short-term tool for addressing depression and suicidal ideation in young adults, demonstrating comparable effectiveness to human-therapist-delivered iCBT only in the initial stages of intervention. The key observation, however, is the plateau effect observed in the AI group's depression scores, which suggests that complex, sustained therapeutic work requires the unique, adaptable engagement and relational depth of a human counselor. While AI's limitations in emotional perception and nuanced language understanding require further refinement, its accessibility and adaptability highlight its potential to bridge critical gaps in mental healthcare as a scalable supplementary or short-term intervention. Future research should prioritize long-term outcomes and exploration of various AI models and intervention approaches to further advance the development and implementation of effective and accessible AI-driven mental health solutions. Declarations Ethics approval and consent to participate The study involving human participants was approved by the Biomedical Ethics Committee of Southern Medical University, China (Reference Number: [2023] No. 53). The study procedures were conducted in accordance with the ethical standards of the responsible committee and with the Declaration of Helsinki. All participants provided written informed consent prior to participation and were informed of their right to withdraw at any time without penalty. Consent for publication Not applicable. (This manuscript does not contain any individual person’s data, image, or video in any form.) Availability of data and materials The data underlying the findings of this study contain sensitive patient information and cannot be publicly shared due to privacy regulations. Anonymized summary statistics are available upon reasonable request from the corresponding author. Competing interests The authors declare that they have no competing interests. Funding This work was supported by the Guangdong Provincial Undergraduate Teaching Quality and Teaching Reform Project (Yue Jiao Gao Han [2024] No. 30) and the Ministry of Education of China, Humanities and Social Sciences Research Project (25YJA190016). Authors' contributions Yiyang Wu : Conceptualization, Data curation, Formal analysis, Software, Supervision, Validation, Visualization, Writing - original draft. Haoran Song: Conceptualization, Data curation, Formal analysis, Project administration, Investigation, Software, Supervision. Weihao Huang: Formal analysis, Investigation, Software, Validation, Visualization. Chen Ye: Investigation, Software. Ruoyu Lin: Investigation, Software.You Wang: Funding acquisition, Resources. Xueling Yang: Conceptualization, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing - review and editing. Acknowledgement We would like to thank the participants and research assistants for their contributions to this study. The intervention team consisted of the following: Xueling Yang (Supervisor), Chenhua Wang (Clinician), Yeya Lu (Clinician), Ruoxi Li (Peer Counselor), Wenxi Huang (Peer Counselor), Xiaoqiao Lu (Peer Counselor), Jiatong Liang (Peer Counselor), Yun Shi Yao (Peer Counselor), Guangying He (Peer Counselor), Zijun Wang (Peer Counselor), Chun Zhan (Peer Counselor), Wenxuan Zhou (Peer Counselor), and Xiaorong Lin (Peer Counselor). Statement During the preparation of this work, the author(s) used Google AI Studio's Gemini model to refine the language and improve the clarity and readability of the manuscript. After using this tool, the author(s) carefully reviewed and edited the content as needed and take(s) full responsibility for the content of the published article. References Smith K. Mental health: A world of depression. Nature 2014;515:180–1. https://doi.org/10.1038/515180a. Shorey S, Ng ED, Wong CHJ. Global prevalence of depression and elevated depressive symptoms among adolescents: A systematic review and meta-analysis. Br J Clin Psychol 2022;61:287–305. https://doi.org/10.1111/bjc.12333. Hofmann SG, Asnaani A, Vonk IJ, et al. The efficacy of cognitive behavioral therapy: A review of meta-analyses. Cogn Ther Res 2012;36:427–40. Luo C, Sanger N, Singhal N, et al. A comparison of electronically-delivered and face to face cognitive behavioural therapies in depressive disorders: A systematic review and meta-analysis. eClinicalMedicine 2020;24:100442. https://doi.org/10.1016/j.eclinm.2020.100442. Hedman E, Ljótsson B, Lindefors N. Cognitive behavior therapy via the Internet: a systematic review of applications, clinical efficacy and cost–effectiveness. Expert Rev Pharmacoecon Outcomes Res 2012;12:745–64. https://doi.org/10.1586/erp.12.67. Muse K, McManus F. A systematic review of methods for assessing competence in cognitive–behavioural therapy. Clin Psychol Rev 2013;33:484–99. https://doi.org/10.1016/j.cpr.2013.01.010. Loughnan SA, Newby JM, Haskelberg H, et al. Internet-based cognitive behavioural therapy (iCBT) for perinatal anxiety and depression versus treatment as usual: study protocol for two randomised controlled trials. Trials 2018;19:56. https://doi.org/10.1186/s13063-017-2422-5. Karyotaki E, Efthimiou O, Miguel C, et al. Internet-based cognitive behavioral therapy for depression: a systematic review and individual patient data network meta-analysis. JAMA Psychiatry 2021;78:361–71. Fitzpatrick KK, Darcy A, Vierhile M. Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Ment Health 2017;4:e19. https://doi.org/10.2196/mental.7785. Green EP, Lai Y, Pearson N, et al. Expanding Access to Perinatal Depression Treatment in Kenya Through Automated Psychological Support: Development and Usability Study. JMIR Form Res 2020;4:e17895. https://doi.org/10.2196/17895. Lan J. The historical mission of generative artificial intelligence and humanities and social sciences - starting from the ChatGPT intelligent revolution Ideological and Theoretical Education. Ideol Theor Educ 2023:12–8. https://doi.org/10.16075/j.cnki.cn31-1220/g4.2023.04.012. Liu J. ChatGPT: perspectives from human–computer interaction and psychology. Front Artif Intell 2024;7:1418869. https://doi.org/10.3389/frai.2024.1418869. Dave T, Athaluri SA, Singh S. ChatGPT in medicine: an overview of its applications, advantages, limitations, future prospects, and ethical considerations. Front Artif Intell 2023;6:1169595. https://doi.org/10.3389/frai.2023.1169595. Eysenbach G. The Role of ChatGPT, Generative Language Models, and Artificial Intelligence in Medical Education: A Conversation With ChatGPT and a Call for Papers. JMIR Med Educ 2023;9:e46885. https://doi.org/10.2196/46885. Sussman MB. The development and effects of a model for training peer-group counselors in a multi-ethnic junior high school. University of Miami; 1973. Tindall JA. Peer counseling: An in-depth look at training peer helpers. ERIC; 1989. Winterton CI, Dunk RDP, Wiles JR. Peer-led team learning for introductory biology: relationships between peer-leader relatability, perceived role model status, and the potential influences of these variables on student learning gains. Discip Interdiscip Sci Educ Res 2020;2:3. https://doi.org/10.1186/s43031-020-00020-9. Beck JS, Beck A, Beck J. Cognitive behavior therapy: basics and beyond. ed. N Y 2011. David B, Burns M. Feeling Good: The New Mood Therapy. NY Signet Books Chin Richard 1980:42–3. Karyotaki E, Furukawa TA, Efthimiou O, et al. Guided or self-guided internet-based cognitive–behavioural therapy (iCBT) for depression? Study protocol of an individual participant data network meta-analysis. BMJ Open 2019;9:e026820. Yin L, Teklu S, Pham H, et al. Validity of the Chinese Language Patient Health Questionnaire 2 and 9: A Systematic Review. Health Equity 2022;6:574–94. https://doi.org/10.1089/heq.2022.0030. Beck AT, Kovacs M, Weissman A. Assessment of suicidal intention: the Scale for Suicide Ideation. J Consult Clin Psychol 1979;47:343. Lin Z, Cheng L, Han X, et al. The Effect of Internet-Based Cognitive Behavioral Therapy on Major Depressive Disorder: Randomized Controlled Trial. J Med Internet Res 2023;25:e42786. https://doi.org/10.2196/42786. Schwarzer R. Generalized self-efficacy scale. Meas Health Psychol User’s Portf Causal Control BeliefsNfer-Nelson 1995. Luszczynska A, Scholz U, Schwarzer R. The General Self-Efficacy Scale: Multicultural Validation Studies. J Psychol 2005;139:439–57. https://doi.org/10.3200/JRLP.139.5.439-457. Glaser B, Strauss A. Discovery of grounded theory: Strategies for qualitative research. Routledge; 2017. Cuijpers P, Noma H, Karyotaki E, et al. Effectiveness and acceptability of cognitive behavior therapy delivery formats in adults with depression: a network meta-analysis. JAMA Psychiatry 2019;76:700–7. Donkin L, Hickie IB, Christensen H, et al. Rethinking the Dose-Response Relationship Between Usage and Outcome in an Online Intervention for Depression: Randomized Controlled Trial. J Med Internet Res 2013;15:e231. https://doi.org/10.2196/jmir.2771. Liu H, Peng H, Song X, et al. Using AI chatbots to provide self-help depression interventions for university students: A randomized trial of effectiveness. Internet Interv 2022;27:100495. https://doi.org/10.1016/j.invent.2022.100495. DARWIN C. The expression of the emotions in man and animals (1872). Portable Darwin 1993:364–93. Quigley L, Dozois DJA, Bagby RM, et al. Cognitive change in cognitive-behavioural therapy v. pharmacotherapy for adult depression: a longitudinal mediation analysis. Psychol Med 2019;49:2626–34. https://doi.org/10.1017/S0033291718003653. Torous J, Staples P, Shanahan M, et al. Utilizing a Personal Smartphone Custom App to Assess the Patient Health Questionnaire-9 (PHQ-9) Depressive Symptoms in Patients With Major Depressive Disorder. JMIR Ment Health 2015;2:e8. https://doi.org/10.2196/mental.3889. Pintelas EG, Kotsilieris T, Livieris IE, et al. A review of machine learning prediction methods for anxiety disorders, 2018, p. 8–15. Tables Table 1 Baseline characteristics of the participants; mean (SD) AI group Human group Control group F/c2 p Age, mean (SD) 20. 37 (1. 83) 20. 41 (2. 39) 20. 43 (1. 52) 0. 120 0. 887 Gender, n (%) Male 8 (26. 67) 11 (40. 74) 6 (0. 20) 3. 081 0. 214 Female 22 (73. 33) 16 (59. 26) 24 (0. 80) Scale, mean (SD) PHQ-9 9. 37(3. 49) 9. 78(3. 08) 11. 17(4. 87) 1. 733 0. 183 BSI-CV 4. 90(3. 06) 3. 93(3. 27) 3. 43(3. 05) 1. 712 0. 187 GSES 19. 80(6. 21) 22. 04(6. 08) 19. 87(6. 00) 1. 217 0. 301 PEQ 120. 00(23. 22) 127. 70(16. 95) 2. 008 0. 162 Table 2 Effect of intervention on primary and secondary outcomes after 2 and 4 weeks Outcome Group Baseline (T0) 2 weeks (T1) 4 weeks (T2) F p Partial η2 PHQ-9 AI 9. 37(3. 49) 7. 00(3. 22) 7. 13(4. 18) Human 9. 78(3. 08) 7. 37(3. 20) 5. 07(3. 03) Control 11. 17(4. 86) 10. 40(4. 56) 10. 40(5. 69) Time Effect 28. 520 <0. 001 0. 253 Group 7. 088 0. 001 0. 144 Interaction 5. 920 <0. 001 0. 124 BCI-CV AI 4. 90(3. 06) 3. 47(2. 80) 2. 80(2. 52) 3. 555 0. 012 Human 3. 93(3. 27) 2. 59(2. 50) 1. 96(2. 18) Control 3. 43(3. 05) 2. 77(3. 00) 3. 30(3. 46) Time Effect 18. 830 <0. 001 0. 183 Group 4. 985 0. 009 0. 106 Interaction 3. 555 0. 012 0. 078 GSES AI 19. 80(6. 21) 26. 03(4. 41) 23. 73(6. 74) 37. 128 <0. 001 Human 22. 04(6. 08) 25. 11(5. 36) 24. 44(7. 08) Control 19. 87(6. 00) 24. 97(2. 98) 19. 57(6. 03) Time Effect 29. 539 <0. 001 0. 260 Group 2. 029 0. 138 0. 046 Interaction 3. 670 0. 009 0. 080 Abbreviations: Degrees of freedom are Greenhouse-Geisser corrected. AI, AI intervention group; Human, Human intervention group; Control, control group with no intervention. Note. T0 = baseline; T1 = week 2; T2 = week 4. Table 3 The Effects of Different Interventions on Depression, Suicidal Ideation, and Self-Efficacy: Intergroup Comparisons Baseline(T0) 2 weeks(T1) 4weeks(T2) Outcome Contrast MD Cohen'd 95% CI MD Cohen'd 95% CI MD Cohen'd 95% CI PHQ-9 AI vs. Human -0. 41 -0. 12 -2. 95, 2. 12 -0. 37 -0. 12 -2. 79, 2. 05 2. 06 0. 55 -0. 84, 4. 96 AI vs. Control -1. 80 -0. 42 -4. 27, 0. 67 -3. 47 -0. 86 -5. 75,- 1. 05 -3. 27 -0. 65 -6. 09, -0. 44 Human vs. Control -1. 39 -0. 33 -3. 92, 1. 15 -3. 10 -0. 76 -5. 45,- 0. 61 -5. 33 -1. 15 -8. 23, -2. 43 BCI-CV AI vs. Human 0. 974 0. 31 -1. 05, 3. 00 0. 874 0. 33 -0. 93, 2. 68 0. 837 0. 35 -0. 97, 2. 65 AI vs. Control 1. 467 0. 48 -0. 50, 3. 44 0. 700 0. 24 -1. 06, 2. 46 -0. 500 -0. 17 -2. 26, 1. 26 Human vs. Control 0. 493 0. 16 -1. 53, 2. 52 -0. 174 -0. 06 -1. 98, 1. 63 -1. 337 -0. 45 -3. 15, 0. 47 GSES AI vs. Human -2. 237 -0. 36 -6. 19, 1. 71 0. 922 0. 19 -1. 88, 3. 72 -0. 711 -0. 10 -5. 00, 3. 58 AI vs. Control -0. 067 -0. 01 -3. 91, 3. 78 1. 067 0. 28 -1. 66, 3. 79 4. 167 0. 65 -0. 01, 8. 34 Human vs. Control 2. 170 0. 36 -1. 78, 6. 12 0. 144 0. 03 -2. 65, 2. 94 4. 878 0. 74 0. 59, 9. 