AI-Assisted Solution-Focused Counseling Training for Novice Mental Health Educators: An Exploratory Study

AI-Assisted Solution-Focused Counseling Training for Novice Mental Health Educators: An Exploratory Study | 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 AI-Assisted Solution-Focused Counseling Training for Novice Mental Health Educators: An Exploratory Study Huabing Liu, Nayila Tuerxun, Jing Chen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9368123/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Teachers increasingly serve as first responders to student mental health crises, yet pre-service teacher education often lacks opportunities for realistic, ethically safe practice of mental health support skills. This exploratory study evaluated a theoretically grounded dual-agent AI simulation system designed to train novice educators in Solution-Focused Brief Therapy (SFBT) skills. Using a quasi-experimental pretest-posttest design, 30 non-psychology graduate education students were divided into an AI-assisted practice group (n = 18) and a peer role-play control group (n = 12) according to class enrolment. The AI system was built around Kolb’s ( 1984 ) experiential learning cycle, operationalized through two interactive spaces: a Demonstration Space and a Practice Space. Both groups showed significant pre-to-post gains in SFBT knowledge and counselling self-efficacy. The AI-assisted group demonstrated a marginal advantage in objective knowledge acquisition, whereas self-efficacy gains were comparable across groups. Within the experimental group, certain AI system evaluations were positively associated with self-efficacy gains. Student feedback was largely positive, though some noted areas for improvement in the system’s authenticity and interactivity. These findings suggest that generative AI simulation offers a scalable and ethically safe complement to traditional peer role-play for foundational mental health support training. The observed asymmetry between knowledge and self-efficacy gains further highlights the nuanced effects of technology-mediated learning on trainee self-perception, with implications for the design of AI-assisted curricula in teacher education and mental health training. AI simulation Solution-Focused Brief Therapy Pre-service teacher education School mental health support training Self-efficacy Figures Figure 1 Figure 2 Figure 3 Figure 4 1 Introduction Mental health difficulties among school-age populations have received increasing attention from policymakers and educators worldwide. The World Health Organization (p. 41, 2022) estimates that one in seven adolescents meets criteria for a mental health disorder, placing considerable demands on school-based support systems. This crisis is particularly acute in China, where prevalence of any mental disorders among school children and adolescents was 17.5% (Li et al., 2022 ). Teachers are frequently the first point of contact for students in distress, and are increasingly expected to provide initial mental health support before professional services become available (Reinke et al., 2011 ). Preparing pre-service teachers to fulfil this role effectively has consequently become a priority in teacher education, yet current training practices leave substantial gaps between theoretical knowledge and applied competence. 1.1 The Theory-Practice Gap in Mental Health Support Training Teacher preparation programs typically provide grounding in developmental psychology and school psychology, but offer limited structured opportunities to apply these concepts in realistic interactions (Koller & Bertel, 2006 ; Reinke et al., 2011 ). Developing genuine mental health support (MHS) competence goes well beyond merely acquiring declarative facts. Instead, the development of school mental health support skills needs active exposure to diverse student cases, accompanied by immediate corrective feedback and sustained deliberate practice (Ericsson, 2004 ). Several evidence-based frameworks have been adapted for educational contexts, including Solution-Focused Brief Therapy, Motivational Interviewing, and strengths-based approaches (Kim & Franklin, 2009 ). Mastery of these techniques requires repeated application with diverse presentations and timely feedback, conditions that conventional teacher education struggles to provide (Mazzer & Rickwood, 2015 ). Peer role-play, the most common practice modality, often fails to replicate the complexity of genuine teacher-student interactions (Rønning & Bjørkly, 2019 ; Gorski et al., 2022 ). Two further constraints compound these limitations. First, mental health support skills training in teacher education operates under severe time pressure: a typical Master of Education program in China allocates only 24–64 hours to this area, compared with the hundreds of supervised practice hours required in professional counselling training (Council for Accreditation of Counseling and Related Educational Programs, 2024 ). Second, having novice teachers practice emerging MHS skills with students experiencing genuine distress raises profound ethical concerns about potential harm in the absence of competence and supervision (Johnson, 2008). These constraints together create a need for practice environments that are scalable, ethically safe, and pedagogically effective. 1.2 Simulation-Based Training as a Pedagogical Bridge To bridge this theory-practice gap without compromising ethical standards, simulation-based training has established itself as an evidence-based pedagogy. In healthcare education, standardized patients have been utilized in the medical field since the 1960s (Flanagan & Cummings, 2023 ), and computerized virtual patients have demonstrated robust effectiveness across numerous trials (Kononowicz et al., 2019 ). This tradition has successfully extended to mental health contexts, where virtual patient simulations increase listening skills, self-efficacy and diagnostic accuracy in screening and assessment (Tanana et al., 2019 ; Washburn et al., 2016 ; Washburn et al., 2020 ). Similarly, in teacher education, clinical simulations have been employed and proved to be beneficial for teachers’ skill development (Dotger, 2013 ; De Coninck et al., 2023 ; Chernikova et al., 2020 ). By providing a safe environment for trial-and-error learning, simulations allow trainees to develop foundational competence while shielding actual humans from the ethical risks associated with novice mistakes. Furthermore, this method effectively bypasses the inherent vulnerabilities of traditional peer role-play. While peer exercises are frequently compromised by time constraints and a lack of realism (Nisar et al., 2021 ; Nestel & Tierney, 2007 ), simulations offer an anytime and anywhere training environment. However, these earlier simulation approaches nevertheless faced practical constraints that limited their applicability in teacher education. Standardized patient programs require trained actors, scheduling infrastructure, and supervisory resources that are difficult to sustain within teacher preparation curricula (Kaplonyi et al., 2017 ). which limited conversational range and ecological validity (Kononowicz et al., 2019 ). Furthermore, barriers such as a lack of transferable learning resources prevent general educators from effectively implementing these complex tools (Wu et al., 2022 ). Unlike these predecessors, generative AI enables the highly natural and flexible conversational dynamics essential to authentic counselling practice (Demszky et al., 2023 ). Furthermore, these modern models effectively dismantle previous technical barriers. Teacher educators without programming expertise can now design and customize sophisticated simulated students using prompts in natural language (Mollick & Mollick, 2023 ). Indeed, a growing body of work has begun to examine how such tools can be integrated into counselling skill training, with initial evidence suggesting that AI-assisted practice can enhance self-efficacy and counseling skills through simulated role-play training (Jeong et al., 2025 ; Maurya, 2024 ). While early evidence-based studies demonstrate the feasibility of AI in clinical training, the broader research landscape remains in its infancy. The theoretical frameworks and specific pedagogy necessary to guide such practice are still lacking (Maurya & DeDiego, 2025 ). Therefore, the current study investigates whether a theoretically grounded, dual-agent AI system can effectively support foundational skill acquisition and self-efficacy among novice educators. 1.3 Solution-Focused Brief Therapy as an Entry Point Originally developed by Steve de Shazer and Insoo Kim Berg, Solution-Focused Brief Therapy (SFBT) emphasizes clients’ existing strengths through structured techniques including the miracle question, scaling questions, and exception-finding (De Shazer & Berg, 1997 ). SFBT’s concrete, stepwise techniques are highly accessible to beginners (De Shazer et al., 2021 ), and its non-pathologizing orientation is well suited to school settings (Kim & Franklin, 2009 ). We selected SFBT as the initial focus skill because its accessibility matches our pre-service teacher sample lacking prior psychology backgrounds. Furthermore, its discrete techniques allow objective assessment, and its structured dialogue patterns are highly amenable to AI simulation with pedagogical scaffolding. While SFBT serves as an ideal foundational test case for this study, the underlying simulation architecture established here holds significant potential for future adaptation to other established therapeutic frameworks. 1.4 Theoretical Framework and Research Variables Existing studies on generative AI-based simulation counseling skills training have rarely been anchored in explicit learning theory, a gap that limits the interpretability of findings and the transferability of design principles (Maurya & DeDiego, 2025 ). The present intervention was developed around Kolb’s ( 1984 ) experiential learning cycle, which comprises four phases: concrete experience, reflective observation, abstract conceptualization, and active experimentation. This framework is well suited to simulation-based skill training because its core logic, iterative cycles of experience and reflection, aligns with what generative AI simulation can offer: repeated practice opportunities with immediate, structured feedback that conventional instruction cannot readily sustain (Ericsson, 2008). The system architecture through which this cycle is operationalized is described in Section 2.4 . 1.4.1 Outcome Variables: Objective Knowledge vs. Subjective Self-Efficacy Two outcomes were selected to evaluate whether the experiential learning cycle, as operationalized through the simulation system, produces meaningful skill development. SFBT knowledge captures the declarative and procedural competence that training aims to develop. Self-efficacy refers to a practitioner's belief in their capacity to perform specific support tasks (Bandura, 1997 ). Humans with higher self-efficacy are more likely to practice with their learned knowledge (Artino, 2012 ; Mazzer & Rickwood, 2015 ). Interestingly, these two outcomes are not assumed to move in parallel. When self-efficacy outpaces actual skill, a pattern documented in both medical education and professional training (Moore & Healy, 2008 ; Dunning, 2011 ), practitioners may engage in overconfident and potentially harmful practice. Whether simulation-based and peer-based training yield different patterns of knowledge and self-efficacy calibration is therefore an empirical question the present study is positioned to examine. 1.4.2 Mechanisms of Learning: Practice Duration vs. System Evaluation While establishing main effects is essential, identifying the specific mechanisms driving AI-mediated learning remains theoretically crucial for informing future design. Two candidate mechanisms are examined, each corresponding to a distinct prerequisite for completing the cycle. Practice duration reflects the degree of investment in the active experimentation phase. Without sufficient engagement at this stage, subsequent reflection and conceptualization are unlikely to be triggered regardless of the quality of available feedback. Although extended engagement is associated with achievement gains in educational settings more broadly (Natividad-Sancho et al., 2024 ), whether this holds within iterative simulation environments remains an open question. Drawing on the Technology Acceptance Model (Davis, 1989 ), learners who reject the system’s instructional utility are unlikely to engage substantively with the reflective and conceptualization phases of the cycle, irrespective of time invested (Al-Fraihat et al., 2020 ). Therefore, we explore candidate mechanisms. 1.5 The Present Study Responding to school mental health needs and research gaps, this study designed a dual-agent AI system to support SFBT training within the practical constraints of pre-service teacher education and therefore evaluate its effectiveness and explore potential mechanisms. The study aims to answer the following three questions: RQ1. Do students using the AI system show significantly greater improvements in SFBT knowledge and MHS self-efficacy compared to those in the peer role-play group? RQ2. Within the AI-simulation group, are systems using duration and students’ evaluations of the AI system associated with their knowledge and self-efficacy change? RQ3. How do pre-service teachers perceive the simulation system as a tool for MHS skill development? 2 Method 2.1 Research Design This study employed a quasi-experimental, pretest-posttest design with non-equivalent control groups to evaluate the AI intervention’s effectiveness. Students were assigned to conditions based on existing class sections. 2.2 Participants Participants were graduate students enrolled in a mandatory 32-hour school psychology course at a Chinese university. All held undergraduate degrees in non-psychology majors and reported no prior counseling training. In the previous semester, students completed a course on student psychological development that covered developmental theories but not clinical methods or therapeutic techniques. Initially, 35 students voluntarily registered for the study (experimental group: n = 20; control group: n = 15). However, five students were excluded from the final analysis due to incomplete data (i.e., missing either pre- or post-test assessments), resulting in a final sample of 30 participants (experimental: n = 18; control: n = 12). Despite this attrition, the groups remained comparable in academic performance, with final course grades differing by only 0.3 points on a 100-point scale (t = 0.421, p = 0.677). All procedures were approved by the university’s Institutional Review Board (# No. H20250564I), and informed consent was obtained from all participants. 2.3 Procedure The intervention occurred during Week 10, when the course covered SFBT. Both groups received the same 70-minute lecture on SFBT principles and applications in school settings. After the lecture, students participated in a 20-minute practice session. The experimental group began with a 15-minute instructor-led demonstration, during which the teacher modeled the use of the AI system by simulating a counseling interaction with the agent. Students observed this demonstration before engaging in approximately 5 minutes of independent practice. The control group practiced through peer role-play for the full 20 minutes. Instructors encouraged but did not require out-of-class practice for both groups. Pre-tests were administered one week before the intervention (Week 9), and post-tests were administered one week after (Week 11). Participants who completed both assessments received a small incentive (approximately 10 RMB). After post-test data collection, the control group received access to the AI system. 