Actionable companion · Feedback & scaffolding

Learning with agency: Breaking Barriers in Student Mental Health Care With AI-Enhanced Group Cognitive Behavioral Therapy: Pilot Feasibility Study

AI-assisted editorial implementation guide, separate from the original research. Proposed actions are not claims or recommendations attributed to the source authors. No independent peer review of this guide is asserted.

Read original paper Search paper agent CC BY 4.0

Evidence and context

Title and abstract screening; stage labels describe reported settings, not verified numerical ages. Design labels require full-text appraisal. Reported stage(s): Higher education. Verify the participant population and task before transferring the findings.

Original authors’ abstract, reproduced under the linked source license:

Abstract Background University students experience elevated psychological distress, with limited access to mental health services. While cognitive behavioral therapy (CBT) demonstrates efficacy for anxiety and depression, treatment gaps persist due to access barriers and insufficient between-session support. Large language model (LLM) chatbots could improve and scale CBT delivery. However, the scientific evaluation of chatbot-enhanced protocols is just emerging. Objective This pilot study aimed to assess the feasibility, acceptability, and preliminary efficacy of an LLM-based ChatBot as an adjunct to group Unified Protocol (UP) therapy for between-session support in university students with subclinical anxiety and depression symptoms. Methods A single-arm feasibility trial recruited university students aged 18 years and older with moderate subclinical symptoms (Social Phobia Inventory: 21‐40, Patient Health Questionnaire-9: 5‐14, or Generalized Anxiety Disorder-7: 5‐14), excluding those with current psychiatric disorders, suicidal ideation, or psychotropic medication use. The intervention comprised 4 weekly group UP counseling sessions complemented by an adjunctive Claude 3.7-Sonnet LLM ChatBot programmed with UP-based therapeutic prompts for between-session support rather than a stand-alone therapeutic agent. Primary feasibility outcomes included treatment adherence, chatbot engagement metrics, and system usability (System Usability Scale). Secondary outcomes assessed changes in generalized anxiety (Generalized Anxiety Disorder-7 Scale), social anxiety (Social Phobia Inventory), depression (Patient Health Questionnaire-9), and well-being (Short Warwick-Edinburgh Mental Wellbeing Scale) using paired t tests. Qualitative feedback was collected through focus group interviews and analyzed using thematic analysis. Results Of 72 screened participants, 37 met eligibility criteria and 19 initiated treatment (mean age 22.06, SD 1.78 years; 70.6% female). Retention was high with 17 completers (10.5% dropout rate). Among completers, 94.1% (16/17) attended ≥3 group sessions. The engagement with the CBT ChatBot was substantial: participants were active on a median of 23 days during the 34-day study period and exchanged a median of 15 messages in total. System usability was rated as excellent (mean 84.94, SD 10.98 out of 100). Pre-to-post comparisons revealed significant improvements in generalized anxiety (mean change −3.00, SD 3.46; t16=3.01, P=.004; Cohen d=0.71) and mental well-being (mean change +2.29, SD 3.65; t16=−2.17, P=.02; Cohen d=0.69). Social anxiety and depression showed nonsignificant trends toward improvement. Qualitative feedback highlighted the CBT ChatBot’s accessibility and nonjudgmental support while noting limitations in personalization. No adverse events or inappropriate chatbot interactions occurred. Conclusions Augmenting a group UP therapy with an LLM ChatBot demonstrated high feasibility, acceptability, and preliminary efficacy signals for university students with subclinical symptoms. The hybrid intervention package achieved strong retention and engagement while maintaining safety. These findings support progression to a randomized controlled trial to definitively evaluate this technology-enhanced approach for expanding access to evidence-based mental health interventions.

What kind of evidence is this?

Conversational & pedagogical agents · Randomized study reported

Stage: Higher education. Screening labels describe reported context and study-design language. They are not a risk-of-bias rating or proof of effectiveness for an age band.

Participant ages · source passage

Methods A single-arm feasibility trial recruited university students aged 18 years and older with moderate subclinical symptoms (Social Phobia Inventory: 21‐40, Patient Health Questionnaire-9: 5‐14, or Generalized Anxiety Disorder-7: 5‐14), excluding those with current psychiatric disorders, suicidal ideation, or psychotropic medication use.