16 Abbreviations: MD, mean difference; AI, AI intervention group; Human, Human intervention group; Control, control group with no intervention. Note. T0 = baseline; T1 = week 2; T2 = week 4. Table 4 Qualitative Themes and Subthemes: AI Intervention vs. Human Intervention Theme Subtheme AI group c Human group Concept Process (Advantages) A1: Reduced time/location constraints 8 7 Cost-effective, saves time A2: Reduced privacy concerns 15 9 Safe environment, increased self-disclosure A3: Reduced stigma 14 10 Reduced psychological burden A4: Sense of interaction 11 16 Provides an outlet for sharing/feeling heard A5: Broad information resources 11 0 Extensive information coverage A6: Comprehension ability 15 14 Flexible counseling direction/captures semantics and key issues (Disadvantages) A7: Weak empathy 17 16 Limited ability to understand circumstances, identify emotional fluctuations, and provide emotional validation A8: Idealistic advice 11 8 Impractical suggestions A9: Weak recognition ability 5 0 Limited capacity to process language and perceive emotions A10: Mechanistic language 15 21 Inflexible and impersonal tone and response style A11: Weak sense of communication 34 53 Formulaic counseling style, poor flow, slow response time A12: Unproductive probing 3 1 Lack of direction in guidance Effectiveness A13: Providing advice 25 20 Offers suggestions for practical problems A14: Emotional support 8 18 Understands client's perspective and improves mood A15: Thought-provoking 18 30 Identifies irrational beliefs and introduces new possibilities A16: Achieved desired outcomes 21 20 Met expectations for counseling outcomes A18: Partially helpful 9 8 Only addressed surface-level issues Future A19: Preference for AI 18 15 Personalized matching, accessibility, stability A20: Preference for human 9 6 Professionalism, emotional needs met, rich content exploration A21: Application and promotion 4 9 widespread adoption and dissemination A22: Scope of application 18 30 Suitable for mild to moderate problems, short-term interventions Note. c The number represents the amount of material with consistent content obtained from node materials of different texts. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 24 Feb, 2026 Read the published version in BMC Psychiatry → Version 1 posted Editorial decision: Revision requested 05 Feb, 2026 Reviews received at journal 02 Feb, 2026 Reviews received at journal 28 Jan, 2026 Reviewers agreed at journal 30 Dec, 2025 Reviewers agreed at journal 28 Dec, 2025 Reviewers agreed at journal 27 Dec, 2025 Reviewers agreed at journal 26 Dec, 2025 Reviewers invited by journal 25 Dec, 2025 Editor assigned by journal 25 Dec, 2025 Editor invited by journal 10 Dec, 2025 Submission checks completed at journal 10 Dec, 2025 First submitted to journal 10 Dec, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8269763","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":566458600,"identity":"f4dc2707-b93a-4a42-9278-ae1f17699c31","order_by":0,"name":"Yiyang Wu","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yiyang","middleName":"","lastName":"Wu","suffix":""},{"id":566458601,"identity":"1222221d-2613-43cf-951f-5459b1ba0164","order_by":1,"name":"Haoran Song","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Haoran","middleName":"","lastName":"Song","suffix":""},{"id":566458602,"identity":"6897ec62-9e22-497f-8a1d-7caa0f139abe","order_by":2,"name":"Chen Ye","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Chen","middleName":"","lastName":"Ye","suffix":""},{"id":566458603,"identity":"fff79623-bfca-47de-b441-b3e67764b375","order_by":3,"name":"Ruoyu Lin","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ruoyu","middleName":"","lastName":"Lin","suffix":""},{"id":566458605,"identity":"7ea6343c-da8e-4e0c-a4d9-55af96c0e94c","order_by":4,"name":"Weihao Huang","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Weihao","middleName":"","lastName":"Huang","suffix":""},{"id":566458606,"identity":"dededf0a-a8b2-43a3-90d5-c7e8ae7e2930","order_by":5,"name":"You Wang","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"You","middleName":"","lastName":"Wang","suffix":""},{"id":566458609,"identity":"76148a4a-da72-473a-aa4c-a90dd94d2842","order_by":6,"name":"Xueling Yang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3UlEQVRIiWNgGAWjYDACCRBRYMPDxt584MCHCqK0MAMJgzQ5fp5jiQdnnCFey2FjyRk+xod5W4jQYXC7/5g0jwFz4oYbPB8O8DYwyPOLHSCg5c5hNskZBmyJG273bjgguYPBcObsBAJabiSzSXww4EnccOfshgOGZxgSDG4ToyXBQALosJwHBxLbiNXywcAA6P0chgMHidEieSPZ2HKGQQIokA0ONpyRIOwXvhuJD2/zVPwHReXjz38qbOT5pQloUTjAwCKBxJfAqRIO5BsYmD8QVjYKRsEoGAUjGgAADjlL4eQVkAAAAAAASUVORK5CYII=","orcid":"","institution":"Southern Medical University","correspondingAuthor":true,"prefix":"","firstName":"Xueling","middleName":"","lastName":"Yang","suffix":""}],"badges":[],"createdAt":"2025-12-03 11:38:22","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8269763/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8269763/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12888-026-07925-1","type":"published","date":"2026-02-24T15:57:58+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":99222832,"identity":"2ac8d217-6453-4400-b1e3-77e7eaea58c4","added_by":"auto","created_at":"2025-12-30 09:57:04","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":79622,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/a4e06febd182d469e50f6ade.docx"},{"id":99222835,"identity":"36f88acc-4f0f-48ca-a181-4e31d8ac1ae9","added_by":"auto","created_at":"2025-12-30 09:57:04","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9066,"visible":true,"origin":"","legend":"","description":"","filename":"580007211c304e4fa988e2d688a2a97b.json","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/67ec4b481f0bc794177e9b0d.json"},{"id":99222833,"identity":"84cdbd83-4b71-4b67-a609-b183c0e7ee8d","added_by":"auto","created_at":"2025-12-30 09:57:04","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":146767,"visible":true,"origin":"","legend":"","description":"","filename":"580007211c304e4fa988e2d688a2a97b1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/dc8b88d5782cc04546870ea8.xml"},{"id":99222836,"identity":"a4748544-3a78-4f06-9703-c47ff2f5276a","added_by":"auto","created_at":"2025-12-30 09:57:04","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139742,"visible":true,"origin":"","legend":"","description":"","filename":"580007211c304e4fa988e2d688a2a97b1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/dd430d6de5132b24d07b0854.xml"},{"id":99318852,"identity":"cb326b4e-3248-44c9-8140-479521d8eec6","added_by":"auto","created_at":"2025-12-31 16:35:26","extension":"html","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":159777,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/f7755b491be4ba2c0efef755.html"},{"id":99222831,"identity":"12dc0b96-1131-4203-8c85-6b5b92c4f5f9","added_by":"auto","created_at":"2025-12-30 09:57:04","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":5713,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/fb9027409816ff7882edbe76.png"},{"id":103765490,"identity":"cd0977c5-efdb-4da3-b1ac-624d589a7944","added_by":"auto","created_at":"2026-03-02 16:03:15","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1243192,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8269763/v1/d0c82cb7-f1f3-49e0-aaa0-cea7cec3f42c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Pilot Randomized Controlled Trial of AI-Delivered vs. Human-delivered iCBT for Depression in Young Adults","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eDepression, characterized by persistent low mood, anhedonia, psychomotor retardation, and in severe cases, suicidal ideation\u0026nbsp;[1], represents a substantial global health challenge. The self-reported prevalence of depressive symptoms has risen steadily between 2001 and 2020, reaching 34%\u0026nbsp;[2], suggesting that over one-third of the global population may experience depressive episodes. Psychotherapy, particularly cognitive behavioral therapy (CBT), has demonstrated efficacy in treating various mental health conditions, including depression\u0026nbsp;[3]. However, access to qualified therapists is often limited, and traditional CBT typically involves extended treatment durations and considerable costs.\u0026nbsp;These limitations are compounded by patient-level barriers, such as stigma surrounding mental illness and reluctance to seek help, further restricting access to and the overall effectiveness of psychological car [4].\u003c/p\u003e\n\u003cp\u003eAlthough effective, traditional CBT for depression presents practical challenges, including lengthy treatment periods, complex therapeutic processes, and substantial costs, prompting the development of internet-based CBT (iCBT)\u0026nbsp;[5].\u0026nbsp;iCBT delivers CBT modules through online platforms using text, videos, or interactive exercises, offering advantages in convenience, cost-effectiveness, and accessibility, and has demonstrated therapeutic benefits in various settings\u0026nbsp;[6]. Randomized controlled trials (RCTs) have shown the effectiveness of iCBT in reducing depressive symptoms among adolescents through diverse formats, including online modules combined with family support and therapist guidance\u0026nbsp;[7]. Moreover, research indicates that guided iCBT, incorporating therapist support, can be particularly beneficial for individuals with moderate to severe depression\u0026nbsp;[8]. A large individual patient data network meta-analysis by\u0026nbsp;Karyotaki et al. (2021)\u0026nbsp;found that while both guided and unguided iCBT were effective compared to control conditions, guided iCBT yielded greater short-term benefits for individuals with higher baseline depression scores. This underscores the potential for tailoring iCBT interventions to individual needs and symptom severity. However, iCBT is not without limitations. Notably, the complexity of certain online programs can lead to reduced treatment adherence due to patient avoidance or resistance\u0026nbsp;[7].\u003c/p\u003e\n\u003cp\u003eThe rapid advancement of artificial intelligence (AI) presents a promising avenue for addressing the limitations of traditional iCBT. Early studies have explored the potential of AI-based interventions in providing mental health support. For instance,\u0026nbsp;Fitzpatrick et al. (2017)\u0026nbsp;demonstrated the effectiveness of a fully automated conversational agent (Woebot) in reducing depressive symptoms among young adults. Similarly,\u0026nbsp;Green et al. (2020)\u0026nbsp;developed and tested a mobile application, \u0026quot;Healthy Moms,\u0026quot; designed to provide automated psychological support for perinatal depression in Kenya. These preliminary findings suggest the feasibility, acceptability, and potential efficacy of AI-driven interventions, particularly in resource-constrained settings where they can expand access to mental health services. These early systems often relied on pre-defined rules and limited natural language processing capabilities. However, recent breakthroughs in deep learning, particularly the development of large language models (LLMs), have revolutionized the potential of AI in mental health. LLMs, such as OpenAI\u0026apos;s ChatGPT, are trained on massive datasets of text and code, enabling them to exhibit sophisticated natural language understanding, generation, and dialogue capabilities\u0026nbsp;[11]. This capacity allows for more natural and engaging user interactions, potentially overcoming the challenges of program complexity and limited personalization observed in earlier iCBT approaches. Integrating the core principles of CBT into LLM-driven systems, delivered through user-friendly iCBT platforms, thus offers a compelling opportunity to enhance the therapeutic experience and improve intervention outcomes.\u003c/p\u003e\n\u003cp\u003eLLMs, as a prominent subset of AI, possess several key advantages that can mitigate the limitations of both traditional CBT and earlier forms of iCBT. First, LLMs can deliver CBT interventions with high fidelity and consistency, ensuring standardized, evidence-based care and minimizing variability inherent in human-delivered therapy. Second, LLMs offer highly personalized and interactive experiences, adapting to individual user needs, preferences, and real-time progress. This dynamic adaptation can enhance user engagement and adherence, a recognized challenge in iCBT\u0026nbsp;[12]. Third, LLMs can provide immediate and readily available 24/7 support, overcoming geographical barriers and scheduling constraints that often limit access to traditional therapy\u0026nbsp;[13]. Furthermore, the capacity of LLMs to process and analyze vast amounts of data enables continuous learning and refinement of the intervention, potentially leading to more effective and efficient therapeutic outcomes over time\u0026nbsp;[14]. These advancements provide a robust theoretical foundation for exploring iCBT powered by advanced LLMs, such as ChatGPT, as a potentially effective and scalable intervention for depression.\u003c/p\u003e\n\u003cp\u003eBased on these theoretical considerations and empirical findings, this randomized controlled trial aims to compare the effectiveness of AI-driven (ChatGPT-powered) iCBT and human peer counselor-delivered iCBT in reducing depressive symptoms and suicidal ideation among young adults. We hypothesize that AI-driven iCBT will demonstrate efficacy in reducing depressive symptoms and suicidal ideation, and we will explore its comparative effectiveness relative to human peer counselor-delivered iCBT. Furthermore, we will explore young adults\u0026apos; experiences with and perceptions of both intervention modalities to understand the potential advantages and disadvantages of AI-delivered iCBT compared to a human-delivered approach.