2.4 The AI-Assisted Practice System 2.4.1 System Architecture We developed a dual-agent AI system on the COZE platform comprising two functional spaces. Rather than following a fixed instructional sequence, the two spaces serve as flexible entry points into Kolb’s (1984; 1995) experiential learning cycle, which are concrete experience, reflective observation, abstract conceptualization, and active experimentation. Learners may enter either space at any point and move between them repeatedly, with each cycle of use informing the next. Demonstration Space. In this space, the learner takes the role of a help-seeker while an AI counsellor demonstrates SFBT techniques in response to a problem the learner presents. This reverse role-play simultaneously provides a concrete experience of the mental health interaction from the student perspective and supports reflective observation of expert technique application (Bandura, 1977). Learners may return to this space after encountering difficulty in practice, using fresh observational experience to revise their understanding before re-attempting. Practice Space. This space supports active experimentation. Two coordinated agents are deployed. A Student Simulation Agent, which portrays adolescents presenting varied psychological concerns (academic stress, peer conflict, family difficulties) with graduated complexity. An Assessment Agent, which analyzes conversation transcripts and delivers immediate feedback structured across three levels (Hattie & Timperley, 2007): task level (identifying correct and incorrect technique application), process level (explaining how specific moves align with SFBT principles), and self-regulation level (prompting learners to monitor and adjust their approach). Each practice attempt constitutes a new concrete experience. The Assessment Agent’s feedback transforms that experience into reflective observation and abstract conceptualization, enabling the learner to enter the next attempt with revised understanding. Repeated use of this space thus generates successive Kolb’s cycles within a single session. Because neither space is a prerequisite to the other, learners can move between them in response to their own learning needs. Learners can return to the Demonstration Space when practice reveals gaps in understanding, or move directly to practice after observation to test what they have seen. 2.4.2 Agent Development Knowledge Base . The agents’ knowledge base included articles on SFBT in educational settings, and representative books focused on SFBT (e.g., De Shazer, 1985; Berg & Shilts, 2005; Kim, 2008). The first author, who is also a licensed counseling psychologist, interacted with the agent to ensure its professionalism in SFBT. Agent Development . The agent prompts were developed through several iterations. Initially, the prompts were based on SFBT principles (Kim & Franklin, 2009) and feedback frameworks (Hattie & Timperley, 2007). An early single-agent design was later separated into distinct simulation and assessment functions to avoid role confusion. Next, three graduate students pilot-tested the system. Based on their feedback, the prompts and models were further refined. Finally, the agents were designed as follows. The Student Simulation Agent was instructed to: (a) present concerns at varying difficulty levels; (b) respond differently to well-executed versus poorly-executed techniques; (c) avoid resolving issues too quickly. The Assessment Agent was instructed to: (a) identify specific SFBT techniques used (e.g., scaling questions, exception-finding, goal-setting); (b) provide feedback at task, process, and self-regulation levels; (c) limit responses to approximately 200 words. 2.5 Measures 2.5.1 SFBT Knowledge Test Given the lack of established SFBT knowledge measures, we developed a situational judgment test based on the suggestions in Hosany, Wellman & Lowe (2007) and Ferraz & Wellman (2009), comprising three scenarios measuring core SFBT and mental health support competencies: focusing on problems versus solutions, focusing on patients’ current strengths and resources, and managing therapeutic silence. Two experienced school mental health practitioners reviewed items for content validity. Each scenario presented a student concern with four response options, one reflecting SFBT- and counseling-consistent approaches. Scores ranged from 0-3, with higher scores indicating better SFBT technique recognition. 2.5.2 Self-Efficacy Scale We adapted the General Self-Efficacy Scale (Schwarzer & Jerusalem, 1995) to assess participants’ self-efficacy in student mental health support. The 10-item scale used 4-point Likert ratings (1 = completely incorrect, 4 = completely correct), with total scores ranging from 10-40. The internal consistency of the adapted scale is 0.86 (Cronbach’s α) in our sample. 2.5.3 AI System Evaluation Questionnaire For the experimental group, we designed a questionnaire assessing their experience with the AI system. The questionnaire included: AI Evaluation Questionnaire (8 items). The questionnaire included 8 items measuring perceived effectiveness of the AI agents. Example items include ‘helped me understand the core values of SFBT’ and ‘increased my learning interest and engagement’ (the complete list of items is presented in Figure 3 in the Results section). All items were rated on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Time Spent (2 items). Participants recalled the approximate time they spent on the Demonstration Space and the Practice Space respectively. Qualitative Feedback (1 open-ended question). We asked participants to identify the helpful aspects of the AI system and suggest improvements. Responses were analyzed thematically to complement quantitative findings. 2.6 Data Analysis Analyses were conducted in Python 3.9.6 (pandas, numpy, scipy, statsmodels) with α = .05. Shapiro–Wilk tests indicated that self-efficacy gain scores met normality assumptions in both groups, whereas knowledge gain scores violated normality in both groups. Levene’s tests confirmed homogeneity of variance. Independent t-tests confirmed no significant baseline differences between groups. For self-efficacy, we used paired t-tests for within-group pre-to-post comparisons, independent t-tests for between-group comparisons, and ANCOVA with pre-test scores as covariates after confirming homogeneity of regression slopes. For SFBT knowledge, which violated normality assumptions, we used Wilcoxon signed-rank tests for within-group comparisons and Mann–Whitney U tests for between-group comparisons, and ANCOVA was additionally conducted on post-test scores with pre-test scores as covariates. Given the small sample size, effect sizes were reported throughout (Cohen’s d for t-tests, partial η² for ANCOVAs) to facilitate interpretation independent of statistical significance. 3 Results Preliminary analyses and baseline equivalence are reported in Section 3.1 , followed by a description of agent usage patterns in the experimental group (Section 3.2 ). Sections 3.3 and 3.4 report intervention effects on SFBT knowledge test scores and student mental health support self-efficacy, respectively. Each section includes between-group comparisons, ANCOVA, and within-group analyses examining agent usage as a potential predictor of gains (see Table 1 ). Student written responses are cited where they directly bear on a finding. 3.1 Preliminary Analyses Distributional assumptions . Shapiro–Wilk tests indicated that self-efficacy gain scores were normally distributed in both the experimental (W = 0.949, p = .417) and control groups (W = 0.966, p = .868). SFBT Test score gains violated normality in both groups (experimental: W = 0.850, p = .008; control: W = 0.768, p = .004); Wilcoxon signed-rank and Mann–Whitney U tests were therefore used for test score outcomes, with parametric tests retained for self-efficacy. Baseline equivalence. Independent-samples t-tests confirmed that the experimental (n = 18) and control (n = 12) groups did not differ significantly on pre-test knowledge scores (t (28) = − 0.812, p = .424, d = 0.302) or self-efficacy (t (28) = 0.555, p = .584, d = − 0.207; Table 2 ). Levene’s tests confirmed homogeneity of variance (all ps > .20). The groups were treated as equivalent at baseline. 3.2 Agent Usage Patterns in the Experimental Group 3.2.1 Time Investment In the Practice Space, 78% of students spent 15 minutes or more (Levels 3–4; M = 2.83, SD = 0.71); only one student (5.6%) spent fewer than five minutes. In the Demo Space, time investment was shorter and more variable: 22.2% spent fewer than five minutes, and equal proportions (38.9% each) fell in the 5–15 and 15–30 minute bands (M = 2.17, SD = 0.79). The two duration variables were not significantly correlated (Spearman ρ = .338, p = .169). 3.2.2 Student Perceptio ns of the Agent Post-intervention evaluations of the agent were positive overall (8-item scale mean: M = 3.74, SD = 0.48). The highest-rated items were ‘facilitates SFBT skill acquisition’ and ‘enhances course interest and engagement’ (both M = 4.00, SD = 0.59; 83.3% agreement). Items addressing understanding of SFBT core values (M = 3.83; 72.2% agreement) and strengthening SFBT identification (M = 3.78; 72.2% agreement) also received relatively high ratings. Sixty-one percent of students agreed that agent practice increased their confidence in working with real students (M = 3.67, SD = 0.77), while 38.9% remained neutral or disagreed. Fifty-six percent expressed a desire for agent use in future courses (M = 3.61, SD = 0.92). One student affirmed the agent’s contribution to skill development directly. They commented “ Using the Practice Space agent genuinely improved my ability to support students with psychological difficulties! ”(translated from Chinese). ‘More effective than in-class role-play’ and ‘willing to use independently after class’ were the two lowest-rated items (both M = 3.50; 44.4% agreement), with half of students neutral on comparative effectiveness. Several students pointed out that AI practice are different from practice in reality. The agent is sometimes wooden, [because] it repeats the same point, whereas real students’ situations are [more] varied. At times, it gives the feeling of playing a game rather than genuinely solving a problem. (translated from Chinese) Another student noted the absence of in-built procedural guidance: For those training for the first time, perhaps an optional example panel or step-by-step instruction bar could be introduced. Practising with a reference [guide] can help consolidate memory; once learners gain proficiency, they can choose to remove the guidance. (translated from Chinese) Their feedback indicates participants’ feelings of using AI agents with their practice and highlights areas for future improvement. 3.3 Effects on SFBT Knowledge Test Scores 3.3.1 Within-Group Pre-to-Post Change Both groups showed significant gains in SFBT knowledge from pre- to post-intervention (Table 1 ). The experimental group improved reliably (W = 10, p = .003; Mdn gain = 1.0, IQR [0.2, 2.0]), as did the control group (W = 0, p = .031; Mdn gain = 0.5, IQR [0.0, 1.0]). Table 1 SFBT Test Score Outcomes by Group Measure Control Experimental Between-Group Comparison Mdn IQR Mdn IQR U p Effect Size (r) Pre-test 1.0 [1.0, 2.0] 1.0 [0.0, 2.0] 124.0 .490 .148 Post-test 2.0 [1.0, 2.0] 2.0 [2.0, 3.0] 147.5 .079 .366 Within-group test (Wilcoxon) W = 0, p = .031* W = 10, p = .003* / ANCOVA controlling for pre-test: F (1, 27) = 3.302, p = .080, partial η² = .109; Madj, ctrl = 1.660, Madj, exp = 2.171; model R² = .347 Note . Test scores were non-normally distributed (Shapiro–Wilk p < .05); * p < .05 3.3.2 Between-Group Comparisons and ANCOVA The Mann–Whitney test found no significant difference in test score gains (U = 137.5, p = .195, rank-biserial r = − 0.273), though post-test scores showed a marginal trend favouring the experimental group (U = 147.5, p = .079, rank-biserial r = − 0.366). After controlling for pre-test scores, the group effect approached but did not reach conventional significance, F (1, 27) = 3.302, p = .080, partial η² = .109, with the experimental group scoring higher after adjustment The pre-test covariate was significant (F (1, 27) = 9.035, p = .006), the homogeneity-of-slopes assumption was met (F = 4.165, p = .052), and the model explained 34.7% of post-test variance. 3.4 Effects on Self-Efficacy Table 2 Self-Efficacy Outcomes by Group Measure Control Experimental Between-Group Comparison M SD M SD t p Effect Size (d) Pre-test 21.25 4.00 20.06 6.68 .56 .583 −.207 Post-test 24.33 5.03 23.61 5.89 .35 .731 −.130 Within-group test (paired t) t (11) = 3.120, p = .010*, d = 0.901 t (17) = 3.625, p = .002*, d = 0.854 ANCOVA controlling for pre-test: F (1, 27) = 0.014, p = .908, partial η² = .001; Madj, ctrl = 23.804, Madj, exp = 23.964; model R² = .588 Note . * p < .05. 3.4.1 Within-Group Pre-to-Post Change Both groups showed significant gains in student mental health support self-efficacy (Table 2 ). The experimental group gained a mean of 3.56 points (SD = 4.16, 95% CI [1.49, 5.63]), t (17) = 3.625, p = .002, d = 0.854. The control group gained a mean of 3.08 points (SD = 3.42, 95% CI [0.91, 5.26]), t (11) = 3.120, p = .010, d = 0.901. 3.4.2 Between-Group Comparisons and ANCOVA Post-test self-efficacy (t (28) = 0.35, p = .731, d = − 0.130) did not differ significantly between groups. ANCOVA confirmed a non-significant group effect (F (1, 27) = 0.014, p = .908, partial η² = .001), with adjusted means of 23.964 (experimental) and 23.804 (control). Pre-test self-efficacy accounted for most of the explained variance (F (1, 27) = 38.302, p < .001, R² = .588), and the homogeneity-of-slopes assumption was satisfied (F = 0.559, p = .461). 3.5 Correlations Table 3 Spearman Rank-Order Correlations Among Agent Usage Variables and Outcome Gains Variable 1 2 3 4 5 1. AE — 2. PD .271 — 3. DD .375 .338 — 4. TG .236 .129 .021 — 5. EG .234 .382 .435† − .021 — M 3.74 2.83 2.17 0.94 3.56 SD 0.48 0.71 0.79 1.00 4.16 Note . AE = Agent Overall Evaluation; PD = Practice Space Duration; DD = Demo Space Duration; TG = Test Score Gain; EG = Self-Efficacy Gain. M and SD for each variable shown in bottom rows. †p < .10 Within the experimental group, no agent usage variable was significantly associated with test score gain: agent overall evaluation (ρ = .236, p = .345), Practice Space duration (ρ = .129, p = .611), and Demo Space duration (ρ = .021, p = .934). Demo Space duration showed a marginally significant positive correlation with self-efficacy gain (ρ = .435, p = .071). Practice Space duration (ρ = .382, p = .118) and agent overall evaluation (ρ = .234, p = .349) were positively but non-significantly associated with gains. Three evaluation items such as ‘the agent enhanced my interest and engagement in the course’ have positive significant correlation relationships with self-efficacy gain (Table 4 ), and no evaluation items have significant correlation relationships with test score gain. Table 4 Spearman Correlations Between AI Agent Evaluation Items and Self-Efficacy Gain Item Statement ρ p 1 The agent helped me understand the core values of SFBT. 0.410 .091 2 Practising with the agent helped me master SFBT skills. 0.168 .505 3 The agent strengthened my identification with SFBT principles. 0.352 .152 4 Agent practice was more effective than in-class role-play. 0.092 .717 5 The agent enhanced my interest and engagement in the course. 0.504 .033* 6 I am willing to use the agent independently after class. 0.194 .442 7 My confidence in real counselling improved after using the agent. 0.553 .017* 8 I hope future courses will also be equipped with practice agents. 0.571 .013* Note . Spearman rank-order correlations. Self-Efficacy Gain = post-test minus pre-test. * p < .05. 