Study methods · source passage

This study was reviewed and approved by the Scientific Council of University Research and Creation of the West University of Timișoara (approval no. 22170, dated March 31, 2025). All procedures were conducted in accordance with the ethical standards of the relevant institutional and national research committees and with the 1964 Helsinki Declaration and its later amendments. Prior to participation, all participants provided written informed consent. Participant data were anonymized and stored securely in accordance with applicable data protection regulations.

Limitations and uncertainty · source passage

The present findings add to a growing consensus that generative AI chatbots can serve as a valuable adjunct in mental health care. Rather than viewing these systems as stand-alone therapists, it may be most productive to deploy them as part of a blended care model—where the chatbot extends the reach of a human clinician. In this hybrid model, the CBT ChatBot was never intended to serve as a stand-alone therapeutic agent. Instead, its role was strictly supportive, functioning as a digital bridge to ensure that the core UP principles taught by the facilitators were practiced in the participant’s daily environment. In our university context, the CBT ChatBot effectively became a 24/7 “coach” that reinforced the UP techniques students learned during the group counseling sessions. This highlights a key advantage of LLM chatbots: they are instantly accessible and infinitely scalable. A single AI can support hundreds of users simultaneously, even at inconvenient times, providing guidance or a compassionate ear when human providers are unavailable. This could help bridge gaps in care, especially for young adults who often struggle to access traditional therapy due to cost, scheduling, or stigma. Indeed, a recent report noted that affordability-related barriers to health care access are highly prevalent, with individuals experiencing greater psychological distress reporting disproportionately higher rates of such barriers [56]. Chatbots might serve as a lower-barrier first line of support—for example, a student experiencing heightened stress or loneliness could turn to the chatbot for coping strategies and emotional support during nights or weekends, potentially preventing escalation of distress. Additionally, LLM chatbots may function as triage or stepped-care tools in public health. They could help monitor symptom changes and alert human clinicians when higher-level intervention is needed. For example, Kuhail et al [20] found that therapists were largely unable to distinguish between transcripts of sessions involving a relational AI chatbot and those involving human therapists and rated the AI-generated interactions as slightly higher in quality, suggesting potential for chatbots to provide preliminary support in contexts of limited mental health resources. Our findings also hint at specific therapeutic roles: the chatbot particularly aided in emotion expression and homework completion—tasks that augment therapy by encouraging daily practice.

Design the system before the pilot

Define whether the agent asks questions, gives hints, explains or assesses. Test hallucinated feedback, excessive praise, misleading confidence and responses when the learner is confused. A conversational interface alone does not establish autonomous agency.

Give feedback on the learner’s reasoning, then fade hints. Record whether the learner can complete the next task without that scaffold.
Include a delayed assessment and an unfamiliar transfer problem without AI. Immediate supported performance cannot establish durable learning.
Measure engagement separately from achievement. Check whether novelty-driven participation persists and whether it translates to independent performance.
Use common rubrics, blind scoring where feasible and a fresh unaided reasoning task. Document permissible assistance and avoid treating AI-detector scores as proof of misconduct.
Co-design accommodations with learners and specialists. Check access and error burden across groups without inferring subgroup effectiveness from small or unreported samples.

A 90-day district pilot

Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
Days 16–30 · Design. Choose one disciplinary task. Require a first attempt, a documented AI consultation, a critique of the output and an independent explanation. Align the comparison group and scoring rubric with the same objective. Name an accountable owner and set the baseline, success threshold and budget before implementation.
Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
Days 61–90 · Decide. Blind-score reasoning and accuracy on a new task without AI. Distinguish model examination scores, student perceptions and actual learner gains; only the last directly tests educational benefit. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Bogdan Tudor Tulbure; Bianca Spătaru; Andrei Rusu; Darian Faur; Daniel Precupaș; Loredana Marcela Trancă; Șerban Gabriel Popa; Adrian Spătaru. Breaking Barriers in Student Mental Health Care With AI-Enhanced Group Cognitive Behavioral Therapy: Pilot Feasibility Study. JMIR Formative Research (2026-06-12). https://doi.org/10.2196/84296

Companion prepared 2026-10-30. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.