\u003c/p\u003e"},{"header":"2. Material and Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 Study Design and ethical approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA randomized controlled trial (August 20\u0026ndash;October 30, 2024) compared AI-driven and peer counselor-delivered internet-based cognitive behavioral therapy (iCBT) for reducing depressive symptoms in young adults. Three parallel groups were used (Fig. 1): AI intervention, peer counselor intervention, and a no-intervention control. Participants received four weeks of tailored online intervention. The study was approved by the Biomedical Ethics Committee of Southern Medical University, China ([2023] No.53, September 2023) and registered with the Chinese Clinical Trial Registry (ChiCTR2400088423 on 19 August 2024) at https://www.chictr.org.cn/about.html. All participants provided informed consent and could withdraw at any time without affecting other participants. All data were anonymized, encrypted, and access-controlled. The reporting of this randomized controlled trial adheres to the Consolidated Standards of Reporting Trials (CONSORT) guidelines.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 Participants and recruitment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants were recruited online using a poster titled \u0026quot;AI Psychotherapy.\u0026quot; Eligibility was assessed via an online screening questionnaire collecting basic demographics (e.g., age), willingness to complete follow-up and permit data export, and medical history (including severe physical/organic brain disease, substance abuse, recent severe depressive episodes/suicide attempts, and previous mental disorder diagnoses). Eligible participants completed pre-interview assessments. All collected data were anonymized, securely stored, and used solely for this study.\u003c/p\u003e\n\u003cp\u003eInclusion criteria were: (1) Patient Health Questionnaire-9 (PHQ-9) score between 5 and 20 (indicating mild to moderate depression); (2) aged 17\u0026ndash;30 years; (3) proficient in using smartphones or tablets and communicating in Chinese; and (4) agreed to complete follow-up questionnaires and permit the export of therapy chat records.\u003c/p\u003e\n\u003cp\u003eExclusion criteria were: (1) presence of severe physical or organic brain diseases; (2) history of alcohol or substance abuse; (3) a severe depressive episode or suicide attempt within one month prior to the study; (4) current receipt of other psychological or medication treatments; or (5) a previous diagnosis of a severe mental disorder (e.g., schizophrenia, bipolar disorder) likely requiring different or more intensive interventions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Randomization and Blinding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA prospective, participant-blinded, parallel-group randomized controlled trial (RCT) compared AI, peer counselor, and control interventions. Participants were randomly assigned (computer-generated list, stratified by baseline PHQ-9 score) to one of three groups. Within the peer counselor group, counselors were randomly assigned to participants.\u003c/p\u003e\n\u003cp\u003eTo maintain consistent perception of the therapeutic modality, all intervention participants were informed their intervention was AI-driven. In the peer counselor group, this was achieved by omitting mention of human involvement in recruitment and providing briefings about the \u0026ldquo;AI system,\u0026rdquo; focusing participants on treatment sessions; non-treatment inquiries were addressed post-study. To further assess the effectiveness of this blinding strategy in the peer counselor group, a debriefing question was included in the post-intervention interview. Participants were asked, \u0026ldquo;During your interactions with the support you received, did you ever suspect that you were communicating with a real person rather than an AI system?\u0026rdquo; The responses to this question revealed that none of the participants in the peer counselor group reported suspecting that they were interacting with a human counselor.\u003c/p\u003e\n\u003cp\u003eBlinding of peer counselors was not feasible due to the nature of peer support. Assessors (including interviewers) were blinded to group assignments using unique participant codes. This extended to all study personnel, preventing interactions that could reveal group allocation. Both active intervention groups received iCBT interventions with identical frequency and duration, differing only in delivery mode, controlling for non-specific therapeutic effects.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 Intervention\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e2.4.1 AI Group\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, the AI intervention group received support through a therapeutic chatbot named \u0026quot;Xiao Zhi.\u0026quot; This chatbot was deployed on QQ, China\u0026apos;s most popular instant messaging platform, and was directly powered by the GPT-3.5 large language model (LLM) architecture, ensuring a focus on the core language model technology. Participants in the AI group accessed and interacted with Xiao Zhi through the publicly available QQ application on various operating systems, including Windows, macOS, Android, and iOS. Participants engaged with Xiao Zhi via text-based input within the QQ interface, which the chatbot processed using natural language understanding technology to generate relevant responses. Mirroring human-delivered CBT, Xiao Zhi guided participants to identify and differentiate their emotions, thoughts, reactions, and behaviors, encouraging the development of more adaptive thought patterns. All conversation records with Xiao Zhi will be encrypted and stored separately from participant identifying information. Only authorized researchers will have access to this data under strict confidentiality agreements and solely for research purposes.\u003c/p\u003e\n\u003cp\u003eA carefully constructed prompt engineering framework was developed to establish Xiao Zhi\u0026apos;s role as a CBT-based psychological counselor. The key components of this framework include precise role definition, specific behavioral instructions, illustrative dialogues, comprehensive CBT guide integration, and system-level prompts. Specific content is detailed in the supplementary material: Prompt Engineering Framework for AI Chatbot (See Supplementary Material 2 for full details).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e2.4.2 Human Group\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe selected peer counseling as the human intervention model for this study due to its recognized benefits in school mental health services. Peer counseling involves semi-professionally trained and supervised students providing listening, support, and counseling to their peers using verbal and nonverbal communication methods\u0026nbsp;[15]\u0026nbsp;. Research highlights the positive impacts of peer counseling, including effective psychological support, a relaxed atmosphere, reduced stigma, increased participation, and enhanced communication among peers\u0026nbsp;[16,17].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo ensure high-quality peer counseling, we implemented a rigorous training and selection process. We recruited 30 peer counselor candidates majoring in applied psychology and psychiatry. Candidates underwent three months of intensive CBT skills training followed by two months of practical experience. We also provided specialized training to familiarize candidates with AI communication methods and enhance their text-based online intervention skills. We used standardized assessment criteria to evaluate candidates\u0026apos; ability to provide iCBT, selecting the top 10 students who met the criteria as peer counselors for the study. Throughout the study, a licensed psychotherapist with over 10 years of CBT experience provided ongoing supervision to all peer counselors to ensure adherence to iCBT protocol. More detailed information on the peer counselor training and supervision process is provided in Supplementary Material 3: Peer Counselor Training and Supervision Manual. Counseling records will be stored in encrypted form on secure servers or using encrypted paper documents kept in secure locations. Only authorized researchers will have access.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e2.4.3 Control Group\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants in the control group were placed on a waitlist for the iCBT intervention. This waitlist control design allowed us to compare the effectiveness of both iCBT delivery methods (AI-driven and peer-counselor-delivered) to a baseline condition, accounting for the potential impact of time and nonspecific factors (e.g., natural recovery, time passage, repeated assessment effects, and the attention received from participating in the study) on symptom change.To further address ethical considerations, participants were instructed to report any engagement in other forms of psychological treatment during the study period to the research staff. This self-reporting allowed us to monitor potential confounding influences on the study outcomes. Additionally, should any participant experience a significant worsening of symptoms or express an urgent need for intervention during the study, they would be provided with referrals to appropriate mental health services.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e2.4.4 CBT Principles and Guidelines\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, both the Xiao Zhi chatbot and peer counselors strictly adhered to core Cognitive Behavioral Therapy (CBT) principles in their interventions, forming the theoretical foundation of our approach. These principles encompass cognitive restructuring (helping individuals identify and change irrational thoughts), behavioral activation (encouraging engagement in activities to improve mood and behavior), problem-solving (teaching effective methods and skills), and emotional regulation (helping individuals identify, understand, and manage emotions). Working together, these principles assist participants in coping with psychological distress from cognitive, behavioral, and emotional perspectives. Rather than adopting a single published guideline, a tailored set of CBT intervention guidelines was developed for this research, based on established CBT theory and rich clinical practice experience, and extensively referencing authoritative works [18, 19]. To ensure professionalism and applicability to online interventions, two psychology experts with doctoral degrees, advanced professional titles, extensive CBT clinical and research experience, specialized training, and relevant professional certifications participated in the guideline development process. Their combined expertise ensured the guidelines adhered to the theoretical framework of CBT and fully considered the unique characteristics of online delivery (e.g., clear/concise language, timely/effective responses, mitigating misunderstandings), thereby ensuring the intervention\u0026apos;s safety and effectiveness.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Instruments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePatient Health Questionnaire-9(PHQ-9)\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe PHQ-9 is a widely used, standardized screening tool designed to assess the severity of depressive symptoms\u0026nbsp;[20]\u0026nbsp;. It consists of nine items, each rated on a 4-point scale (0 = not at all, 1 = several days, 2 = more than half the days, 3 = nearly every day) based on the frequency of symptoms experienced over the past two weeks. The total score ranges from 0 to 27 and is used to categorize depression severity: none (0-4), mild (5-9), moderate (10-14), and severe (15 or greater). The PHQ-9 has demonstrated good internal consistency (\u0026alpha; = 0.86-0.89) and criterion validity\u0026nbsp;[20]. Multiple studies have validated the reliability and validity of the PHQ-9 in Chinese populations. For example, a systematic review by\u0026nbsp;Yin\u003c/p\u003e\n\u003cp\u003eet al. (2022)\u0026nbsp;indicates that the Chinese version of the PHQ-9 demonstrates good psychometric properties across various populations, including acceptable internal consistency, good discriminant validity, and criterion validity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eBeck Scale for Suicide Ideation-Chinese Version(BSI-CV)\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used the Chinese version of the Beck Scale for Suicide Ideation (BSI-CV) to assess participants\u0026apos; current suicide ideation intensity. This 19-item scale uses a 3-point Likert response format (0-2) for each item, resulting in a total score ranging from 0 to 38. Higher BSI-CV scores indicate higher levels of current suicide ideation. The BSI-CV is a commonly used tool for suicide risk assessment and can identify individuals at elevated risk of suicide, aiding clinicians in risk assessment and intervention\u0026nbsp;[22]\u0026nbsp;.\u0026nbsp;Liu et al. (2023)\u0026nbsp;demonstrated good reliability and validity of the BSI-CV in Chinese adults.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eGeneral Self-Efficacy Scale (GSES)\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study used the General Self-Efficacy Scale (GSES) to assess participants\u0026apos; self-efficacy. The GSES is a widely used 10-item scale designed to assess individuals\u0026apos; general beliefs in their ability to cope with a broad range of life challenges\u0026nbsp;[24]\u0026nbsp;. Each item is rated on a 4-point scale ranging from 1 (not at all true) to 4 (exactly true). Sample items include: \u0026quot;I can manage to solve difficult problems if I try hard enough\u0026quot; and \u0026quot;If I am in trouble, I can usually think of a solution\u0026quot;. The total score ranges from 10 to 40, with higher scores indicating higher levels of self-efficacy. The GSES has been widely used globally and validated in various cultural contexts, demonstrating good reliability and validity\u0026nbsp;[25]\u0026nbsp;.