4 Discussions 4.1 Discussion of Main Findings Pre-post and between-group comparisons consistently indicated that AI-assisted psychology practice yielded comparable or superior outcomes relative to traditional classroom practice in developing students’ mental health support skills and self-efficacy. This pattern aligns with our educational hypotheses articulated in the introduction and method sections, and corroborates studies demonstrating that AI-simulated training enhances learning outcomes in counseling education (Jeong et al., 2025 ; Maurya, 2024 ). The present study extends this literature to the domain of solution-focused counseling training, a domain previously dominated by didactic instruction and conventional role-play methods. The study further demonstrates the applicability of such AI agent designs to novice educators in mental health support contexts, an area that has received limited empirical attention. Notably, meaningful gains in self-efficacy were observed even when training was confined to a single counseling theory. Interestingly, the two outcome variables exhibited divergent patterns of group differences. The experimental group demonstrated a marginal advantage on the knowledge test following covariate adjustment, whereas the group effect on self-efficacy was negligible. This asymmetric pattern, whereby knowledge gains outpaced self-efficacy gains in the AI-assisted condition, warrants closer interpretive attention. We offer two tentative, and potentially complementary, explanations regarding the knowledge-efficacy asymmetric pattern. The first draws on the Dunning-Kruger Effect, which describes the tendency for individuals to overestimate their competence in early stages of learning, followed by a decline in perceived confidence as awareness of domain complexity increases (Kruger & Dunning, 1999 ). Evidence from medical and professional training similarly suggests that more competent trainees frequently report lower self-efficacy than less proficient peers, particularly during early-to-intermediate stages of skill development (Tzamaras et al., 2024 ). From this perspective, the experimental group’s more substantial knowledge acquisition may have simultaneously expanded their awareness of the complexity inherent in solution-focused theory, thereby tempering confidence gains that might otherwise have accompanied skill development. Conventional role-play, by contrast, affords social validation and peer affirmation that may sustain or inflate perceived competence independently of objective performance. The second explanation concerns the perceived authenticity of AI simulations. Student feedback indicated that some participants found the AI interlocutor mechanical rather than genuinely humanistic, which may have attenuated confidence gains irrespective of actual skill improvement. This interpretation is consistent with broader evidence that learners’ trust in, and engagement with, technology-mediated interactions is sensitive to perceived anthropomorphism and relational authenticity (Ho et al., 2018 ; Zhang et al., 2024 ). These two mechanisms are not mutually exclusive; they may operate simultaneously, apply differentially across students depending on prior technology experience or interpersonal orientation, or interact in ways that produce the aggregate pattern observed here. Systematic investigation of their relative contributions remains an important direction for future research. With respect to engagement metrics, neither practice time nor self-efficacy gains showed a significant association, nor did practice time predict knowledge gains. Observation time, however, showed a tentative positive association with self-efficacy but not with knowledge acquisition. This pattern is theoretically coherent within Bandura’s (1977) social cognitive framework. Observation of modeled performance serves as a vicarious mastery experience that may be particularly beneficial for novice learners in enhancing their perceived self-efficacy. Knowledge development, by contrast, tends to be slow and incremental (Falender & Shafranske, 2004 ). The non-significant role of practice time further suggests that other factors, such as engagement quality, learners’ readiness, may warrant further investigation Finally, as an exploratory finding, learners who evaluated the AI agent more favorably reported higher self-efficacy. This association has two potential interpretations. Students who perceived greater value in AI-mediated practice may have engaged more deeply with the simulated scenarios, thereby reinforcing confidence in their developing skills. Alternatively, those who explored the curriculum more thoroughly may have formed both more favorable appraisals of the agent and stronger self-efficacy beliefs through cumulative experience. The directionality and underlying mechanisms of this association remain to be clarified. 4.2 Limitations and Future Directions Several limitations should be considered when interpreting these findings. First, the sample was drawn from a single institution and comprised predominantly novice educators, which constrains the generalizability of results to other populations and contexts. Sample size was further limited by the natural classroom setting and the voluntary nature of participation, which together restricted statistical power and may have precluded the detection of subtler group differences. Second, the knowledge test contained a limited number of items with scarce reliability and validity. To our knowledge, no SFBT validated instrument currently exists, and the test used here represents a preliminary attempt at a SFBT test. Knowledge-related findings should therefore be treated as exploratory, and future research should prioritize the development of psychometrically sound assessment tools for this domain, alongside the incorporation of behavioral or observational measures of counseling competence. Third, both outcomes were assessed within a single training cycle and a specific counseling theory, leaving open questions about the ecological validity. Future studies should employ longitudinal designs and agents to practice multiple counseling theories to determine whether effects are sustained or prevalent. Fourth, the internal mechanisms underlying the observed training effects remain unclear. None of the engagement variables examined were consistently associated with knowledge or self-efficacy gains, which may partly reflect the limited sample size. Future research incorporating a richer set of process variables, such as engagement quality and interaction patterns with the AI agent, alongside larger samples, would enable more systematic investigation of how AI-mediated practice translates into competence development. 4.3 Implications The present findings have several implications for curriculum development and educational policy in teachers’ mental health training. First, the demonstrated effectiveness of AI-assisted practice suggests that institutions and program developers could incorporate AI simulation as a structured component of mental health skills training curricula. Given that meaningful gains in skill and self-efficacy can be achieved through training focused on a single counseling approach, even modestly resourced programs may benefit from targeted AI-assisted modules without requiring wholesale curriculum redesign. This is particularly relevant in educational contexts where opportunities for supervised practice are limited and ethical constraints restrict direct client contact for novice learners. At the policy level, these results provide preliminary empirical support for investing in the development of AI-based training tools for novice mental health educators, a population that has historically received limited attention in both research and professional development policy. Future curriculum frameworks should additionally consider the intentional placement of demonstration-based activities, given their observed association with self-efficacy development, as a low-cost and scalable design feature applicable across diverse training contexts. A third implication concerns the intentional sequencing of demonstration and practice within novice learner training. The present findings suggest that demonstration-based activities may warrant prioritization in the early stages of training. Practice-based components, while essential, may require a longer developmental window before their effects become fully apparent. Training programs and curriculum designers should therefore avoid drawing premature conclusions about the effectiveness of practice-based elements based solely on immediate post-training assessments. Instead, a staged curriculum design, one that leads with structured demonstration before transitioning to guided practice, may better align with the developmental needs of novice learners. This also highlights the importance of incorporating longitudinal assessment into program evaluation frameworks, so that the delayed but potentially substantial benefits of practice-based learning are not overlooked. 5 Conclusions This study examined the effectiveness of AI-assisted simulation in solution-focused counseling training for novice mental health educators, a population and domain that have received limited empirical attention. The findings demonstrate that AI-mediated practice yielded comparable or marginally superior outcomes relative to traditional role-play across both knowledge and self-efficacy measures, supporting the viability of AI simulation as a structured training modality in school mental health education. The divergent pattern observed between the two outcomes, whereby the AI-assisted group gained more on knowledge but not on self-efficacy, represents a theoretically meaningful finding that points to the complex interplay between competence development and self-perception in technology-mediated learning environments. To our knowledge, this study is among the first to evaluate AI-assisted training within a solution-focused counseling framework, and contributes a preliminary empirical foundation to an emerging area of research at the intersection of AI, school mental health education, and professional skill development. Beyond its methodological contributions, this study opens up a practical training scenario for novice educators operating in school-based mental health contexts, addressing a gap that has long existed in teacher education systems where opportunities for systematic psychological practice are often insufficient or absent. AI-assisted simulation thus holds promise not only as a research tool, but as an accessible and scalable means of strengthening the mental health support capacity of educators who occupy a critical yet underserved role in students’ well-being. Nevertheless, the exploratory nature of this work, combined with limitations in sample size and measurement, necessitates cautious interpretation of the findings. Future research should build on these results by employing larger and more diverse samples, developing psychometrically validated assessment tools for this domain, and systematically investigating the design features and learning mechanisms that underpin effective AI-mediated school mental health training. References Al-Fraihat, D., Joy, M., Masa'deh, R., & Sinclair, J. (2020). Evaluating E-learning systems success: An empirical study. Computers in Human Behavior, 102 , 67–86. https://doi.org/10.1016/j.chb.2019.08.004 Artino, A. R. (2012). Academic self-efficacy: From educational theory to instructional practice. Perspectives on Medical Education , 1, 76–85. https://doi.org/10.1007/s40037-012-0012-5 Bandura, A. (1997). Self-efficacy: The exercise of control . W. H. Freeman. Berg, I. K., & Shilts, L. (2005). Classroom solutions: Woww coaching . Brief Family Therapy Center. Chernikova, O., Heitzmann, N., Stadler, M., Holzberger, D., Seidel, T., & Fischer, F. (2020). Simulation-based learning in higher education: A meta-analysis. Review of Educational Research, 90 (4), 499–541. https://doi.org/10.3102/0034654320933544 Council for Accreditation of Counseling and Related Educational Programs. (2024). 2024 CACREP Standards. https://www.cacrep.org/for-programs/2024-cacrep-standards/ Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13 (3), 319–340. https://doi.org/10.2307/249008 De Coninck, K., Keppens, K., Valcke, M., Dehaene, H., De Neve, J., & Vanderlinde, R. (2023). Exploring the effectiveness of clinical simulations to develop student teachers’ parent-teacher communication competence. Research Papers in Education , 38 (1), 69–101. https://doi.org/10.1080/02671522.2021.1961291 De Shazer, S. (1985). Keys to solutions in brief therapy . W. W. Norton. De Shazer, S., & Berg, I. K. (1997). ‘What works?’ Remarks on research aspects of solution‐focused brief therapy. J ournal of Family therapy , 19 (2), 121-124. https://doi.org/10.1111/1467-6427.00043 De Shazer, S., Dolan, Y., Korman, H., Trepper, T., McCollum, E., & Berg, I. K. (2021). More than miracles: The state of the art of solution-focused brief therapy . Routledge. Demszky, D., Yang, D., Yeager, D. S., Bryan, C. J., Clapper, M., Chandhok, S., ... & Pennebaker, J. W. (2023). Using large language models in psychology. Nature Reviews Psychology , 2 (11), 688-701. https://doi.org/10.1038/s44159-023-00241-5 Dotger, B. H. (2013). Clinical simulations for teacher development: A companion manual for teachers . Emerald Publishing Limited. Dunning, D. (2011). The Dunning–Kruger effect: On being ignorant of one's own ignorance. In Advances in experimental social psychology (Vol. 44, pp. 247-296). Academic Press. https://doi.org/10.1016/B978-0-12-385522-0.00005-6 Ericsson, K. A. (2004). Deliberate practice and the acquisition and maintenance of expert performance in medicine and related domains. Academic medicine , 79 (Supplement_2), S70-S81. https://doi.org/10.1097/00001888-200410001-00022 Ferraz, H., & Wellman, N. (2009). Fostering a culture of engagement: an evaluation of a 2‐day training in solution‐focused brief therapy for mental health workers. Journal of Psychiatric and Mental Health Nursing , 16 (4), 326-334. https://doi.org/10.1111/j.1365-2850.2008.01374.x Flanagan, O. L., & Cummings, K. M. (2023). Standardized Patients in Medical Education: A Review of the Literature. Cureus , 15 (7), e42027. https://doi.org/10.7759/cureus.42027 Falender, C. A., & Shafranske, E. P. (2004). Clinical supervision: A competency-based approach . American Psychological Association. https://doi.org/10.1037/10806-000 Gorski, S., Prokop-Dorner, A., Pers, M., Stalmach-Przygoda, A., Malecki, Ł., Cebula, G., & Bombeke, K. (2022). The use of simulated patients is more effective than student role playing in fostering patient-centred attitudes during communication skills training: A mixed method study. BioMed Research International , 2022, 1498692. https://doi.org/10.1155/2022/1498692 Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77 (1), 81–112. https://doi.org/10.3102/003465430298487 Ho, A., Hancock, J., & Miner, A. S. (2018). Psychological, relational, and emotional effects of self-disclosure after conversations with a chatbot. Journal of Communication , 68 (4), 712-733. https://doi.org/10.1093/joc/jqy026 Hosany, Z., Wellman, N., & Lowe, T. (2007). Fostering a culture of engagement: A pilot study of the outcomes of training mental health nurses working in two UK acute admission units in brief solution-focused therapy techniques. Journal of Psychiatric and Mental Health Nursing , 14 , 688–695. https://doi.org/10.1111/j.1365-2850.2007.01161.x Jeong, D., Lee, H., Yang, H., Park, J. C., & Lee, S. M. (2025). AI integration in counseling training: Aiding counselors‐in‐training in self‐efficacy enhancement and anxiety reduction. Journal of Counseling & Development, 103 (3), 346–360. https://doi.org/10.1002/jcad.12560 Johnson, W. B., Elman, N. S., Forrest, L., Robiner, W. N., Rodolfa, E., & Schaffer, J. B. (2008). Addressing professional competence problems in trainees: Some ethical considerations. Professional Psychology: Research and Practice , 39 (6), 589–599. https://doi.org/10.1037/a0014264 Kaplonyi, J., Bowles, K. A., Nestel, D., Kiegaldie, D., Maloney, S., Haines, T., & Williams, C. (2017). Understanding the impact of simulated patients on health care learners’ communication skills: a systematic review. Medical education , 51 (12), 1209-1219. https://doi.org/10.1111/medu.13387 Kim, J. S. (2008). Examining the effectiveness of solution-focused brief therapy: A meta-analysis. Research on Social Work Practice , 18 (2), 107-116. https://doi.org/10.1177/1049731507307807 Kim, J. S., & Franklin, C. (2009). Solution-focused brief therapy in schools: A review of the outcome literature. Children and Youth Services Review , 31 (4), 464–470. https://doi.org/10.1016/j.childyouth.2008.10.002 Kolb, D.A. (1984) Experiential Learning: Experience as the Source of Learning and Developmen t. Prentice-Hall, Inc. Englewood Cliffs, NJ. Koller, J. R., & Bertel, J. M. (2006). Responding to today's mental health needs of children, familes and schools: revisiting the preservice training and preparation of school-based personnel. Education and treatment of children , 29 (2), 197-217. https://www.jstor.org/stable/42899882 Kononowicz, A. A., Woodham, L. A., Edelbring, S., Stathakarou, N., Davies, D., Saxena, N., ... & Zary, N. (2019). Virtual patients in health professions education: Systematic review and meta-analysis by the Digital Health Education Collaboration. Journal of Medical Internet Research, 21 (7), e13259. https://doi.org/10.2196/14676 Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77 (6), 1121–1134. https://doi.org/10.1037/0022-3514.77.6.1121 Li, F., Cui, Y., Li, Y., Guo, L., Ke, X., Liu, J., ... & Leckman, J. F. (2022). Prevalence of mental disorders in school children and adolescents in China: diagnostic data from detailed clinical assessments of 17,524 individuals. Journal of Child Psychology and Psychiatry , 63 (1), 34-46. https://doi.org/10.1111/jcpp.13445 Maurya, R. K. (2024). Using AI based chatbot ChatGPT for