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSelf-compiled Participant Expectation Questionnaire (PEQ)\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo examine potential baseline differences in treatment expectations across groups, we developed a 22-item Participant Expectation Questionnaire. The questionnaire\u0026apos;s design was informed by the bifactor model proposed by Sun Qiwu (citation), dividing the items into two dimensions: expectations of the counselor\u0026apos;s professionalism and expectations of the participant\u0026apos;s personal commitment. We also drew upon the Expectations About Counseling - Brief (EAC-B) scale (citation), adapting items to specifically measure participants\u0026apos; subjective expectations regarding AI-delivered psychological counseling. The first 20 items use a 5-point Likert scale (e.g., strongly disagree to strongly agree), while the final two items employ a 10-point scale (e.g., not at all likely to extremely likely). Higher scores across all items indicate higher overall expectations of psychological counseling. Total scores are calculated by summing the responses to all 22 items, with a maximum possible score of 160. This questionnaire served as an operational measure to compare baseline treatment expectations across the three groups, aiming to control for potential confounding effects of pre-existing expectations on intervention outcomes. It was not used as an outcome variable to assess changes in expectations following the intervention.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSemi-Structured Interviews\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGrounded in the principles of grounded theory\u0026nbsp;[26]\u0026nbsp;, we conducted semi-structured interviews to explore the underlying mechanisms driving differences in intervention effects. The interview guide was developed iteratively, starting with a comprehensive literature review and initial data collection, followed by refinement based on preliminary investigations and expert consultation. The interviews centered around three key areas of inquiry: participants\u0026apos; experiences and psychological changes during the intervention, perceived similarities and differences between AI and human intervention, and preferences for future intervention modalities and the reasons behind those preferences.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 Study Procedure\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePilot Study\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePrior to the formal study, we conducted a pilot study with six participants (three in the AI intervention group and three in the human intervention group). Participants were informed that they were taking part in an experiment on AI psychological intervention. They then engaged in text-based counseling sessions and post-intervention interviews. The pilot study included an operational check to assess the effectiveness of the blinding procedures, specifically whether participants in the human intervention group believed their counselors were AI. Results confirmed that participants in the human intervention group did indeed perceive their counselors as AI.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eFormal Study\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInterested individuals first completed a screening questionnaire. Those meeting the inclusion criteria proceeded to the formal study, which consisted of a four-week intervention phase. Participants in the AI intervention group received iCBT from the chatbot \u0026quot;Xiao Zhi, \u0026quot; while those in the human intervention group received iCBT from peer counselors. Each participant engaged in two one-hour intervention sessions per week, totaling eight hours of intervention over the four weeks. We administered the PHQ-9, BSI-CV, and GSES scales at baseline, after two weeks, and again after four weeks to assess changes in depressive symptoms, suicidal ideation, and general self-efficacy. All three groups of participants received three measurements. Following the intervention, we conducted semi-structured interviews with all participants to explore their experiences and perceptions of the intervention process. Participants were asked to review their text-based counseling records prior to the interview. All interviews were audio-recorded for subsequent qualitative analysis(See Figure 1).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.7 Debriefing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUpon completion of the study, all participants attended a debriefing session. As a core ethical measure to address the deception inherent in the participant-blinding procedure, the research team fully explained the rationale for the study design, including the use of peer counselors posing as AI in the human intervention group. Participants had the opportunity to ask questions about the study design and share their feedback, thoughts, and feelings about their experiences.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.8 Sample size and data analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSample size was calculated using G*Power 3.1 software, with a significance level (\u0026alpha;) set at 0.05 and power (1-\u0026beta;) at 0.80. The effect size (f) was a key consideration. Previous research has shown that internet interventions (including iCBT) typically produce moderate to large effect sizes in improving depressive symptoms\u0026nbsp;[8,20,27]\u0026nbsp;. Furthermore, studies by\u0026nbsp;Fitzpatrick et al. (2017)\u0026nbsp;have also shown that AI-driven interventions can effectively provide mental health support. Based on these findings and referencing the typical range of effect sizes found in previous iCBT research, we hypothesized that AI interventions would have similar effects to human interventions, and ultimately set the effect size (f) at 0.30. Based on these parameters, a minimum of 75 participants were required. Considering a potential attrition rate of 10% and other unforeseen factors, we increased the recruitment target to 84 participants.\u003c/p\u003e\n\u003cp\u003eWe analyzed quantitative data using SPSS version 27. We compared baseline characteristics between the three groups using analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. To assess changes in depressive symptoms, suicidal ideation, and self-efficacy over time and between groups, we employed repeated measures ANOVA.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe analyzed qualitative data from the semi-structured interviews using NVivo 12 Plus. We transcribed the audio recordings verbatim and employed an open coding approach to identify key themes and patterns. The initial codes were then categorized and integrated into axial codes, allowing us to compare and contrast the experiences and perceptions of participants in the AI and human intervention groups.\u0026nbsp;\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Participant Characteristics and flow\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOf the 388 individuals who responded to recruitment materials, 90 met the eligibility criteria and were enrolled in the study. Of these, 87 participants completed the intervention, resulting in a dropout rate of 3. 3% (3 out of 90). The majority of participants were students (n = 79), with a small number employed (n = 5) or reporting other occupations (n = 3). Participants\u0026apos; ages ranged from 18 to 27 years (mean = 20. 46, standard deviation = 1. 94).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Baseline Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe conducted one-way ANOVAs to examine potential differences in baseline characteristics between the three study groups. As shown in Table 1, we found no statistically significant differences between groups in terms of age, gender ratio, depression level (PHQ-9), participant expectation (PEQ), suicide ideation (BSI-CV), or self-efficacy (GSES), indicating that the randomization process resulted in comparable baseline characteristics across the groups. These results support the validity of conducting subsequent analyses to compare the effects of the interventions (see Table 1).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Quantitative Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo examine the effects of the interventions on psychological well-being, repeated measures ANOVAs were conducted for each outcome variable (PHQ-9, BSI-CV, GSES). Time (baseline, second week, post-intervention) was treated as the within-subjects factor, and Group (AI, Human, Control) was the between-subjects factor. Greenhouse-Geisser corrections were applied when sphericity assumptions were violated. Bonferroni corrections were used for post-hoc pairwise comparisons. The results are presented in Tables 2 and 3.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePrimary outcome variable: Depression symptoms\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs shown in Table 2, there was a significant main effect of Time on depressive symptoms as measured by the PHQ-9, \u003cem\u003eF\u003c/em\u003e = 28. 520, \u003cem\u003ep\u003c/em\u003e \u0026lt;0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 253. This indicates that depressive symptoms significantly decreased from baseline to both 2 weeks and 4 weeks across all groups. There was also a significant main effect of Group on depressive symptoms(Across three time points), \u003cem\u003eF\u003c/em\u003e = 7. 088, \u003cem\u003ep\u003c/em\u003e = 0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 144, suggesting that the type of intervention (AI, Human, or Control) had a distinct impact on depressive symptoms. However, the significant Time x Group interaction effect, \u003cem\u003eF\u003c/em\u003e = 5. 920, \u003cem\u003ep\u003c/em\u003e \u0026lt;0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 124, indicated that the rate of change in depressive symptoms over time differed between the intervention groups. To further investigate this interaction, we conducted a simple effects analysis to assess the effect of group at each time period.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAt baseline, there was no significant effect of Group on PHQ-9 scores, \u003cem\u003eF\u003c/em\u003e = 1. 733, \u003cem\u003ep\u003c/em\u003e = 0. 183, partial \u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 040. Post hoc analyses with Bonferroni correction showed that there was no significant difference in depression levels at baseline (T0) among the three groups, suggesting that the data were comparable. At week two, there was a significant effect of Group on PHQ-9 scores, \u003cem\u003eF\u003c/em\u003e = 7. 410, \u003cem\u003ep\u003c/em\u003e = 0. 001, partial \u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 150. Post hoc analyses with Bonferroni correction showed significant differences between the AI and control groups (d = -0. 86) and between the human and control groups (\u003cem\u003ed\u0026nbsp;\u003c/em\u003e= -0. 76), however there were no significant differences between the AI and human groups (\u003cem\u003ed\u0026nbsp;\u003c/em\u003e= -0. 12). At week four, there was a significant effect of Group on PHQ-9 scores, \u003cem\u003eF\u003c/em\u003e = 10. 334, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0. 001, partial \u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 197. Post hoc analyses with Bonferroni correction showed significant differences between the AI and control groups (\u003cem\u003ed\u003c/em\u003e = -0. 65) and between the human and control groups (\u003cem\u003ed\u003c/em\u003e = -1. 15), however there were no significant differences between the AI and human groups (\u003cem\u003ed\u003c/em\u003e = 0. 55)(See Table 3). Furthermore, the AI group showed a significant reduction in depression symptoms from baseline to the second week (T0 to T1, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0. 05), with no further significant change observed from the second week to post-intervention (T1 to T2). Conversely, the human intervention group demonstrated a continuous and significant downward trend in depression scores across all time points (T0, T1, and T2,\u003cem\u003e\u0026nbsp;p\u003c/em\u003e \u0026lt; 0. 05).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eSecondary outcome variables: Suicide ideation and self-efficacy\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs shown in Table 2, there was a significant main effect of Time on suicidal ideation as measured by the BCI-CV, \u003cem\u003eF\u003c/em\u003e = 18. 830, \u003cem\u003ep\u003c/em\u003e \u0026lt;0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 183. This indicates that suicidal ideation significantly decreased from baseline to both 2 weeks and 4 weeks across all groups. There was also a significant main effect of Group on suicidal ideation(Across three time points), \u003cem\u003eF\u003c/em\u003e = 4. 985, \u003cem\u003ep\u0026nbsp;\u003c/em\u003e= 0. 009, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 106 , suggesting that the type of intervention (AI, Human, or Control) had a distinct impact on suicidal ideation. The significant Time x Group interaction effect, \u003cem\u003eF\u003c/em\u003e = 3. 555, \u003cem\u003ep\u003c/em\u003e =0. 012, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 078, suggests that the rate of change in suicidal ideation differed among the intervention groups over time. Simple effects analyses were conducted to further examine this interaction.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUnexpectedly, there was no significant effect of group on BCI-CV scores at baseline, week 2 and week 4. While no significant differences between groups were found, analysis of within-group changes highlighted important trends(See Table 3). For the AI group, there was a significant effect of Time on BCI-CV scores,\u003cem\u003e\u0026nbsp;F\u003c/em\u003e = 9. 797, \u003cem\u003ep\u003c/em\u003e \u0026lt; 0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 191, with a moderate decrease from baseline to 2 weeks (\u003cem\u003ed\u003c/em\u003e = 0. 48) and from baseline to 4 weeks (\u003cem\u003ed\u0026nbsp;\u003c/em\u003e= 0. 74). The Human group also showed a significant effect of Time on BCI-CV scores, \u003cem\u003eF\u0026nbsp;\u003c/em\u003e= 7. 682, \u003cem\u003ep\u003c/em\u003e =0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 156, with a large reduction from baseline to 2 weeks (\u003cem\u003ed\u0026nbsp;\u003c/em\u003e= 0. 46) and from baseline to 4 weeks (\u003cem\u003ed\u0026nbsp;\u003c/em\u003e= 0. 71). In contrast, the Control group did not exhibit significant changes in BCI-CV scores over time, \u003cem\u003eF\u003c/em\u003e =2. 324, \u003cem\u003ep\u003c/em\u003e=0. 104, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 053.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThere was a significant main effect of Time on self-efficacy as measured by the GSES, \u003cem\u003eF\u0026nbsp;\u003c/em\u003e= 29. 539,\u003cem\u003e\u0026nbsp;p\u0026nbsp;\u003c/em\u003e\u0026lt; 0. 001, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 260, indicating significant improvements in self-efficacy from baseline to both 2 weeks and 4 weeks across all groups. The Time x Group interaction effect was significant, \u003cem\u003eF\u003c/em\u003e = 3. 670, \u003cem\u003ep\u0026nbsp;\u003c/em\u003e=0. 009, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0. 080, indicated that the rate of change in depressive symptoms over time differed between the intervention groups. To further investigate this interaction, we conducted a simple effects analysis to assess the effect of group at each time period.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAt baseline, there was no significant main effect of Group on GSES scores,\u0026nbsp;\u003cem\u003eF\u003c/em\u003e = 1.217,\u0026nbsp;\u003cem\u003ep\u003c/em\u003e = 0.301, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0.028, indicating that self-efficacy levels were comparable across the AI, Human, and Control groups. Similarly, at two weeks, the group effect was not significant,\u0026nbsp;\u003cem\u003eF\u003c/em\u003e = 0.533,\u0026nbsp;\u003cem\u003ep\u003c/em\u003e = 0.589, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0.013. However, by week four, a significant main effect of Group emerged,\u0026nbsp;\u003cem\u003eF\u003c/em\u003e = 4.639,\u0026nbsp;\u003cem\u003ep\u003c/em\u003e = 0.012, partial\u0026nbsp;\u003cem\u003e\u0026eta;\u003c/em\u003e\u003csup\u003e2\u003c/sup\u003e = 0.099. Post hoc analyses with Bonferroni correction revealed a substantial increase in self-efficacy within the Human group compared to the Control group (d = 0.74). But there were no significant differences in self-efficacy between the Human and AI groups at any time point (See Table 3). Furthermore, we found that the AI group showed significant improvements in self-efficacy from baseline to post-intervention (T0 to T2,\u0026nbsp;\u003cem\u003ep\u003c/em\u003e \u0026lt; 0.001), specifically, the AI group showed significant improvements from baseline to two weeks post-intervention (T0 to T2,\u0026nbsp;\u003cem\u003ep\u0026nbsp;\u003c/em\u003e\u0026lt; 0.001), but there was no significant difference between baseline and four weeks post-intervention. While the human intervention group also displayed an increase in self-efficacy from baseline to post-intervention, this change did not reach statistical significance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Qualitative Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe conducted a qualitative analysis of the semi-structured interview transcripts using an open coding approach to identify key themes and patterns. From the initial codes, we developed three core categories representing the central themes emerging from the data: Intervention Process, Intervention Effects, and Intervention Prospects (See Table 4).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eIntervention Process: Advantages and Disadvantages\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAdvantages: Participants in the AI intervention group highlighted several key advantages of AI-delivered iCBT, particularly its convenience and privacy. They appreciated the ability to access intervention regardless of time or location, saving both time and money. Many also felt that AI intervention alleviated privacy concerns, fostering a sense of safety and anonymity. This was especially true for those with social anxiety, who found the absence of a human counselor less intimidating, reducing stigma and encouraging self-disclosure. Participants in the human intervention group, on the other hand, valued the emotional interaction and nuanced understanding provided by their peer counselors, emphasizing the richer emotional experience.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDisadvantages: Despite its advantages, AI-delivered iCBT also presented challenges. Participants noted difficulties the chatbot encountered in recognizing subtle emotional cues in language, limiting its ability to fully grasp emotional complexity and nuances. They felt that the AI, compared to human counselors, struggled to accurately perceive and validate their emotional dynamics, sometimes leading to formulaic responses or inappropriate questioning techniques that felt intrusive or \u0026quot;prying.\u0026quot;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eIntervention Effects: Perceived Efficacy and Support\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhile both the AI and human intervention groups acknowledged that the iCBT interventions provided practical suggestions and successfully initiated positive intervention outcomes, the qualitative data also highlighted the distinct difference in support depth. Both groups reported emotional support, but the human intervention was described as offering more profound validation and empathy, which is crucial for maintaining therapeutic motivation and sustained symptom improvement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eIntervention Prospects: Future Preferences and Applications\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants expressed varying preferences for future modalities. While a significant number of participants were optimistic about the potential of AI-driven interventions, this preference stemmed primarily from its perceived advantages of personalized matching, cost-effectiveness, and a consistent process. Conversely, those who maintained a preference for human intervention highlighted the need for nuanced emotional understanding and flexibility\u0026mdash;qualities they associated with human interaction, even when reflecting on the \u0026apos;AI\u0026apos; support they received.\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e\u003cstrong\u003eCurrent Findings\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis randomized controlled trial examined the comparative effectiveness of AI-delivered iCBT and human-delivered iCBT for reducing depressive symptoms and suicidal ideation among young adults. Importantly, our study sought to move beyond simply comparing efficacy by exploring users\u0026apos; experiences and perceptions of both intervention modalities.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOur findings demonstrate that AI-delivered iCBT can significantly reduce depressive symptoms in young adults, aligning with previous research highlighting the therapeutic potential of AI in mental health interventions\u0026nbsp;[28,29]. Moreover, we found that both AI and human-delivered iCBT led to significant reductions in suicidal ideation. These findings underscore the potential of both approaches in addressing these critical mental health concerns, particularly among young adults where suicide risk is a major concern.\u003c/p\u003e\n\u003cp\u003eCrucially, the primary distinction between the intervention modalities emerged in the sustained therapeutic benefit. Our study revealed no significant differences in effectiveness between AI-delivered and human-therapist-delivered iCBT on any of the measured outcomes at week two. However, a key observation is that while the AI group showed significant improvement from baseline to week two in depression scores, this benefit did not continue from week two to week four, exhibiting a plateau effect. This stands in sharp contrast to the human intervention group, which demonstrated a continuous and significant downward trend in depression scores across all time points. This observation, even without resulting in a statistically significant difference between the two groups at week four, suggests that the human element possesses a unique, sustained advantage in maintaining therapeutic momentum and achieving more durable gains.\u003c/p\u003e\n\u003cp\u003eRegarding self-efficacy, a significant time effect was observed, indicating a general trend of increasing scores from baseline to week two, followed by a decrease by week four in all groups. However, this trend was not differentially influenced by the intervention type, as no significant differences were found between the AI and Human groups at any assessment point. This observed pattern in self-efficacy, distinct from the plateau effect seen in depression scores, suggests that changes in self-efficacy may be attributable to factors other than the specific intervention modality itself. Therefore, the current findings offer limited insight into the specific impact of AI-delivered iCBT on self-efficacy, warranting further investigation with more refined methodologies and longer follow-up periods to disentangle the observed trend from potential confounding factors.\u003c/p\u003e\n\u003cp\u003eDelving deeper into the qualitative data, we uncovered several key themes that shed light on the nuances of user experiences with AI-delivered iCBT. As anticipated, participants appreciated the advantages of AI, such as temporal and spatial flexibility and privacy. These benefits echo prior findings, suggesting that individuals, particularly those with social anxiety or introversion, may feel more comfortable disclosing personal information to AI systems\u0026nbsp;[30\u0026ndash;32]. The anonymity provided by AI can foster a sense of safety and reduce the stigma often associated with seeking mental health care, potentially leading to more honest self-disclosure and deeper engagement with the therapeutic process.\u003c/p\u003e\n\u003cp\u003eHowever, our qualitative findings also revealed limitations in AI-delivered iCBT, particularly regarding the chatbot\u0026apos;s capacity for understanding and responding to complex human language and applying therapeutic techniques with the same flexibility as a human therapist. Participants noted instances where the AI struggled to fully grasp emotional nuances and provide tailored responses, resulting in a perceived lack of personalized support. This limitation aligns with previous studies highlighting challenges in AI\u0026apos;s ability to process and interpret complex human communication\u0026nbsp;[10,33]. Participants reported that the AI often failed to recognize subtle shifts in emotional tone and complex semantic layers, demonstrating a significant weakness in clinical intuition and judgment. Its responses were sometimes perceived as rigid, formulaic, and poorly timed, especially during deeper explorations involving core beliefs. This tendency towards mechanical application of techniques, rather than flexible, personalized support, led some participants to feel unheard or misunderstood. This resultant loss of engagement and therapeutic momentum is hypothesized to directly contribute to the observed plateau effect in the AI group\u0026apos;s depression scores, contrasting sharply with the sustained improvement demonstrated by the human group. Interestingly, despite participants in the human group believing they were interacting with an AI, this group showed sustained improvement. This suggests that the human counselors\u0026apos; superior ability to interpret complex emotional context and provide nuanced, non-repetitive responses\u0026mdash;which is a function of advanced cognitive and empathetic ability\u0026mdash;rather than the explicit awareness of a human relationship, may be the key driver for long-term therapeutic gains. The current AI, while able to generate basic empathetic responses, hit a \u0026apos;cognitive ceiling\u0026apos; when attempting to fully address deeper core beliefs. This superiority is rooted in the human capacity for nuanced interpretation (i.e., listening for the nuance or \u0026apos;between-the-lines\u0026apos; meaning) and providing non-formulaic, adaptive engagement necessary for deep cognitive and emotional restructuring. Furthermore, the analysis