practicing counseling skills through role-play. Journal of Creativity in Mental Health , 19 (4), 513-528. https://doi.org/10.1080/15401383.2023.2297857 Maurya, R. K., & DeDiego, A. C. (2025). Artificial intelligence integration in counsellor education and supervision: A roadmap for future directions and research inquiries. Counselling and Psychotherapy Research , 25 (1), e12727. https://doi.org/10.1002/capr.12727 Mazzer, K. R., & Rickwood, D. J. (2015). Teachers' role breadth and perceived efficacy in supporting student mental health. Advances in school mental health promotion , 8 (1), 29-41. https://doi.org/10.1080/1754730x.2014.978119 Mollick, E., & Mollick, L. (2023). Assigning AI: Seven approaches for students, with prompts. arXiv preprint arXiv:2306.10052 . https://doi.org/10.48550/arXiv.2306.10052 Moore, D. A., & Healy, P. J. (2008). The trouble with overconfidence. Psychological Review , 115 (2), 502–517. https://doi.org/10.1037/0033-295X.115.2.502 Natividad-Sancho, L., Serradell, O., Sanmamed, A. F. F., & Garcia-Yeste, C. (2024). Review of the scientific literature on the impact of extending learning time with vulnerable groups. Children and Youth Services Review , 163 , 107809. https://doi.org/10.1016/j.childyouth.2024.107809 Nestel, D., & Tierney, T. (2007). Role-play for medical students learning about communication: Guidelines for maximising benefits. BMC Medical Education, 7 (1), 3. https://doi.org/10.1186/1472-6920-7-3 Nisar, S., Mahboob, U., Khan, R. A., & Rehman, D. (2021). Challenges of peer assisted learning in online clinical skills training of ophthalmology module. B MC medical education , 21 (1), 530. https://doi.org/10.1186/s12909-021-02959-3 Reinke, W. M., Stormont, M., Herman, K. C., Puri, R., & Goel, N. (2011). Supporting children's mental health in schools: Teacher perceptions of needs, roles, and barriers. School Psychology Quarterly, 26 (1), 1–13. https://doi.org/10.1037/a0022714 Rønning, S. B., & Bjørkly, S. (2019). The use of clinical role-play and reflection in learning therapeutic communication skills in mental health education: An integrative review. Advances in Medical Education and Practice , 10 , 415–425. https://doi.org/10.2147/AMEP.S202115 Schwarzer, R., & Jerusalem, M. (1995). Generalized self-efficacy scale. In J. Weinman, S. Wright, & M. Johnston (Eds.), Measures in health psychology: A user's portfolio. Causal and control beliefs (pp. 35–37). NFER-NELSON. Tanana, M. J., Soma, C. S., Srikumar, V., Atkins, D. C., & Imel, Z. E. (2019). Development and evaluation of ClientBot: Patient-like conversational agent to train basic counseling skills. Journal of medical Internet research , 21 (7), e12529. https://doi.org/10.2196/12529 Tzamaras, H., Sinz, E., Yang, Moore, J., & Miller, S. (2024). Competence over confidence: Uncovering lower self-efficacy for women residents during central venous catheterization training. BMC Medical Education , 24 , 923. https://doi.org/10.1186/s12909-024-05747-x Washburn, M., Bordnick, P., & Rizzo, A. S. (2016). A pilot feasibility study of virtual patient simulation to enhance social work students’ brief mental health assessment skills. Social Work in Health Care , 55 (9), 675–693. https://doi.org/10.1080/00981389.2016.1210715 Washburn, M., Parrish, D. E., & Bordnick, P. S. (2020). Virtual patient simulations for brief assessment of mental health disorders in integrated care settings. Social Work in Mental Health , 18 (2), 121-148. https://doi.org/10.1080/15332985.2017.1336743 World Health Organization. (2022). World mental health report: transforming mental health for all. World Health Organization. https://iris.who.int/handle/10665/356119 Wu Q., Wang Y., Lu L., Chen Y., Long H. & Wang J. (2022). Virtual Simulation in Undergraduate Medical Education: A Scoping Review of Recent Practice. Front. Med . 9:855403. https://doi.org/10.3389/fmed.2022.855403 Zhang, S., Zhao, X., Nan, D., & Kim, J. (2024). Beyond learning with cold machine: interpersonal communication skills as anthropomorphic cue of AI instructor. International Journal of Educational Technology in Higher Education , 21. https://doi.org/10.1186/s41239-024-00465-2. Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted 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-9368123","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":620288393,"identity":"21b48b53-3883-4774-b069-9d6f31d2125f","order_by":0,"name":"Huabing Liu","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0002-5704-6726","institution":"Shanghai Jiao Tong University School of Education","correspondingAuthor":true,"prefix":"","firstName":"Huabing","middleName":"","lastName":"Liu","suffix":""},{"id":620288394,"identity":"baaeed34-bca2-4cda-8a96-99700376f1d5","order_by":1,"name":"Nayila Tuerxun","email":"","orcid":"","institution":"Shanghai Jiao Tong University School of Education","correspondingAuthor":false,"prefix":"","firstName":"Nayila","middleName":"","lastName":"Tuerxun","suffix":""},{"id":620288395,"identity":"21a15edf-1a40-4ce1-9c10-176ad4e12cd9","order_by":2,"name":"Jing Chen","email":"","orcid":"","institution":"Shanghai Jiao Tong University School of Education","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2026-04-09 11:46:52","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9368123/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9368123/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106742810,"identity":"bddc6124-3df2-4bc7-bf15-e6a58935f50d","added_by":"auto","created_at":"2026-04-13 04:19:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":182873,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSystem design mapped onto Kolb’s experiential learning cycle\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9368123/v1/d645b93b3ffde1c6080857ae.png"},{"id":106742811,"identity":"cbea552b-0f93-4394-8f0a-639f77c75f54","added_by":"auto","created_at":"2026-04-13 04:19:00","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":295887,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eArchitecture of the Practice Agent\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9368123/v1/246605284edf4404ed30e17a.png"},{"id":106960046,"identity":"ad1e6587-8304-4a10-86fe-b26f8354cf1e","added_by":"auto","created_at":"2026-04-15 09:18:14","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":53036,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDistribution of time investment in the Practice Space and Demo Space\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9368123/v1/e5dc0f1d53843a1b1f4931db.png"},{"id":106742812,"identity":"add453c9-76e6-419a-ad9c-e209d5a40b9c","added_by":"auto","created_at":"2026-04-13 04:19:00","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":89172,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudent evaluations of the AI agent across eight perception items. The Agree (%) column indicates the percentage of students rating each item ≥ 4.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-9368123/v1/944439c51cd5b34923f9c3f9.png"},{"id":108490642,"identity":"45c360f0-9a24-4a37-a1b5-0450d69c6bd4","added_by":"auto","created_at":"2026-05-05 09:45:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":945589,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9368123/v1/74c5e4e4-356a-43f0-af74-d1dfa78292e5.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eAI-Assisted Solution-Focused Counseling Training for Novice Mental Health Educators: An Exploratory Study\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eMental health difficulties among school-age populations have received increasing attention from policymakers and educators worldwide. The World Health Organization (p. 41, 2022) estimates that one in seven adolescents meets criteria for a mental health disorder, placing considerable demands on school-based support systems. This crisis is particularly acute in China, where prevalence of any mental disorders among school children and adolescents was 17.5% (Li et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Teachers are frequently the first point of contact for students in distress, and are increasingly expected to provide initial mental health support before professional services become available (Reinke et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Preparing pre-service teachers to fulfil this role effectively has consequently become a priority in teacher education, yet current training practices leave substantial gaps between theoretical knowledge and applied competence.\u003c/p\u003e \u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e1.1 The Theory-Practice Gap in Mental Health Support Training\u003c/h2\u003e \u003cp\u003eTeacher preparation programs typically provide grounding in developmental psychology and school psychology, but offer limited structured opportunities to apply these concepts in realistic interactions (Koller \u0026amp; Bertel, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2006\u003c/span\u003e; Reinke et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Developing genuine mental health support (MHS) competence goes well beyond merely acquiring declarative facts. Instead, the development of school mental health support skills needs active exposure to diverse student cases, accompanied by immediate corrective feedback and sustained deliberate practice (Ericsson, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2004\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSeveral evidence-based frameworks have been adapted for educational contexts, including Solution-Focused Brief Therapy, Motivational Interviewing, and strengths-based approaches (Kim \u0026amp; Franklin, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Mastery of these techniques requires repeated application with diverse presentations and timely feedback, conditions that conventional teacher education struggles to provide (Mazzer \u0026amp; Rickwood, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Peer role-play, the most common practice modality, often fails to replicate the complexity of genuine teacher-student interactions (R\u0026oslash;nning \u0026amp; Bj\u0026oslash;rkly, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Gorski et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTwo further constraints compound these limitations. First, mental health support skills training in teacher education operates under severe time pressure: a typical Master of Education program in China allocates only 24\u0026ndash;64 hours to this area, compared with the hundreds of supervised practice hours required in professional counselling training (Council for Accreditation of Counseling and Related Educational Programs, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Second, having novice teachers practice emerging MHS skills with students experiencing genuine distress raises profound ethical concerns about potential harm in the absence of competence and supervision (Johnson, 2008). These constraints together create a need for practice environments that are scalable, ethically safe, and pedagogically effective.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.2 Simulation-Based Training as a Pedagogical Bridge\u003c/h2\u003e \u003cp\u003eTo bridge this theory-practice gap without compromising ethical standards, simulation-based training has established itself as an evidence-based pedagogy. In healthcare education, standardized patients have been utilized in the medical field since the 1960s (Flanagan \u0026amp; Cummings, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), and computerized virtual patients have demonstrated robust effectiveness across numerous trials (Kononowicz et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). This tradition has successfully extended to mental health contexts, where virtual patient simulations increase listening skills, self-efficacy and diagnostic accuracy in screening and assessment (Tanana et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Washburn et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Washburn et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Similarly, in teacher education, clinical simulations have been employed and proved to be beneficial for teachers\u0026rsquo; skill development (Dotger, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; De Coninck et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Chernikova et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). By providing a safe environment for trial-and-error learning, simulations allow trainees to develop foundational competence while shielding actual humans from the ethical risks associated with novice mistakes. Furthermore, this method effectively bypasses the inherent vulnerabilities of traditional peer role-play. While peer exercises are frequently compromised by time constraints and a lack of realism (Nisar et al., \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Nestel \u0026amp; Tierney, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2007\u003c/span\u003e), simulations offer an anytime and anywhere training environment.\u003c/p\u003e \u003cp\u003eHowever, these earlier simulation approaches nevertheless faced practical constraints that limited their applicability in teacher education. Standardized patient programs require trained actors, scheduling infrastructure, and supervisory resources that are difficult to sustain within teacher preparation curricula (Kaplonyi et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). which limited conversational range and ecological validity (Kononowicz et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Furthermore, barriers such as a lack of transferable learning resources prevent general educators from effectively implementing these complex tools (Wu et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Unlike these predecessors, generative AI enables the highly natural and flexible conversational dynamics essential to authentic counselling practice (Demszky et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Furthermore, these modern models effectively dismantle previous technical barriers. Teacher educators without programming expertise can now design and customize sophisticated simulated students using prompts in natural language (Mollick \u0026amp; Mollick, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIndeed, a growing body of work has begun to examine how such tools can be integrated into counselling skill training, with initial evidence suggesting that AI-assisted practice can enhance self-efficacy and counseling skills through simulated role-play training (Jeong et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Maurya, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). While early evidence-based studies demonstrate the feasibility of AI in clinical training, the broader research landscape remains in its infancy. The theoretical frameworks and specific pedagogy necessary to guide such practice are still lacking (Maurya \u0026amp; DeDiego, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Therefore, the current study investigates whether a theoretically grounded, dual-agent AI system can effectively support foundational skill acquisition and self-efficacy among novice educators.