of future preferences revealed a practical functional distinction: participants viewed AI as suitable for accessible, short-term support and initial symptom management, while human counselors were deemed essential for addressing complex, long-term psychological issues requiring genuine relational interaction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFuture research and Limitations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDespite these findings, our study has limitations that warrant consideration. First, our sample primarily comprised young adults, potentially limiting the generalizability of our findings to other age groups. Future research should encompass more diverse samples to determine if AI-delivered iCBT demonstrates comparable effectiveness across different demographics and mental health needs. Second, while we assessed emotional responses through self-report measures and qualitative interviews, our study design relied solely on text-based communication. This approach, while aligning with the nature of many digital mental health interventions, limits our understanding of how AI might perform in interventions incorporating audio or visual cues. Future studies incorporating multimodal communication channels could provide richer insights into the nuances of AI-driven empathy and emotional support. Third, our study focused on short-term outcomes. Exploring the long-term efficacy and sustainability of AI-delivered iCBT, compared to human-delivered approaches, is crucial for determining its potential as a scalable and sustainable mental health solution. Finally, our use of a waitlist control group has limitations. While it allowed us to assess the overall effectiveness of both interventions compared to no intervention, it may not have fully controlled for non-specific effects associated with engaging with an online platform and the expectation of receiving support. Future research could consider employing an active control group (e.g., an attention-matched control group receiving a non-therapeutic online application) to better isolate the specific therapeutic effects of the AI and human-delivered iCBT. Our study focused on a single AI chatbot utilizing a specific iCBT program. Future research should explore users\u0026apos; experiences with various AI models and therapeutic approaches to determine the generalizability of our findings and identify optimal combinations of AI technology and therapeutic techniques.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn conclusion, our findings indicate that AI-delivered iCBT can be a promising short-term tool for addressing depression and suicidal ideation in young adults, demonstrating comparable effectiveness to human-therapist-delivered iCBT only in the initial stages of intervention. The key observation, however, is the plateau effect observed in the AI group\u0026apos;s depression scores, which suggests that complex, sustained therapeutic work requires the unique, adaptable engagement and relational depth of a human counselor. While AI\u0026apos;s limitations in emotional perception and nuanced language understanding require further refinement, its accessibility and adaptability highlight its potential to bridge critical gaps in mental healthcare as a scalable supplementary or short-term intervention. Future research should prioritize long-term outcomes and exploration of various AI models and intervention approaches to further advance the development and implementation of effective and accessible AI-driven mental health solutions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study involving human participants was approved by the Biomedical Ethics Committee of Southern Medical University, China (Reference Number: [2023] No. 53). The study procedures were conducted in accordance with the ethical standards of the responsible committee and with the Declaration of Helsinki. All participants provided written informed consent prior to participation and were informed of their right to withdraw at any time without penalty.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. (This manuscript does not contain any individual person\u0026rsquo;s data, image, or video in any form.)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data underlying the findings of this study contain sensitive patient information and cannot be publicly shared due to privacy regulations. Anonymized summary statistics are available upon reasonable request from the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Guangdong Provincial Undergraduate Teaching Quality and Teaching Reform Project (Yue Jiao Gao Han [2024] No. 30) and the Ministry of Education of China, Humanities and Social Sciences Research Project (25YJA190016).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYiyang Wu : Conceptualization, Data curation, Formal analysis, Software, Supervision, Validation, Visualization, Writing - original draft. Haoran Song: Conceptualization, Data curation, Formal analysis, Project administration, Investigation, Software, Supervision. Weihao Huang: Formal analysis, Investigation, Software, Validation, Visualization. Chen Ye: Investigation, Software. Ruoyu Lin: Investigation, Software.You Wang: Funding acquisition, Resources. Xueling Yang: Conceptualization, Funding acquisition, Methodology, Project administration, Resources, Supervision, Writing - review and editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to thank the participants and research assistants for their contributions to this study. The intervention team consisted of the following: Xueling Yang (Supervisor), Chenhua Wang (Clinician), Yeya Lu (Clinician), Ruoxi Li (Peer Counselor), Wenxi Huang (Peer Counselor), Xiaoqiao Lu (Peer Counselor), Jiatong Liang (Peer Counselor), Yun Shi Yao (Peer Counselor), Guangying He (Peer Counselor), Zijun Wang (Peer Counselor), Chun Zhan (Peer Counselor), Wenxuan Zhou (Peer Counselor), and Xiaorong Lin (Peer Counselor).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDuring the preparation of this work, the author(s) used Google AI Studio\u0026apos;s Gemini model to refine the language and improve the clarity and readability of the manuscript. After using this tool, the author(s) carefully reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSmith K. Mental health: A world of depression. Nature 2014;515:180\u0026ndash;1. https://doi.org/10.1038/515180a.\u003c/li\u003e\n\u003cli\u003eShorey S, Ng ED, Wong CHJ. Global prevalence of depression and elevated depressive symptoms among adolescents: A systematic review and meta-analysis. Br J Clin Psychol 2022;61:287\u0026ndash;305. https://doi.org/10.1111/bjc.12333.\u003c/li\u003e\n\u003cli\u003eHofmann SG, Asnaani A, Vonk IJ, et al. The efficacy of cognitive behavioral therapy: A review of meta-analyses. Cogn Ther Res 2012;36:427\u0026ndash;40.\u003c/li\u003e\n\u003cli\u003eLuo C, Sanger N, Singhal N, et al. A comparison of electronically-delivered and face to face cognitive behavioural therapies in depressive disorders: A systematic review and meta-analysis. eClinicalMedicine 2020;24:100442. https://doi.org/10.1016/j.eclinm.2020.100442.\u003c/li\u003e\n\u003cli\u003eHedman E, Lj\u0026oacute;tsson B, Lindefors N. Cognitive behavior therapy via the Internet: a systematic review of applications, clinical efficacy and cost\u0026ndash;effectiveness. Expert Rev Pharmacoecon Outcomes Res 2012;12:745\u0026ndash;64. https://doi.org/10.1586/erp.12.67.\u003c/li\u003e\n\u003cli\u003eMuse K, McManus F. A systematic review of methods for assessing competence in cognitive\u0026ndash;behavioural therapy. Clin Psychol Rev 2013;33:484\u0026ndash;99. https://doi.org/10.1016/j.cpr.2013.01.010.\u003c/li\u003e\n\u003cli\u003eLoughnan SA, Newby JM, Haskelberg H, et al. Internet-based cognitive behavioural therapy (iCBT) for perinatal anxiety and depression versus treatment as usual: study protocol for two randomised controlled trials. Trials 2018;19:56. https://doi.org/10.1186/s13063-017-2422-5.\u003c/li\u003e\n\u003cli\u003eKaryotaki E, Efthimiou O, Miguel C, et al. Internet-based cognitive behavioral therapy for depression: a systematic review and individual patient data network meta-analysis. JAMA Psychiatry 2021;78:361\u0026ndash;71.\u003c/li\u003e\n\u003cli\u003eFitzpatrick KK, Darcy A, Vierhile M. Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Ment Health 2017;4:e19. https://doi.org/10.2196/mental.7785.\u003c/li\u003e\n\u003cli\u003eGreen EP, Lai Y, Pearson N, et al. Expanding Access to Perinatal Depression Treatment in Kenya Through Automated Psychological Support: Development and Usability Study. JMIR Form Res 2020;4:e17895. https://doi.org/10.2196/17895.\u003c/li\u003e\n\u003cli\u003eLan J. The historical mission of generative artificial intelligence and humanities and social sciences - starting from the ChatGPT intelligent revolution Ideological and Theoretical Education. Ideol Theor Educ 2023:12\u0026ndash;8. https://doi.org/10.16075/j.cnki.cn31-1220/g4.2023.04.012.\u003c/li\u003e\n\u003cli\u003eLiu J. ChatGPT: perspectives from human\u0026ndash;computer interaction and psychology. Front Artif Intell 2024;7:1418869. https://doi.org/10.3389/frai.2024.1418869.\u003c/li\u003e\n\u003cli\u003eDave T, Athaluri SA, Singh S. ChatGPT in medicine: an overview of its applications, advantages, limitations, future prospects, and ethical considerations. Front Artif Intell 2023;6:1169595. https://doi.org/10.3389/frai.2023.1169595.\u003c/li\u003e\n\u003cli\u003eEysenbach G. The Role of ChatGPT, Generative Language Models, and Artificial Intelligence in Medical Education: A Conversation With ChatGPT and a Call for Papers. JMIR Med Educ 2023;9:e46885. https://doi.org/10.2196/46885.\u003c/li\u003e\n\u003cli\u003eSussman MB. The development and effects of a model for training peer-group counselors in a multi-ethnic junior high school. University of Miami; 1973.\u003c/li\u003e\n\u003cli\u003eTindall JA. Peer counseling: An in-depth look at training peer helpers. ERIC; 1989.\u003c/li\u003e\n\u003cli\u003eWinterton CI, Dunk RDP, Wiles JR. Peer-led team learning for introductory biology: relationships between peer-leader relatability, perceived role model status, and the potential influences of these variables on student learning gains. Discip Interdiscip Sci Educ Res 2020;2:3. https://doi.org/10.1186/s43031-020-00020-9.\u003c/li\u003e\n\u003cli\u003eBeck JS, Beck A, Beck J. Cognitive behavior therapy: basics and beyond. ed. N Y 2011.\u003c/li\u003e\n\u003cli\u003eDavid B, Burns M. Feeling Good: The New Mood Therapy. NY Signet Books Chin Richard 1980:42\u0026ndash;3.\u003c/li\u003e\n\u003cli\u003eKaryotaki E, Furukawa TA, Efthimiou O, et al. Guided or self-guided internet-based cognitive\u0026ndash;behavioural therapy (iCBT) for depression? Study protocol of an individual participant data network meta-analysis. BMJ Open 2019;9:e026820.\u003c/li\u003e\n\u003cli\u003eYin L, Teklu S, Pham H, et al. Validity of the Chinese Language Patient Health Questionnaire 2 and 9: A Systematic Review. Health Equity 2022;6:574\u0026ndash;94. https://doi.org/10.1089/heq.2022.0030.\u003c/li\u003e\n\u003cli\u003eBeck AT, Kovacs M, Weissman A. Assessment of suicidal intention: the Scale for Suicide Ideation. J Consult Clin Psychol 1979;47:343.\u003c/li\u003e\n\u003cli\u003eLin Z, Cheng L, Han X, et al. The Effect of Internet-Based Cognitive Behavioral Therapy on Major Depressive Disorder: Randomized Controlled Trial. J Med Internet Res 2023;25:e42786. https://doi.org/10.2196/42786.\u003c/li\u003e\n\u003cli\u003eSchwarzer R. Generalized self-efficacy scale. Meas Health Psychol User\u0026rsquo;s Portf Causal Control BeliefsNfer-Nelson 1995.\u003c/li\u003e\n\u003cli\u003eLuszczynska A, Scholz U, Schwarzer R. The General Self-Efficacy Scale: Multicultural Validation Studies. J Psychol 2005;139:439\u0026ndash;57. https://doi.org/10.3200/JRLP.139.5.439-457.\u003c/li\u003e\n\u003cli\u003eGlaser B, Strauss A. Discovery of grounded theory: Strategies for qualitative research. Routledge; 2017.\u003c/li\u003e\n\u003cli\u003eCuijpers P, Noma H, Karyotaki E, et al. Effectiveness and acceptability of cognitive behavior therapy delivery formats in adults with depression: a network meta-analysis. JAMA Psychiatry 2019;76:700\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003eDonkin L, Hickie IB, Christensen H, et al. Rethinking the Dose-Response Relationship Between Usage and Outcome in an Online Intervention for Depression: Randomized Controlled Trial. J Med Internet Res 2013;15:e231. https://doi.org/10.2196/jmir.2771.\u003c/li\u003e\n\u003cli\u003eLiu H, Peng H, Song X, et al. Using AI chatbots to provide self-help depression interventions for university students: A randomized trial of effectiveness. Internet Interv 2022;27:100495. https://doi.org/10.1016/j.invent.2022.100495.\u003c/li\u003e\n\u003cli\u003eDARWIN C. The expression of the emotions in man and animals (1872). Portable Darwin 1993:364\u0026ndash;93.\u003c/li\u003e\n\u003cli\u003eQuigley L, Dozois DJA, Bagby RM, et al. Cognitive change in cognitive-behavioural therapy \u003cem\u003ev.\u003c/em\u003e pharmacotherapy for adult depression: a longitudinal mediation analysis. Psychol Med 2019;49:2626\u0026ndash;34. https://doi.org/10.1017/S0033291718003653.\u003c/li\u003e\n\u003cli\u003eTorous J, Staples P, Shanahan M, et al. Utilizing a Personal Smartphone Custom App to Assess the Patient Health Questionnaire-9 (PHQ-9) Depressive Symptoms in Patients With Major Depressive Disorder. JMIR Ment Health 2015;2:e8. https://doi.org/10.2196/mental.3889.