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e1.3 Solution-Focused Brief Therapy as an Entry Point\u003c/h2\u003e \u003cp\u003eOriginally developed by Steve de Shazer and Insoo Kim Berg, Solution-Focused Brief Therapy (SFBT) emphasizes clients\u0026rsquo; existing strengths through structured techniques including the miracle question, scaling questions, and exception-finding (De Shazer \u0026amp; Berg, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). SFBT\u0026rsquo;s concrete, stepwise techniques are highly accessible to beginners (De Shazer et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), and its non-pathologizing orientation is well suited to school settings (Kim \u0026amp; Franklin, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2009\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe selected SFBT as the initial focus skill because its accessibility matches our pre-service teacher sample lacking prior psychology backgrounds. Furthermore, its discrete techniques allow objective assessment, and its structured dialogue patterns are highly amenable to AI simulation with pedagogical scaffolding. While SFBT serves as an ideal foundational test case for this study, the underlying simulation architecture established here holds significant potential for future adaptation to other established therapeutic frameworks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e1.4 Theoretical Framework and Research Variables\u003c/h2\u003e \u003cp\u003eExisting studies on generative AI-based simulation counseling skills training have rarely been anchored in explicit learning theory, a gap that limits the interpretability of findings and the transferability of design principles (Maurya \u0026amp; DeDiego, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The present intervention was developed around Kolb\u0026rsquo;s (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e1984\u003c/span\u003e) experiential learning cycle, which comprises four phases: concrete experience, reflective observation, abstract conceptualization, and active experimentation. This framework is well suited to simulation-based skill training because its core logic, iterative cycles of experience and reflection, aligns with what generative AI simulation can offer: repeated practice opportunities with immediate, structured feedback that conventional instruction cannot readily sustain (Ericsson, 2008). The system architecture through which this cycle is operationalized is described in Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e2.4\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e1.4.1 Outcome Variables: Objective Knowledge vs. Subjective Self-Efficacy\u003c/h2\u003e \u003cp\u003eTwo outcomes were selected to evaluate whether the experiential learning cycle, as operationalized through the simulation system, produces meaningful skill development. SFBT knowledge captures the declarative and procedural competence that training aims to develop. Self-efficacy refers to a practitioner's belief in their capacity to perform specific support tasks (Bandura, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e1997\u003c/span\u003e). Humans with higher self-efficacy are more likely to practice with their learned knowledge (Artino, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Mazzer \u0026amp; Rickwood, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eInterestingly, these two outcomes are not assumed to move in parallel. When self-efficacy outpaces actual skill, a pattern documented in both medical education and professional training (Moore \u0026amp; Healy, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Dunning, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), practitioners may engage in overconfident and potentially harmful practice. Whether simulation-based and peer-based training yield different patterns of knowledge and self-efficacy calibration is therefore an empirical question the present study is positioned to examine.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e1.4.2 Mechanisms of Learning: Practice Duration vs. System Evaluation\u003c/h2\u003e \u003cp\u003eWhile establishing main effects is essential, identifying the specific mechanisms driving AI-mediated learning remains theoretically crucial for informing future design. Two candidate mechanisms are examined, each corresponding to a distinct prerequisite for completing the cycle. Practice duration reflects the degree of investment in the active experimentation phase. Without sufficient engagement at this stage, subsequent reflection and conceptualization are unlikely to be triggered regardless of the quality of available feedback. Although extended engagement is associated with achievement gains in educational settings more broadly (Natividad-Sancho et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), whether this holds within iterative simulation environments remains an open question. Drawing on the Technology Acceptance Model (Davis, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e1989\u003c/span\u003e), learners who reject the system\u0026rsquo;s instructional utility are unlikely to engage substantively with the reflective and conceptualization phases of the cycle, irrespective of time invested (Al-Fraihat et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Therefore, we explore candidate mechanisms.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e1.5 The Present Study\u003c/h2\u003e \u003cp\u003eResponding to school mental health needs and research gaps, this study designed a dual-agent AI system to support SFBT training within the practical constraints of pre-service teacher education and therefore evaluate its effectiveness and explore potential mechanisms. The study aims to answer the following three questions:\u003c/p\u003e \u003cp\u003eRQ1. Do students using the AI system show significantly greater improvements in SFBT knowledge and MHS self-efficacy compared to those in the peer role-play group?\u003c/p\u003e \u003cp\u003eRQ2. Within the AI-simulation group, are systems using duration and students\u0026rsquo; evaluations of the AI system associated with their knowledge and self-efficacy change?\u003c/p\u003e \u003cp\u003eRQ3. How do pre-service teachers perceive the simulation system as a tool for MHS skill development?\u003c/p\u003e \u003c/div\u003e"},{"header":"2 Method","content":"\u003cp\u003e2.1 Research Design\u003c/p\u003e\n\u003cp\u003eThis study employed a quasi-experimental, pretest-posttest design with non-equivalent control groups to evaluate the AI intervention\u0026rsquo;s effectiveness. Students were assigned to conditions based on existing class sections.\u003c/p\u003e\n\u003cp\u003e2.2 Participants\u003c/p\u003e\n\u003cp\u003eParticipants were graduate students enrolled in a mandatory 32-hour school psychology course at a Chinese university. All held undergraduate degrees in non-psychology majors and reported no prior counseling training. In the previous semester, students completed a course on student psychological development that covered developmental theories but not clinical methods or therapeutic techniques.\u003c/p\u003e\n\u003cp\u003eInitially, 35 students voluntarily registered for the study (experimental group: n = 20; control group: n = 15). However, five students were excluded from the final analysis due to incomplete data (i.e., missing either pre- or post-test assessments), resulting in a final sample of 30 participants (experimental: n = 18; control: n = 12). Despite this attrition, the groups remained comparable in academic performance, with final course grades differing by only 0.3 points on a 100-point scale (t = 0.421, p = 0.677). All procedures were approved by the university\u0026rsquo;s Institutional Review Board (# No. H20250564I), and informed consent was obtained from all participants.\u003c/p\u003e\n\u003cp\u003e2.3 Procedure\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe intervention occurred during Week 10, when the course covered SFBT. Both groups received the same 70-minute lecture on SFBT principles and applications in school settings. After the lecture, students participated in a 20-minute practice session. The experimental group began with a 15-minute instructor-led demonstration, during which the teacher modeled the use of the AI system by simulating a counseling interaction with the agent. Students observed this demonstration before engaging in approximately 5 minutes of independent practice. The control group practiced through peer role-play for the full 20 minutes. Instructors encouraged but did not require out-of-class practice for both groups.\u003c/p\u003e\n\u003cp\u003ePre-tests were administered one week before the intervention (Week 9), and post-tests were administered one week after (Week 11). Participants who completed both assessments received a small incentive (approximately 10 RMB). After post-test data collection, the control group received access to the AI system.\u003c/p\u003e\n\u003cp\u003e2.4 The AI-Assisted Practice System\u003c/p\u003e\n\u003cp\u003e2.4.1 System Architecture\u003c/p\u003e\n\u003cp\u003eWe developed a dual-agent AI system on the COZE platform comprising two functional spaces. Rather than following a fixed instructional sequence, the two spaces serve as flexible entry points into Kolb\u0026rsquo;s (1984; 1995) experiential learning cycle, which are concrete experience, reflective observation, abstract conceptualization, and active experimentation. Learners may enter either space at any point and move between them repeatedly, with each cycle of use informing the next.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eDemonstration Space.\u003c/em\u003e\u003c/strong\u003e In this space, the learner takes the role of a help-seeker while an AI counsellor demonstrates SFBT techniques in response to a problem the learner presents. This reverse role-play simultaneously provides a concrete experience of the mental health\u0026nbsp;interaction from the student perspective and supports reflective observation of expert technique application (Bandura, 1977). Learners may return to this space after encountering difficulty in practice, using fresh observational experience to revise their understanding before re-attempting.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003ePractice Space.\u0026nbsp;\u003c/em\u003e\u003c/strong\u003eThis space supports active experimentation. Two coordinated agents are deployed. A Student Simulation Agent, which portrays adolescents presenting varied psychological concerns (academic stress, peer conflict, family difficulties) with graduated complexity. An Assessment Agent, which analyzes conversation transcripts and delivers immediate feedback structured across three levels (Hattie \u0026amp; Timperley, 2007): task level (identifying correct and incorrect technique application), process level (explaining how specific moves align with SFBT principles), and self-regulation level (prompting learners to monitor and adjust their approach). Each practice attempt constitutes a new concrete experience. The Assessment Agent\u0026rsquo;s feedback transforms that experience into reflective observation and abstract conceptualization, enabling the learner to enter the next attempt with revised understanding. Repeated use of this space thus generates successive Kolb\u0026rsquo;s cycles within a single session.\u003c/p\u003e\n\u003cp\u003eBecause neither space is a prerequisite to the other, learners can move between them in response to their own learning needs. Learners can return to the Demonstration Space when practice reveals gaps in understanding, or move directly to practice after observation to test what they have seen.\u003c/p\u003e\n\u003ch3\u003e2.4.2 Agent Development\u003c/h3\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eKnowledge Base\u003c/em\u003e\u003c/strong\u003e. The agents\u0026rsquo; knowledge base included articles on SFBT in educational settings, and representative books focused on SFBT (e.g., De Shazer, 1985; Berg \u0026amp; Shilts, 2005; Kim, 2008). The first author, who is also a licensed counseling psychologist, interacted with the agent to ensure its professionalism in SFBT.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAgent Development\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e.\u0026nbsp;\u003c/em\u003e\u003c/strong\u003eThe agent prompts were developed through several iterations. Initially, the prompts were based on SFBT principles (Kim \u0026amp; Franklin, 2009) and feedback frameworks (Hattie \u0026amp; Timperley, 2007). An early single-agent design was later separated into distinct simulation and assessment functions to avoid role confusion. Next, three graduate students pilot-tested the system. \u0026nbsp;Based on their feedback, the prompts and models were further refined.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFinally, the agents were designed as follows.\u003c/p\u003e\n\u003cp\u003eThe Student Simulation Agent\u0026nbsp;was instructed to: (a) present concerns at varying difficulty levels; (b) respond differently to well-executed versus poorly-executed techniques; (c) avoid resolving issues too quickly.\u003c/p\u003e\n\u003cp\u003eThe Assessment Agent was instructed to: (a) identify specific SFBT techniques used (e.g., scaling questions, exception-finding, goal-setting); (b) provide feedback at task, process, and self-regulation levels; (c) limit responses to approximately 200 words.\u003c/p\u003e\n\u003ch2\u003e2.5 Measures\u003c/h2\u003e\n\u003ch3\u003e2.5.1 SFBT Knowledge Test\u003c/h3\u003e\n\u003cp\u003eGiven the lack of established SFBT knowledge measures, we developed a situational judgment test based on the suggestions in Hosany, Wellman \u0026amp; Lowe (2007) and Ferraz \u0026amp; Wellman (2009), comprising three scenarios measuring core SFBT and mental health support competencies: focusing on problems versus solutions, focusing on patients\u0026rsquo; current strengths and resources, and managing therapeutic silence. Two experienced school mental health practitioners reviewed items for content validity. Each scenario presented a student concern with four response options, one reflecting SFBT- and counseling-consistent approaches. Scores ranged from 0-3, with higher scores indicating better SFBT technique recognition.\u003c/p\u003e\n\u003ch3\u003e2.5.2 Self-Efficacy Scale\u003c/h3\u003e\n\u003cp\u003eWe adapted the General Self-Efficacy Scale (Schwarzer \u0026amp; Jerusalem, 1995) to assess participants\u0026rsquo; self-efficacy in student mental health support. The 10-item scale used 4-point Likert ratings (1 = completely incorrect, 4 = completely correct), with total scores ranging from 10-40. The internal consistency of the adapted scale is 0.86 (Cronbach\u0026rsquo;s \u0026alpha;) in our sample.\u003c/p\u003e\n\u003ch3\u003e2.5.3 AI System Evaluation Questionnaire\u003c/h3\u003e\n\u003cp\u003eFor the experimental group, we designed a questionnaire assessing their experience with the AI system. The questionnaire included:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eAI Evaluation Questionnaire\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e(8 items). The questionnaire included 8 items measuring perceived effectiveness of the AI agents. Example items include \u0026lsquo;helped me understand the core values of SFBT\u0026rsquo; and \u0026lsquo;increased my learning interest and engagement\u0026rsquo; (the complete list of items is presented in Figure 3 in the Results section). All items were rated on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTime Spent\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e(2 items). Participants recalled the approximate time they spent on the Demonstration Space and the Practice Space respectively.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eQualitative Feedback\u003c/em\u003e\u003c/strong\u003e (1 open-ended question). We asked participants to identify the helpful aspects of the AI system and suggest improvements. Responses were analyzed thematically to complement quantitative findings.\u003c/p\u003e\n\u003ch2\u003e2.6 Data Analysis\u003c/h2\u003e\n\u003cp\u003eAnalyses were conducted in Python 3.9.6 (pandas, numpy, scipy, statsmodels) with \u0026alpha; = .05. Shapiro\u0026ndash;Wilk tests indicated that self-efficacy gain scores met normality assumptions in both groups, whereas knowledge gain scores violated normality in both groups. Levene\u0026rsquo;s tests confirmed homogeneity of variance. Independent t-tests confirmed no significant baseline differences between groups.\u003c/p\u003e\n\u003cp\u003eFor self-efficacy, we used paired t-tests for within-group pre-to-post comparisons, independent t-tests for between-group comparisons, and ANCOVA with pre-test scores as covariates after confirming homogeneity of regression slopes. For SFBT knowledge, which violated normality assumptions, we used Wilcoxon signed-rank tests for within-group comparisons and Mann\u0026ndash;Whitney U tests for between-group comparisons, and ANCOVA was additionally conducted on post-test scores with pre-test scores as covariates. Given the small sample size, effect sizes were reported throughout (Cohen\u0026rsquo;s d for t-tests, partial \u0026eta;\u0026sup2; for ANCOVAs) to facilitate interpretation independent of statistical significance.\u003c/p\u003e"},{"header":"3 Results","content":"\u003cp\u003ePreliminary analyses and baseline equivalence are reported in Section \u003cspan refid=\"Sec22\" class=\"InternalRef\"\u003e3.1\u003c/span\u003e, followed by a description of agent usage patterns in the experimental group (Section \u003cspan refid=\"Sec23\" class=\"InternalRef\"\u003e3.2\u003c/span\u003e). Sections \u003cspan refid=\"Sec26\" class=\"InternalRef\"\u003e3.3\u003c/span\u003e and \u003cspan refid=\"Sec29\" class=\"InternalRef\"\u003e3.4\u003c/span\u003e report intervention effects on SFBT knowledge test scores and student mental health support self-efficacy, respectively. Each section includes between-group comparisons, ANCOVA, and within-group analyses examining agent usage as a potential predictor of gains (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Student written responses are cited where they directly bear on a finding.\u003c/p\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Preliminary Analyses\u003c/h2\u003e \u003cp\u003e \u003cb\u003eDistributional assumptions\u003c/b\u003e. Shapiro\u0026ndash;Wilk tests indicated that self-efficacy gain scores were normally distributed in both the experimental (W\u0026thinsp;=\u0026thinsp;0.949, p = .417) and control groups (W\u0026thinsp;=\u0026thinsp;0.966, p = .868). SFBT Test score gains violated normality in both groups (experimental: W\u0026thinsp;=\u0026thinsp;0.850, \u003cem\u003ep\u003c/em\u003e = .008; control: W\u0026thinsp;=\u0026thinsp;0.768, \u003cem\u003ep\u003c/em\u003e = .004); Wilcoxon signed-rank and Mann\u0026ndash;Whitney U tests were therefore used for test score outcomes, with parametric tests retained for self-efficacy.\u003c/p\u003e \u003cp\u003e \u003cb\u003eBaseline equivalence.\u003c/b\u003e Independent-samples t-tests confirmed that the experimental (n\u0026thinsp;=\u0026thinsp;18) and control (n\u0026thinsp;=\u0026thinsp;12) groups did not differ significantly on pre-test knowledge scores (t\u003csub\u003e(28)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.812, \u003cem\u003ep\u003c/em\u003e = .424, d\u0026thinsp;=\u0026thinsp;0.302) or self-efficacy (t\u003csub\u003e(28)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.555, \u003cem\u003ep\u003c/em\u003e = .584, d\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.207; Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Levene\u0026rsquo;s tests confirmed homogeneity of variance (all ps \u0026gt; .20). The groups were treated as equivalent at baseline.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Agent Usage Patterns in the Experimental Group\u003c/h2\u003e \u003cdiv id=\"Sec24\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1 Time Investment\u003c/h2\u003e \u003cp\u003eIn the Practice Space, 78% of students spent 15 minutes or more (Levels 3\u0026ndash;4; M\u0026thinsp;=\u0026thinsp;2.83, SD\u0026thinsp;=\u0026thinsp;0.71); only one student (5.6%) spent fewer than five minutes. In the Demo Space, time investment was shorter and more variable: 22.2% spent fewer than five minutes, and equal proportions (38.9% each) fell in the 5\u0026ndash;15 and 15\u0026ndash;30 minute bands (M\u0026thinsp;=\u0026thinsp;2.17, SD\u0026thinsp;=\u0026thinsp;0.79). The two duration variables were not significantly correlated (Spearman ρ\u0026thinsp;=\u0026thinsp;.338, p = .169).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003e\u003cem\u003e3.2.2 Student Perceptio\u003c/em\u003ens of the Agent\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003ePost-intervention evaluations of the agent were positive overall (8-item scale mean: M\u0026thinsp;=\u0026thinsp;3.74, SD\u0026thinsp;=\u0026thinsp;0.48). The highest-rated items were \u0026lsquo;facilitates SFBT skill acquisition\u0026rsquo; and \u0026lsquo;enhances course interest and engagement\u0026rsquo; (both M\u0026thinsp;=\u0026thinsp;4.00, SD\u0026thinsp;=\u0026thinsp;0.59; 83.3% agreement). Items addressing understanding of SFBT core values (M\u0026thinsp;=\u0026thinsp;3.83; 72.2% agreement) and strengthening SFBT identification (M\u0026thinsp;=\u0026thinsp;3.78; 72.2% agreement) also received relatively high ratings. Sixty-one percent of students agreed that agent practice increased their confidence in working with real students (M\u0026thinsp;=\u0026thinsp;3.67, SD\u0026thinsp;=\u0026thinsp;0.77), while 38.9% remained neutral or disagreed. Fifty-six percent expressed a desire for agent use in future courses (M\u0026thinsp;=\u0026thinsp;3.61, SD\u0026thinsp;=\u0026thinsp;0.92). One student affirmed the agent\u0026rsquo;s contribution to skill development directly. They commented \u0026ldquo;\u003cem\u003eUsing the Practice Space agent genuinely improved my ability to support students with psychological difficulties! \u0026rdquo;(translated from Chinese).\u003c/em\u003e\u003c/p\u003e \u003cp\u003e\u0026lsquo;More effective than in-class role-play\u0026rsquo; and \u0026lsquo;willing to use independently after class\u0026rsquo; were the two lowest-rated items (both M\u0026thinsp;=\u0026thinsp;3.50; 44.4% agreement), with half of students neutral on comparative effectiveness. Several students pointed out that AI practice are different from practice in reality.\u003c/p\u003e \u003cp\u003e \u003cem\u003eThe agent is sometimes wooden, [because] it repeats the same point, whereas real students\u0026rsquo; situations are [more] varied. At times, it gives the feeling of playing a game rather than genuinely solving a problem. (translated from Chinese)\u003c/em\u003e \u003c/p\u003e \u003cp\u003eAnother student noted the absence of in-built procedural guidance:\u003c/p\u003e \u003cp\u003e \u003cem\u003eFor those training for the first time, perhaps an optional example panel or step-by-step instruction bar could be introduced. Practising with a reference [guide] can help consolidate memory; once learners gain proficiency, they can choose to remove the guidance. (translated from Chinese)\u003c/em\u003e \u003c/p\u003e \u003cp\u003eTheir feedback indicates participants\u0026rsquo; feelings of using AI agents with their practice and highlights areas for future improvement.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Effects on SFBT Knowledge Test Scores\u003c/h2\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Within-Group Pre-to-Post Change\u003c/h2\u003e \u003cp\u003eBoth groups showed significant gains in SFBT knowledge from pre- to post-intervention (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The experimental group improved reliably (W\u0026thinsp;=\u0026thinsp;10, p = .003; Mdn gain\u0026thinsp;=\u0026thinsp;1.0, IQR [0.2, 2.0]), as did the control group (W\u0026thinsp;=\u0026thinsp;0, p = .031; Mdn gain\u0026thinsp;=\u0026thinsp;0.5, IQR [0.0, 1.0]).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSFBT Test Score Outcomes by Group\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMeasure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eExperimental\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e \u003cp\u003eBetween-Group Comparison\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMdn\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIQR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMdn\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIQR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eU\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eEffect Size (r)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePre-test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[1.0, 2.0]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[0.0, 2.0]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e124.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e.490\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e.148\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePost-test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[1.0, 2.0]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e[2.0, 3.0]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e147.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e.079\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e.366\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWithin-group test (Wilcoxon)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eW\u0026thinsp;=\u0026thinsp;0, \u003cem\u003ep\u003c/em\u003e = .031*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eW\u0026thinsp;=\u0026thinsp;10, \u003cem\u003ep\u003c/em\u003e = .003*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e \u003cp\u003e/\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"8\" nameend=\"c8\" namest=\"c1\"\u003e \u003cp\u003eANCOVA controlling for pre-test: F (1, 27)\u0026thinsp;=\u0026thinsp;3.302, p = .080, partial η\u0026sup2; = .109; Madj, ctrl\u0026thinsp;=\u0026thinsp;1.660, Madj, exp\u0026thinsp;=\u0026thinsp;2.171; model R\u0026sup2; = .347\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003e\u003cem\u003eNote\u003c/em\u003e. Test scores were non-normally distributed (Shapiro\u0026ndash;Wilk \u003cem\u003ep\u003c/em\u003e \u0026lt; .05); *\u003cem\u003ep\u003c/em\u003e \u0026lt; .05\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Between-Group Comparisons and ANCOVA\u003c/h2\u003e \u003cp\u003eThe Mann\u0026ndash;Whitney test found no significant difference in test score gains (U\u0026thinsp;=\u0026thinsp;137.5, p = .195, rank-biserial r\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.273), though post-test scores showed a marginal trend favouring the experimental group (U\u0026thinsp;=\u0026thinsp;147.5, p = .079, rank-biserial r\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.366). After controlling for pre-test scores, the group effect approached but did not reach conventional significance, F \u003csub\u003e(1, 27)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;3.302, \u003cem\u003ep\u003c/em\u003e = .080, partial η\u0026sup2; = .109, with the experimental group scoring higher after adjustment The pre-test covariate was significant (F \u003csub\u003e(1, 27)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;9.035, \u003cem\u003ep\u003c/em\u003e = .006), the homogeneity-of-slopes assumption was met (F\u0026thinsp;=\u0026thinsp;4.165, \u003cem\u003ep\u003c/em\u003e = .052), and the model explained 34.7% of post-test variance.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Effects on Self-Efficacy\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSelf-Efficacy Outcomes by Group\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMeasure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eExperimental\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e \u003cp\u003eBetween-Group Comparison\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eM\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eM\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003et\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eEffect Size (d)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePre-test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21.25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e.583\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u0026minus;.207\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePost-test\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e.731\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u0026minus;.130\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWithin-group test (paired t)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003et\u003csub\u003e(11)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;3.120, \u003cem\u003ep\u003c/em\u003e = .010*,\u003c/p\u003e \u003cp\u003ed\u0026thinsp;=\u0026thinsp;0.901\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003et\u003csub\u003e(17)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;3.625, \u003cem\u003ep\u003c/em\u003e = .002*, d\u0026thinsp;=\u0026thinsp;0.854\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"8\" nameend=\"c8\" namest=\"c1\"\u003e \u003cp\u003eANCOVA controlling for pre-test: F \u003csub\u003e(1, 27)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.014, p = .908, partial η\u0026sup2; = .001; Madj, ctrl\u0026thinsp;=\u0026thinsp;23.804, Madj, exp\u0026thinsp;=\u0026thinsp;23.964; model R\u0026sup2; = .588\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003e\u003cem\u003eNote\u003c/em\u003e. *\u003cem\u003ep\u003c/em\u003e \u0026lt; .05.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec30\" class=\"Section3\"\u003e \u003ch2\u003e3.4.1 Within-Group Pre-to-Post Change\u003c/h2\u003e \u003cp\u003eBoth groups showed significant gains in student mental health support self-efficacy (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). The experimental group gained a mean of 3.56 points (SD\u0026thinsp;=\u0026thinsp;4.16, 95% CI [1.49, 5.63]), t\u003csub\u003e(17)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;3.625, \u003cem\u003ep\u003c/em\u003e = .002, d\u0026thinsp;=\u0026thinsp;0.854. The control group gained a mean of 3.08 points (SD\u0026thinsp;=\u0026thinsp;3.42, 95% CI [0.91, 5.26]), t\u003csub\u003e(11)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;3.120, \u003cem\u003ep\u003c/em\u003e = .010, d\u0026thinsp;=\u0026thinsp;0.901.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec31\" class=\"Section3\"\u003e \u003ch2\u003e3.4.2 Between-Group Comparisons and ANCOVA\u003c/h2\u003e \u003cp\u003ePost-test self-efficacy (t\u003csub\u003e(28)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.35, \u003cem\u003ep\u003c/em\u003e = .731, d\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.130) did not differ significantly between groups. ANCOVA confirmed a non-significant group effect (F \u003csub\u003e(1, 27)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.014, \u003cem\u003ep\u003c/em\u003e = .908, partial η\u0026sup2; = .001), with adjusted means of 23.964 (experimental) and 23.804 (control). Pre-test self-efficacy accounted for most of the explained variance (F \u003csub\u003e(1, 27)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;38.302, \u003cem\u003ep\u003c/em\u003e \u0026lt; .001, R\u0026sup2; = .588), and the homogeneity-of-slopes assumption was satisfied (F\u0026thinsp;=\u0026thinsp;0.559, \u003cem\u003ep\u003c/em\u003e = .461).\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec32\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Correlations\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSpearman Rank-Order Correlations Among Agent Usage Variables and Outcome Gains\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1. AE\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2. PD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e.271\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3. DD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e.375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.338\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4. TG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e.236\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.129\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5. EG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e.234\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e.382\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.435\u0026dagger;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026minus;\u0026thinsp;.021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3.56\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e4.16\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003e\u003cem\u003eNote\u003c/em\u003e. AE\u0026thinsp;=\u0026thinsp;Agent Overall Evaluation; PD\u0026thinsp;=\u0026thinsp;Practice Space Duration; DD\u0026thinsp;=\u0026thinsp;Demo Space Duration; TG\u0026thinsp;=\u0026thinsp;Test Score Gain; EG\u0026thinsp;=\u0026thinsp;Self-Efficacy Gain. M and SD for each variable shown in bottom rows. \u0026dagger;p \u0026lt; .10\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eWithin the experimental group, no agent usage variable was significantly associated with test score gain: agent overall evaluation (ρ\u0026thinsp;=\u0026thinsp;.236, \u003cem\u003ep\u003c/em\u003e = .345), Practice Space duration (ρ\u0026thinsp;=\u0026thinsp;.129, \u003cem\u003ep\u003c/em\u003e = .611), and Demo Space duration (ρ\u0026thinsp;=\u0026thinsp;.021, \u003cem\u003ep\u003c/em\u003e = .934).\u003c/p\u003e \u003cp\u003eDemo Space duration showed a marginally significant positive correlation with self-efficacy gain (ρ\u0026thinsp;=\u0026thinsp;.435, \u003cem\u003ep\u003c/em\u003e = .071). Practice Space duration (ρ\u0026thinsp;=\u0026thinsp;.382, \u003cem\u003ep\u003c/em\u003e = .118) and agent overall evaluation (ρ\u0026thinsp;=\u0026thinsp;.234, \u003cem\u003ep\u003c/em\u003e = .349) were positively but non-significantly associated with gains.\u003c/p\u003e \u003cp\u003eThree evaluation items such as \u0026lsquo;the agent enhanced my interest and engagement in the course\u0026rsquo; have positive significant correlation relationships with self-efficacy gain (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e), and no evaluation items have significant correlation relationships with test score gain.