\u003c/li\u003e\n\u003cli\u003ePintelas EG, Kotsilieris T, Livieris IE, et al. A review of machine learning prediction methods for anxiety disorders, 2018, p. 8\u0026ndash;15.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u0026nbsp;\u003c/strong\u003eBaseline characteristics of the participants; mean (SD)\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"105%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003eAI group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003eHuman group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003eControl group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.3093%;\"\u003e\n \u003cp\u003eF/c2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.27835%;\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eAge, mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e20. 37 (1. 83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e20. 41 (2. 39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e20. 43 (1. 52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e0. 120\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e0. 887\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eGender, n (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e8 (26. 67)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e11 (40. 74)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e6 (0. 20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e3. 081\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e0. 214\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e22 (73. 33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e16 (59. 26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e24 (0. 80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eScale, mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003ePHQ-9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e9. 37(3. 49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e9. 78(3. 08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e11. 17(4. 87)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e1. 733\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e0. 183\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eBSI-CV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e4. 90(3. 06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e3. 93(3. 27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e3. 43(3. 05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e1. 712\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e0. 187\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003eGSES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e19. 80(6. 21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e22. 04(6. 08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e19. 87(6. 00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e1. 217\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e0. 301\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 23.7113%;\"\u003e\n \u003cp\u003ePEQ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 20.6186%;\"\u003e\n \u003cp\u003e120. 00(23. 22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 18.5567%;\"\u003e\n \u003cp\u003e127. 70(16. 95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.5258%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.3093%;\"\u003e\n \u003cp\u003e2. 008\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 9.27835%;\"\u003e\n \u003cp\u003e0. 162\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u0026nbsp;\u003c/strong\u003eEffect of intervention on primary and secondary outcomes after 2 and 4 weeks\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"112%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003eOutcome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003eBaseline (T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e2 weeks (T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e4 weeks (T2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003eF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003ePartial \u0026eta;2\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003ePHQ-9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e9. 37(3. 49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e7. 00(3. 22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e7. 13(4. 18)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e9. 78(3. 08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e7. 37(3. 20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e5. 07(3. 03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eControl\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e11. 17(4. 86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e10. 40(4. 56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e10. 40(5. 69)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eTime Effect\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e28. 520\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e\u0026lt;0. 001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 253\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e7. 088\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e0. 001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 144\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eInteraction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e5. 920\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e\u0026lt;0. 001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 124\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003eBCI-CV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e4. 90(3. 06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e3. 47(2. 80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e2. 80(2. 52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e3. 555\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e0. 012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e3. 93(3. 27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e2. 59(2. 50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e1. 96(2. 18)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eControl\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e3. 43(3. 05)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e2. 77(3. 00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e3. 30(3. 46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eTime Effect\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e18. 830\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e\u0026lt;0. 001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 183\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e4. 985\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e0. 009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 106\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eInteraction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e3. 555\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e0. 012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 078\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\n \u003cp\u003eGSES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e19. 80(6. 21)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e26. 03(4. 41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e23. 73(6. 74)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e37. 128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e\u0026lt;0. 001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eHuman\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e22. 04(6. 08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e25. 11(5. 36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e24. 44(7. 08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eControl\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\n \u003cp\u003e19. 87(6. 00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003e24. 97(2. 98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\n \u003cp\u003e19. 57(6. 03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eTime Effect\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e29. 539\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e\u0026lt;0. 001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 260\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e2. 029\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e0. 138\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 046\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 11.5789%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\n \u003cp\u003eInteraction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.7368%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 13.6842%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.8421%;\"\u003e\u0026nbsp;\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e3. 670\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.47368%;\"\u003e\n \u003cp\u003e0. 009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.5263%;\"\u003e\n \u003cp\u003e0. 080\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAbbreviations: Degrees of freedom are Greenhouse-Geisser corrected. AI, AI intervention group; Human, Human intervention group; Control, control group with no intervention.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNote. T0 = baseline; T1 = week 2; T2 = week 4.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u0026nbsp;\u003c/strong\u003eThe Effects of Different Interventions on Depression, Suicidal Ideation, and Self-Efficacy: Intergroup Comparisons\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"116%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 176px;\"\u003e\n \u003cp\u003eBaseline(T0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 168px;\"\u003e\n \u003cp\u003e2 weeks(T1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" style=\"width: 163px;\"\u003e\n \u003cp\u003e4weeks(T2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eOutcome\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eContrast\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003eMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003eCohen\u0026apos;d\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003eMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003eCohen\u0026apos;d\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003eMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003eCohen\u0026apos;d\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e95% CI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003ePHQ-9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eAI vs. Human\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e-0. 41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-2. 95,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2. 12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e-0. 37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-2. 79,\u003c/p\u003e\n \u003cp\u003e2. 05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e2. 06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e0. 55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-0. 84,\u003c/p\u003e\n \u003cp\u003e4. 96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eAI vs. Control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e-1. 80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-4. 27,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0. 67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e-3. 47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-5. 75,-\u003c/p\u003e\n \u003cp\u003e1. 05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e-3. 27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e-0. 65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-6. 09,\u003c/p\u003e\n \u003cp\u003e-0. 44\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eHuman vs. Control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e-1. 39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-3. 92,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e1. 15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e-3. 10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-5. 45,-\u003c/p\u003e\n \u003cp\u003e0. 61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e-5. 33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e-1. 15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-8. 23,\u003c/p\u003e\n \u003cp\u003e-2. 43\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eBCI-CV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eAI vs. Human\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e0. 974\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-1. 05,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e3. 00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e0. 874\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-0. 93,\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;2. 68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e0. 837\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e0. 35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-0. 97,\u003c/p\u003e\n \u003cp\u003e2. 65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eAI vs. Control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e1. 467\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-0. 50,\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;3. 44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e0. 700\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-1. 06,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2. 46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e-0. 500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e-0. 