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSpearman Correlations Between AI Agent Evaluation Items and Self-Efficacy Gain\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eItem\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStatement\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eρ\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe agent helped me understand the core values of SFBT.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.410\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.091\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePractising with the agent helped me master SFBT skills.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.505\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe agent strengthened my identification with SFBT principles.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.352\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.152\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAgent practice was more effective than in-class role-play.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.092\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.717\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eThe agent enhanced my interest and engagement in the course.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.504\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.033*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI am willing to use the agent independently after class.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.194\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.442\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMy confidence in real counselling improved after using the agent.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.553\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.017*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eI hope future courses will also be equipped with practice agents.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.013*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e\u003cem\u003eNote\u003c/em\u003e. Spearman rank-order correlations. Self-Efficacy Gain\u0026thinsp;=\u0026thinsp;post-test minus pre-test. *\u003cem\u003ep\u003c/em\u003e \u0026lt; .05.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4 Discussions","content":"\u003cdiv id=\"Sec34\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Discussion of Main Findings\u003c/h2\u003e \u003cp\u003ePre-post and between-group comparisons consistently indicated that AI-assisted psychology practice yielded comparable or superior outcomes relative to traditional classroom practice in developing students\u0026rsquo; mental health support skills and self-efficacy. This pattern aligns with our educational hypotheses articulated in the introduction and method sections, and corroborates studies demonstrating that AI-simulated training enhances learning outcomes in counseling education (Jeong et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Maurya, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The present study extends this literature to the domain of solution-focused counseling training, a domain previously dominated by didactic instruction and conventional role-play methods. The study further demonstrates the applicability of such AI agent designs to novice educators in mental health support contexts, an area that has received limited empirical attention. Notably, meaningful gains in self-efficacy were observed even when training was confined to a single counseling theory.\u003c/p\u003e \u003cp\u003eInterestingly, the two outcome variables exhibited divergent patterns of group differences. The experimental group demonstrated a marginal advantage on the knowledge test following covariate adjustment, whereas the group effect on self-efficacy was negligible. This asymmetric pattern, whereby knowledge gains outpaced self-efficacy gains in the AI-assisted condition, warrants closer interpretive attention.\u003c/p\u003e \u003cp\u003eWe offer two tentative, and potentially complementary, explanations regarding the knowledge-efficacy asymmetric pattern. The first draws on the Dunning-Kruger Effect, which describes the tendency for individuals to overestimate their competence in early stages of learning, followed by a decline in perceived confidence as awareness of domain complexity increases (Kruger \u0026amp; Dunning, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e1999\u003c/span\u003e). Evidence from medical and professional training similarly suggests that more competent trainees frequently report lower self-efficacy than less proficient peers, particularly during early-to-intermediate stages of skill development (Tzamaras et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). From this perspective, the experimental group\u0026rsquo;s more substantial knowledge acquisition may have simultaneously expanded their awareness of the complexity inherent in solution-focused theory, thereby tempering confidence gains that might otherwise have accompanied skill development. Conventional role-play, by contrast, affords social validation and peer affirmation that may sustain or inflate perceived competence independently of objective performance.\u003c/p\u003e \u003cp\u003eThe second explanation concerns the perceived authenticity of AI simulations. Student feedback indicated that some participants found the AI interlocutor mechanical rather than genuinely humanistic, which may have attenuated confidence gains irrespective of actual skill improvement. This interpretation is consistent with broader evidence that learners\u0026rsquo; trust in, and engagement with, technology-mediated interactions is sensitive to perceived anthropomorphism and relational authenticity (Ho et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). These two mechanisms are not mutually exclusive; they may operate simultaneously, apply differentially across students depending on prior technology experience or interpersonal orientation, or interact in ways that produce the aggregate pattern observed here. Systematic investigation of their relative contributions remains an important direction for future research.\u003c/p\u003e \u003cp\u003eWith respect to engagement metrics, neither practice time nor self-efficacy gains showed a significant association, nor did practice time predict knowledge gains. Observation time, however, showed a tentative positive association with self-efficacy but not with knowledge acquisition. This pattern is theoretically coherent within Bandura\u0026rsquo;s (1977) social cognitive framework. Observation of modeled performance serves as a vicarious mastery experience that may be particularly beneficial for novice learners in enhancing their perceived self-efficacy. Knowledge development, by contrast, tends to be slow and incremental (Falender \u0026amp; Shafranske, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2004\u003c/span\u003e). The non-significant role of practice time further suggests that other factors, such as engagement quality, learners\u0026rsquo; readiness, may warrant further investigation\u003c/p\u003e \u003cp\u003eFinally, as an exploratory finding, learners who evaluated the AI agent more favorably reported higher self-efficacy. This association has two potential interpretations. Students who perceived greater value in AI-mediated practice may have engaged more deeply with the simulated scenarios, thereby reinforcing confidence in their developing skills. Alternatively, those who explored the curriculum more thoroughly may have formed both more favorable appraisals of the agent and stronger self-efficacy beliefs through cumulative experience. The directionality and underlying mechanisms of this association remain to be clarified.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec35\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Limitations and Future Directions\u003c/h2\u003e \u003cp\u003eSeveral limitations should be considered when interpreting these findings. First, the sample was drawn from a single institution and comprised predominantly novice educators, which constrains the generalizability of results to other populations and contexts. Sample size was further limited by the natural classroom setting and the voluntary nature of participation, which together restricted statistical power and may have precluded the detection of subtler group differences.\u003c/p\u003e \u003cp\u003eSecond, the knowledge test contained a limited number of items with scarce reliability and validity. To our knowledge, no SFBT validated instrument currently exists, and the test used here represents a preliminary attempt at a SFBT test. Knowledge-related findings should therefore be treated as exploratory, and future research should prioritize the development of psychometrically sound assessment tools for this domain, alongside the incorporation of behavioral or observational measures of counseling competence.\u003c/p\u003e \u003cp\u003eThird, both outcomes were assessed within a single training cycle and a specific counseling theory, leaving open questions about the ecological validity. Future studies should employ longitudinal designs and agents to practice multiple counseling theories to determine whether effects are sustained or prevalent.\u003c/p\u003e \u003cp\u003eFourth, the internal mechanisms underlying the observed training effects remain unclear. None of the engagement variables examined were consistently associated with knowledge or self-efficacy gains, which may partly reflect the limited sample size. Future research incorporating a richer set of process variables, such as engagement quality and interaction patterns with the AI agent, alongside larger samples, would enable more systematic investigation of how AI-mediated practice translates into competence development.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec36\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Implications\u003c/h2\u003e \u003cp\u003eThe present findings have several implications for curriculum development and educational policy in teachers\u0026rsquo; mental health training. First, the demonstrated effectiveness of AI-assisted practice suggests that institutions and program developers could incorporate AI simulation as a structured component of mental health skills training curricula. Given that meaningful gains in skill and self-efficacy can be achieved through training focused on a single counseling approach, even modestly resourced programs may benefit from targeted AI-assisted modules without requiring wholesale curriculum redesign. This is particularly relevant in educational contexts where opportunities for supervised practice are limited and ethical constraints restrict direct client contact for novice learners.\u003c/p\u003e \u003cp\u003eAt the policy level, these results provide preliminary empirical support for investing in the development of AI-based training tools for novice mental health educators, a population that has historically received limited attention in both research and professional development policy. Future curriculum frameworks should additionally consider the intentional placement of demonstration-based activities, given their observed association with self-efficacy development, as a low-cost and scalable design feature applicable across diverse training contexts.\u003c/p\u003e \u003cp\u003eA third implication concerns the intentional sequencing of demonstration and practice within novice learner training. The present findings suggest that demonstration-based activities may warrant prioritization in the early stages of training. Practice-based components, while essential, may require a longer developmental window before their effects become fully apparent. Training programs and curriculum designers should therefore avoid drawing premature conclusions about the effectiveness of practice-based elements based solely on immediate post-training assessments. Instead, a staged curriculum design, one that leads with structured demonstration before transitioning to guided practice, may better align with the developmental needs of novice learners. This also highlights the importance of incorporating longitudinal assessment into program evaluation frameworks, so that the delayed but potentially substantial benefits of practice-based learning are not overlooked.\u003c/p\u003e \u003c/div\u003e"},{"header":"5 Conclusions","content":"\u003cp\u003eThis study examined the effectiveness of AI-assisted simulation in solution-focused counseling training for novice mental health educators, a population and domain that have received limited empirical attention. The findings demonstrate that AI-mediated practice yielded comparable or marginally superior outcomes relative to traditional role-play across both knowledge and self-efficacy measures, supporting the viability of AI simulation as a structured training modality in school mental health education. The divergent pattern observed between the two outcomes, whereby the AI-assisted group gained more on knowledge but not on self-efficacy, represents a theoretically meaningful finding that points to the complex interplay between competence development and self-perception in technology-mediated learning environments.\u003c/p\u003e \u003cp\u003eTo our knowledge, this study is among the first to evaluate AI-assisted training within a solution-focused counseling framework, and contributes a preliminary empirical foundation to an emerging area of research at the intersection of AI, school mental health education, and professional skill development. Beyond its methodological contributions, this study opens up a practical training scenario for novice educators operating in school-based mental health contexts, addressing a gap that has long existed in teacher education systems where opportunities for systematic psychological practice are often insufficient or absent. AI-assisted simulation thus holds promise not only as a research tool, but as an accessible and scalable means of strengthening the mental health support capacity of educators who occupy a critical yet underserved role in students\u0026rsquo; well-being. Nevertheless, the exploratory nature of this work, combined with limitations in sample size and measurement, necessitates cautious interpretation of the findings. Future research should build on these results by employing larger and more diverse samples, developing psychometrically validated assessment tools for this domain, and systematically investigating the design features and learning mechanisms that underpin effective AI-mediated school mental health training.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAl-Fraihat, D., Joy, M., Masa\u0026apos;deh, R., \u0026amp; Sinclair, J. (2020). Evaluating E-learning systems success: An empirical study. \u003cem\u003eComputers in Human Behavior, 102\u003c/em\u003e, 67\u0026ndash;86. https://doi.org/10.1016/j.chb.2019.08.004\u003c/li\u003e\n \u003cli\u003eArtino, A. R. (2012). Academic self-efficacy: From educational theory to instructional practice. \u003cem\u003ePerspectives on Medical Education\u003c/em\u003e, 1, 76\u0026ndash;85. https://doi.org/10.1007/s40037-012-0012-5\u003c/li\u003e\n \u003cli\u003eBandura, A. (1997). \u003cem\u003eSelf-efficacy: The exercise of control\u003c/em\u003e. W. H. Freeman.\u003c/li\u003e\n \u003cli\u003eBerg, I. K., \u0026amp; Shilts, L. (2005). \u003cem\u003eClassroom solutions:\u003c/em\u003e \u003cem\u003eWoww coaching\u003c/em\u003e. Brief Family Therapy Center.\u003c/li\u003e\n \u003cli\u003eChernikova, O., Heitzmann, N., Stadler, M., Holzberger, D., Seidel, T., \u0026amp; Fischer, F. (2020). Simulation-based learning in higher education: A meta-analysis. \u003cem\u003eReview of Educational Research, 90\u003c/em\u003e(4), 499\u0026ndash;541. https://doi.org/10.3102/0034654320933544\u003c/li\u003e\n \u003cli\u003eCouncil for Accreditation of Counseling and Related Educational Programs. (2024). \u003cem\u003e2024 CACREP Standards.\u0026nbsp;\u003c/em\u003ehttps://www.cacrep.org/for-programs/2024-cacrep-standards/\u003c/li\u003e\n \u003cli\u003eDavis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. \u003cem\u003eMIS Quarterly, 13\u003c/em\u003e(3), 319\u0026ndash;340. https://doi.org/10.2307/249008\u003c/li\u003e\n \u003cli\u003eDe Coninck, K., Keppens, K., Valcke, M., Dehaene, H., De Neve, J., \u0026amp; Vanderlinde, R. (2023). Exploring the effectiveness of clinical simulations to develop student teachers\u0026rsquo; parent-teacher communication competence.\u003cem\u003e\u0026nbsp;Research Papers in Education\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(1), 69\u0026ndash;101. https://doi.org/10.1080/02671522.2021.1961291\u003c/li\u003e\n \u003cli\u003eDe Shazer, S. (1985). \u003cem\u003eKeys to solutions in brief therapy\u003c/em\u003e. W. W. Norton.\u003c/li\u003e\n \u003cli\u003eDe Shazer, S., \u0026amp; Berg, I. K. (1997). \u0026lsquo;What works?\u0026rsquo; Remarks on research aspects of solution‐focused brief therapy. \u003cem\u003eJ\u003c/em\u003e\u003cem\u003eournal of Family therapy\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(2), 121-124. https://doi.org/10.1111/1467-6427.00043\u003c/li\u003e\n \u003cli\u003eDe Shazer, S., Dolan, Y., Korman, H., Trepper, T., McCollum, E., \u0026amp; Berg, I. K. (2021). \u003cem\u003eMore than miracles: The state of the art of solution-focused brief therapy\u003c/em\u003e. Routledge.\u003c/li\u003e\n \u003cli\u003eDemszky, D., Yang, D., Yeager, D. S., Bryan, C. J., Clapper, M., Chandhok, S., ... \u0026amp; Pennebaker, J. W. (2023). Using large language models in psychology. \u003cem\u003eNature Reviews Psychology\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(11), 688-701. https://doi.org/10.1038/s44159-023-00241-5\u003c/li\u003e\n \u003cli\u003eDotger, B. H. (2013). \u003cem\u003eClinical simulations for teacher development: A companion manual for teachers\u003c/em\u003e. Emerald Publishing Limited.\u003c/li\u003e\n \u003cli\u003eDunning, D. (2011). The Dunning\u0026ndash;Kruger effect: On being ignorant of one\u0026apos;s own ignorance. \u003cem\u003eIn Advances in experimental social psychology\u0026nbsp;\u003c/em\u003e(Vol. 44, pp. 247-296). Academic Press. https://doi.org/10.1016/B978-0-12-385522-0.00005-6\u003c/li\u003e\n \u003cli\u003eEricsson, K. A. (2004). Deliberate practice and the acquisition and maintenance of expert performance in medicine and related domains. \u003cem\u003eAcademic medicine\u003c/em\u003e, \u003cem\u003e79\u003c/em\u003e(Supplement_2), S70-S81. https://doi.org/10.1097/00001888-200410001-00022\u003c/li\u003e\n \u003cli\u003eFerraz, H., \u0026amp; Wellman, N. (2009). Fostering a culture of engagement: an evaluation of a 2‐day training in solution‐focused brief therapy for mental health workers. \u003cem\u003eJournal of Psychiatric and Mental Health Nursing\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(4), 326-334. https://doi.org/10.1111/j.1365-2850.2008.01374.x\u003c/li\u003e\n \u003cli\u003eFlanagan, O. L., \u0026amp; Cummings, K. M. (2023). Standardized Patients in Medical Education: A Review of the Literature. \u003cem\u003eCureus\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(7), e42027. https://doi.org/10.7759/cureus.42027\u003c/li\u003e\n \u003cli\u003eFalender, C. A., \u0026amp; Shafranske, E. P. (2004). \u003cem\u003eClinical supervision: A competency-based approach\u003c/em\u003e. American Psychological Association. https://doi.org/10.1037/10806-000\u003c/li\u003e\n \u003cli\u003eGorski, S., Prokop-Dorner, A., Pers, M., Stalmach-Przygoda, A., Malecki, Ł., Cebula, G., \u0026amp; Bombeke, K. (2022). The use of simulated patients is more effective than student role playing in fostering patient-centred attitudes during communication skills training: A mixed method study. \u003cem\u003eBioMed Research International\u003c/em\u003e, 2022, 1498692. https://doi.org/10.1155/2022/1498692\u003c/li\u003e\n \u003cli\u003eHattie, J., \u0026amp; Timperley, H. (2007). The power of feedback. \u003cem\u003eReview of Educational Research, 77\u003c/em\u003e(1), 81\u0026ndash;112. https://doi.org/10.3102/003465430298487\u003c/li\u003e\n \u003cli\u003eHo, A., Hancock, J., \u0026amp; Miner, A. S. (2018). Psychological, relational, and emotional effects of self-disclosure after conversations with a chatbot. \u003cem\u003eJournal of Communication\u003c/em\u003e, \u003cem\u003e68\u003c/em\u003e(4), 712-733. https://doi.org/10.1093/joc/jqy026\u003c/li\u003e\n \u003cli\u003eHosany, Z., Wellman, N., \u0026amp; Lowe, T. (2007). Fostering a culture of engagement: A pilot study of the outcomes of training mental health nurses working in two UK acute admission units in brief solution-focused therapy techniques. \u003cem\u003eJournal of Psychiatric and Mental Health Nursing\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e, 688\u0026ndash;695. https://doi.org/10.1111/j.1365-2850.2007.01161.x\u003c/li\u003e\n \u003cli\u003eJeong, D., Lee, H., Yang, H., Park, J. C., \u0026amp; Lee, S. M. (2025). AI integration in counseling training: Aiding counselors‐in‐training in self‐efficacy enhancement and anxiety reduction. \u003cem\u003eJournal of Counseling \u0026amp; Development, 103\u003c/em\u003e(3), 346\u0026ndash;360. https://doi.org/10.1002/jcad.12560\u003c/li\u003e\n \u003cli\u003eJohnson, W. B., Elman, N. S., Forrest, L., Robiner, W. N., Rodolfa, E., \u0026amp; Schaffer, J. B. (2008). Addressing professional competence problems in trainees: Some ethical considerations. \u003cem\u003eProfessional Psychology: Research and Practice\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(6), 589\u0026ndash;599. https://doi.org/10.1037/a0014264\u003c/li\u003e\n \u003cli\u003eKaplonyi, J., Bowles, K. A., Nestel, D., Kiegaldie, D., Maloney, S., Haines, T., \u0026amp; Williams, C. (2017). Understanding the impact of simulated patients on health care learners\u0026rsquo; communication skills: a systematic review. \u003cem\u003eMedical education\u003c/em\u003e, \u003cem\u003e51\u003c/em\u003e(12), 1209-1219. https://doi.org/10.1111/medu.13387\u003c/li\u003e\n \u003cli\u003eKim, J. S. (2008). Examining the effectiveness of solution-focused brief therapy: A meta-analysis. \u003cem\u003eResearch on Social Work Practice\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 107-116. https://doi.org/10.1177/1049731507307807\u003c/li\u003e\n \u003cli\u003eKim, J. S., \u0026amp; Franklin, C. (2009). Solution-focused brief therapy in schools: A review of the outcome literature. \u003cem\u003eChildren and Youth Services Review\u003c/em\u003e, \u003cem\u003e31\u003c/em\u003e(4), 464\u0026ndash;470. https://doi.org/10.1016/j.childyouth.2008.10.002\u003c/li\u003e\n \u003cli\u003eKolb, D.A. (1984) \u003cem\u003eExperiential Learning: Experience as the Source of Learning and Developmen\u003c/em\u003et. Prentice-Hall, Inc. Englewood Cliffs, NJ.\u003c/li\u003e\n \u003cli\u003eKoller, J. R., \u0026amp; Bertel, J. M. (2006). Responding to today\u0026apos;s mental health needs of children, familes and schools: revisiting the preservice training and preparation of school-based personnel. \u003cem\u003eEducation and treatment of children\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(2), 197-217. https://www.jstor.org/stable/42899882\u003c/li\u003e\n \u003cli\u003eKononowicz, A. A., Woodham, L. A., Edelbring, S., Stathakarou, N., Davies, D., Saxena, N., ... \u0026amp; Zary, N. (2019). Virtual patients in health professions education: Systematic review and meta-analysis by the Digital Health Education Collaboration. \u003cem\u003eJournal of Medical Internet Research, 21\u003c/em\u003e(7), e13259. https://doi.org/10.2196/14676\u003c/li\u003e\n \u003cli\u003eKruger, J., \u0026amp; Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one\u0026apos;s own incompetence lead to inflated self-assessments. \u003cem\u003eJournal of Personality and Social Psychology, 77\u003c/em\u003e(6), 1121\u0026ndash;1134. https://doi.org/10.1037/0022-3514.77.6.1121\u003c/li\u003e\n \u003cli\u003eLi, F., Cui, Y., Li, Y., Guo, L., Ke, X., Liu, J., ... \u0026amp; Leckman, J. F. (2022). Prevalence of mental disorders in school children and adolescents in China: diagnostic data from detailed clinical assessments of 17,524 individuals. \u003cem\u003eJournal of Child Psychology and Psychiatry\u003c/em\u003e, \u003cem\u003e63\u003c/em\u003e(1), 34-46. https://doi.org/10.1111/jcpp.13445\u003c/li\u003e\n \u003cli\u003eMaurya, R. K. (2024). Using AI based chatbot ChatGPT for practicing counseling skills through role-play. \u003cem\u003eJournal of Creativity in Mental Health\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(4), 513-528. https://doi.org/10.1080/15401383.2023.2297857\u003c/li\u003e\n \u003cli\u003eMaurya, R. K., \u0026amp; DeDiego, A. C. (2025). Artificial intelligence integration in counsellor education and supervision: A roadmap for future directions and research inquiries. \u003cem\u003eCounselling and Psychotherapy Research\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(1), e12727. https://doi.org/10.1002/capr.12727\u003c/li\u003e\n \u003cli\u003eMazzer, K. R., \u0026amp; Rickwood, D. J. (2015). Teachers\u0026apos; role breadth and perceived efficacy in supporting student mental health. \u003cem\u003eAdvances in school mental health promotion\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(1), 29-41. https://doi.org/10.1080/1754730x.2014.978119\u003c/li\u003e\n \u003cli\u003eMollick, E., \u0026amp; Mollick, L. (2023). Assigning AI: Seven approaches for students, with prompts. \u003cem\u003earXiv preprint arXiv:2306.10052\u003c/em\u003e. https://doi.org/10.48550/arXiv.2306.10052\u003c/li\u003e\n \u003cli\u003eMoore, D. A., \u0026amp; Healy, P. J. (2008). The trouble with overconfidence. \u003cem\u003ePsychological Review\u003c/em\u003e, \u003cem\u003e115\u003c/em\u003e(2), 502\u0026ndash;517. https://doi.org/10.1037/0033-295X.115.2.502\u003c/li\u003e\n \u003cli\u003eNatividad-Sancho, L., Serradell, O., Sanmamed, A. F. F., \u0026amp; Garcia-Yeste, C. (2024). Review of the scientific literature on the impact of extending learning time with vulnerable groups. \u003cem\u003eChildren and Youth Services Review\u003c/em\u003e, \u003cem\u003e163\u003c/em\u003e, 107809. https://doi.org/10.1016/j.childyouth.2024.107809\u003c/li\u003e\n \u003cli\u003eNestel, D., \u0026amp; Tierney, T. (2007). Role-play for medical students learning about communication: Guidelines for maximising benefits. \u003cem\u003eBMC Medical Education, 7\u003c/em\u003e(1), 3. https://doi.org/10.1186/1472-6920-7-3\u003c/li\u003e\n \u003cli\u003eNisar, S., Mahboob, U., Khan, R. A., \u0026amp; Rehman, D. (2021). Challenges of peer assisted learning in online clinical skills training of ophthalmology module. \u003cem\u003eB\u003c/em\u003e\u003cem\u003eMC medical education\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(1), 530. https://doi.org/10.1186/s12909-021-02959-3\u003c/li\u003e\n \u003cli\u003eReinke, W. M., Stormont, M., Herman, K. C., Puri, R., \u0026amp; Goel, N. (2011). Supporting children\u0026apos;s mental health in schools: Teacher perceptions of needs, roles, and barriers. \u003cem\u003eSchool Psychology Quarterly, 26\u003c/em\u003e(1), 1\u0026ndash;13. https://doi.org/10.1037/a0022714\u003c/li\u003e\n \u003cli\u003eR\u0026oslash;nning, S. B., \u0026amp; Bj\u0026oslash;rkly, S. (2019). The use of clinical role-play and reflection in learning therapeutic communication skills in mental health education: An integrative review. \u003cem\u003eAdvances in Medical Education and Practice\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, 415\u0026ndash;425. https://doi.org/10.2147/AMEP.S202115\u003c/li\u003e\n \u003cli\u003eSchwarzer, R., \u0026amp; Jerusalem, M. (1995). Generalized self-efficacy scale. In J. Weinman, S. Wright, \u0026amp; M. Johnston (Eds.), \u003cem\u003eMeasures in health psychology: A user\u0026apos;s portfolio. Causal and control beliefs\u003c/em\u003e (pp. 35\u0026ndash;37). NFER-NELSON.\u003c/li\u003e\n \u003cli\u003eTanana, M. J., Soma, C. S., Srikumar, V., Atkins, D. C., \u0026amp; Imel, Z. E. (2019). Development and evaluation of ClientBot: Patient-like conversational agent to train basic counseling skills. \u003cem\u003eJournal of medical Internet research\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(7), e12529. https://doi.org/10.2196/12529\u003c/li\u003e\n \u003cli\u003eTzamaras, H., Sinz, E., Yang, Moore, J., \u0026amp; Miller, S. (2024). Competence over confidence: Uncovering lower self-efficacy for women residents during central venous catheterization training. \u003cem\u003eBMC Medical Education\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e, 923. https://doi.org/10.1186/s12909-024-05747-x\u003c/li\u003e\n \u003cli\u003eWashburn, M., Bordnick, P., \u0026amp; Rizzo, A. S. (2016). A pilot feasibility study of virtual patient simulation to enhance social work students\u0026rsquo; brief mental health assessment skills. \u003cem\u003eSocial Work in Health Care\u003c/em\u003e, \u003cem\u003e55\u003c/em\u003e(9), 675\u0026ndash;693. https://doi.org/10.1080/00981389.2016.1210715\u003c/li\u003e\n \u003cli\u003eWashburn, M., Parrish, D. E., \u0026amp; Bordnick, P. S. (2020). Virtual patient simulations for brief assessment of mental health disorders in integrated care settings. \u003cem\u003eSocial Work in Mental Health\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 121-148. https://doi.org/10.1080/15332985.2017.1336743\u003c/li\u003e\n \u003cli\u003eWorld Health Organization. (2022).\u003cem\u003e\u0026nbsp;World mental health report: transforming mental health for all.\u0026nbsp;\u003c/em\u003eWorld Health Organization. https://iris.who.int/handle/10665/356119\u003c/li\u003e\n \u003cli\u003eWu Q., Wang Y., Lu L., Chen Y., Long H. \u0026amp; Wang J. (2022). Virtual Simulation in Undergraduate Medical Education: A Scoping Review of Recent Practice. \u003cem\u003eFront. Med\u003c/em\u003e. 9:855403. https://doi.org/10.3389/fmed.2022.855403\u003c/li\u003e\n \u003cli\u003eZhang, S., Zhao, X., Nan, D., \u0026amp; Kim, J. (2024). Beyond learning with cold machine: interpersonal communication skills as anthropomorphic cue of AI instructor. \u003cem\u003eInternational Journal of Educational Technology in Higher Education\u003c/em\u003e, 21. https://doi.org/10.1186/s41239-024-00465-2.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"AI simulation, Solution-Focused Brief Therapy, Pre-service teacher education, School mental health support training, Self-efficacy","lastPublishedDoi":"10.21203/rs.3.rs-9368123/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9368123/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTeachers increasingly serve as first responders to student mental health crises, yet pre-service teacher education often lacks opportunities for realistic, ethically safe practice of mental health support skills. This exploratory study evaluated a theoretically grounded dual-agent AI simulation system designed to train novice educators in Solution-Focused Brief Therapy (SFBT) skills. Using a quasi-experimental pretest-posttest design, 30 non-psychology graduate education students were divided into an AI-assisted practice group (n\u0026thinsp;=\u0026thinsp;18) and a peer role-play control group (n\u0026thinsp;=\u0026thinsp;12) according to class enrolment. The AI system was built around Kolb\u0026rsquo;s (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e1984\u003c/span\u003e) experiential learning cycle, operationalized through two interactive spaces: a Demonstration Space and a Practice Space. Both groups showed significant pre-to-post gains in SFBT knowledge and counselling self-efficacy. The AI-assisted group demonstrated a marginal advantage in objective knowledge acquisition, whereas self-efficacy gains were comparable across groups. Within the experimental group, certain AI system evaluations were positively associated with self-efficacy gains. Student feedback was largely positive, though some noted areas for improvement in the system\u0026rsquo;s authenticity and interactivity. These findings suggest that generative AI simulation offers a scalable and ethically safe complement to traditional peer role-play for foundational mental health support training. The observed asymmetry between knowledge and self-efficacy gains further highlights the nuanced effects of technology-mediated learning on trainee self-perception, with implications for the design of AI-assisted curricula in teacher education and mental health training.\u003c/p\u003e","manuscriptTitle":"AI-Assisted Solution-Focused Counseling Training for Novice Mental Health Educators: An Exploratory Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-13 04:18:54","doi":"10.21203/rs.3.rs-9368123/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ca941c52-d424-4d53-a22d-c65a03a6770b","owner":[],"postedDate":"April 13th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-13T04:18:54+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-13 04:18:54","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9368123","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9368123","identity":"rs-9368123","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}