17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-2. 26,\u003c/p\u003e\n \u003cp\u003e1. 26\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eHuman vs. Control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e0. 493\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-1. 53,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e2. 52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e-0. 174\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-1. 98,\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;1. 63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e-1. 337\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e-0. 45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-3. 15,\u003c/p\u003e\n \u003cp\u003e0. 47\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003eGSES\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eAI vs. Human\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e-2. 237\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-6. 19,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e1. 71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e0. 922\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-1. 88,\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;3. 72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e-0. 711\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e-0. 10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-5. 00,\u003c/p\u003e\n \u003cp\u003e3. 58\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eAI vs. Control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e-0. 067\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e-0. 01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-3. 91,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e3. 78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e1. 067\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-1. 66,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e3. 79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e4. 167\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e0. 65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e-0. 01,\u003c/p\u003e\n \u003cp\u003e8. 34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 85px;\"\u003e\n \u003cp\u003eHuman vs. Control\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 62px;\"\u003e\n \u003cp\u003e2. 170\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-1. 78,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e6. 12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 56px;\"\u003e\n \u003cp\u003e0. 144\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 54px;\"\u003e\n \u003cp\u003e0. 03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 58px;\"\u003e\n \u003cp\u003e-2. 65,\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;2. 94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 55px;\"\u003e\n \u003cp\u003e4. 878\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 49px;\"\u003e\n \u003cp\u003e0. 74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003e0. 59,\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e9. 16\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eAbbreviations: MD, mean difference; AI, AI intervention group; Human, Human intervention group; Control, control group with no intervention.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNote. T0 = baseline; T1 = week 2; T2 = week 4.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4\u0026nbsp;\u003c/strong\u003eQualitative Themes and Subthemes: AI Intervention vs. Human Intervention\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"111%\" class=\"fr-table-selection-hover\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eTheme\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eSubtheme\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003eAI group\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10px;\"\u003e\n \u003cp\u003eHuman group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eConcept\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eProcess\u003c/p\u003e\n \u003cp\u003e(Advantages)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA1: Reduced time/location constraints\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eCost-effective, saves time\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA2: Reduced privacy concerns\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eSafe environment, increased self-disclosure\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA3: Reduced stigma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eReduced psychological burden\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA4: Sense of interaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eProvides an outlet for sharing/feeling heard\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA5: Broad information resources\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eExtensive information coverage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA6: Comprehension ability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFlexible counseling direction/captures semantics and key issues\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003e(Disadvantages)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA7: Weak empathy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eLimited ability to understand circumstances, identify emotional fluctuations, and provide emotional validation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA8: Idealistic advice\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eImpractical suggestions\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA9: Weak recognition ability\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eLimited capacity to process language and perceive emotions\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA10: Mechanistic language\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eInflexible and impersonal tone and response style\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA11: Weak sense of communication\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eFormulaic counseling style, poor flow, slow response time\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA12: Unproductive probing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eLack of direction in guidance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eEffectiveness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA13: Providing advice\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eOffers suggestions for practical problems\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA14: Emotional support\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eUnderstands client\u0026apos;s perspective and improves mood\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA15: Thought-provoking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eIdentifies irrational beliefs and introduces new possibilities\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA16: Achieved desired outcomes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eMet expectations for counseling outcomes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA18: Partially helpful\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eOnly addressed surface-level issues\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 18px;\"\u003e\n \u003cp\u003eFuture\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA19: Preference for AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003ePersonalized matching, accessibility, stability\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA20: Preference for human\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eProfessionalism, emotional needs met, rich content exploration\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA21: Application and promotion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003ewidespread adoption and dissemination\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 18px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26px;\"\u003e\n \u003cp\u003eA22: Scope of application\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 8px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10px;\"\u003e\n \u003cp\u003e30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 35px;\"\u003e\n \u003cp\u003eSuitable for mild to moderate problems, short-term interventions\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eNote. \u003csup\u003ec\u003c/sup\u003e The number represents the amount of material with consistent content obtained from node materials of different texts.\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-psychiatry","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bpsy","sideBox":"Learn more about [BMC Psychiatry](http://bmcpsychiatry.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bpsy/default.aspx","title":"BMC Psychiatry","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"AI, depression, suicide ideation, internet-based cognitive behavioral therapy, randomized controlled trial, young adults, user experience","lastPublishedDoi":"10.21203/rs.3.rs-8269763/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8269763/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e This randomized controlled trial compared AI-driven and human peer counselor-delivered internet-based Cognitive Behavioral Therapy (iCBT) for depressive symptoms (primary outcome) in young adults. Addressing a gap in literature, we explored the comparative effectiveness and acceptability/perception of these two modalities of online intervention.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eNinety young adults were randomized to AI-driven iCBT, human peer counselor iCBT (participant-blinded), or a waitlist control. Interventions consisted of eight hours over four weeks. Depressive symptoms (primary outcome), suicidal ideation, and self-efficacy were assessed at baseline, week two, and week four. Qualitative analysis explored participant perceptions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e Both iCBT interventions significantly reduced depressive symptoms (\u003cem\u003ep\u003c/em\u003e\u0026lt; .05). No significant difference was observed between intervention groups at week two. The AI group improved significantly from baseline to week two (\u003cem\u003ep\u003c/em\u003e=0.004) but showed no further significant reduction by week four (a plateau effect), while the human group demonstrated continued improvement. Qualitative analysis indicated that participants valued AI’s convenience and accessibility but expressed concerns regarding its emotional understanding and personalization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e AI-iCBT shows promising short-term efficacy, comparable to human counseling up to week two, but its limitations in emotional perception and sustaining therapeutic momentum resulted in a plateau effect. Future AI development must focus on improving emotional interaction and personalized support.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTrial registration\u003c/strong\u003e: ChiCTR2400088423. Registered on 19 August 2024.\u003c/p\u003e","manuscriptTitle":"A Pilot Randomized Controlled Trial of AI-Delivered vs. Human-delivered iCBT for Depression in Young Adults","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-30 09:56:59","doi":"10.21203/rs.3.rs-8269763/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-02-05T07:20:43+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-02T18:51:22+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-29T03:44:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"329982257739563359389852596846879648464","date":"2025-12-30T15:27:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"115813311616654390428222581997978678735","date":"2025-12-28T18:17:12+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"119333158503867749453139239143009065052","date":"2025-12-27T17:12:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"243342950942549731309433867801408423096","date":"2025-12-26T15:10:43+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-12-25T10:30:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-25T10:10:15+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-12-10T13:45:06+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-12-10T06:03:56+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Psychiatry","date":"2025-12-10T05:54:38+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-psychiatry","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bpsy","sideBox":"Learn more about [BMC Psychiatry](http://bmcpsychiatry.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bpsy/default.aspx","title":"BMC Psychiatry","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"90d3ab42-5622-419f-9021-56991582278e","owner":[],"postedDate":"December 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-03-02T16:01:07+00:00","versionOfRecord":{"articleIdentity":"rs-8269763","link":"https://doi.org/10.1186/s12888-026-07925-1","journal":{"identity":"bmc-psychiatry","isVorOnly":false,"title":"BMC Psychiatry"},"publishedOn":"2026-02-24 15:57:58","publishedOnDateReadable":"February 24th, 2026"},"versionCreatedAt":"2025-12-30 09:56:59","video":"","vorDoi":"10.1186/s12888-026-07925-1","vorDoiUrl":"https://doi.org/10.1186/s12888-026-07925-1","workflowStages":[]},"version":"v1","identity":"rs-8269763","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8269763","identity":"rs-8269763","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}