The transition to college represents a critical developmental period marked by academic, social, and financial stressors, which can significantly impact mental health. According to the American College Health Association (1), approximately 80% of college students report feeling overwhelmed by their responsibilities, while 75% lack access to adequate mental health services (2). These challenges were exacerbated by the COVID-19 pandemic, which triggered unprecedented disruptions to campus life. A large-scale survey of over 45,000 undergraduate and graduate students revealed that 35% met the criteria for major depressive disorder and 39% for generalized anxiety disorder post-pandemic (3). Contributing factors included social isolation from remote learning, health-related fears, financial instability, and abrupt lifestyle changes (4). Despite this growing need, systemic barriers to traditional mental health care, such as counseling, exist. According to Ebert et al. (5), stigma is a critical barrier among college students accessing mental health care and many would rather seek self-help. Other barriers include high treatment cost, limited availability of providers, and long waiting periods (6, 7). These barriers leave many college students seeking mental health care without timely support.

To address these gaps, scalable and accessible interventions are urgently needed. Artificial intelligence (AI)-driven chatbots have emerged as a promising solution, offering 24/7 availability, anonymity to reduce stigma, and low-cost delivery of evidence-based strategies such as cognitive-behavioral techniques (CBT) and mindfulness (8–10). Preliminary studies suggest chatbots may improve emotional well-being by providing psychoeducation, mood tracking, and coping skill development (2, 11). The use of psychological theories like CBT, which targets maladaptive thought patterns, chatbots provide structured interventions, including mood tracking, psychoeducation, and coping skill development (12). Their conversational nature fosters therapeutic alliance, mimicking aspects of human interaction while remaining scalable (13, 14). Unlike traditional telehealth, chatbots can deliver consistent, tailored support without requiring extensive infrastructure, making them particularly suitable for college settings (15). However, there is a lack of reviews focused solely on college students, and the evidence remains fragmented, with variability in chatbot design (e.g., rule-based vs. AI-driven), target outcomes (e.g., anxiety reduction vs. general well-being), and methodological rigor. Additionally, limited research explores barriers to engagement, such as privacy concerns or user preferences for human interaction (6).

This systematic rapid review aims to synthesize existing evidence on the effectiveness of chatbots in improving mental health and well-being among college students. “Mental health” is operationalized using DSM-5 diagnostic criteria, validated measures of negative affect (e.g., PHQ-9 for depression), and subjective well-being scales. Secondary objectives include identifying (1) barriers to chatbot adoption, (2) design features linked to efficacy, and (3) gaps in current research. The study addresses the question: What evidence exists regarding the effectiveness of chatbots in improving mental health outcomes and well-being in college students, and what factors influence their implementation?

The decision to conduct a rapid review was driven by the need to synthesize emerging evidence on this topic in a timely manner to inform ongoing interventions. Given the evolving nature of the subject and its relevance to current priorities, as well as lack of randomized controlled trials, we aimed to provide a concise, evidence-informed synthesis that could be accessible within a shorter time frame. We would also like to note the small number of included studies reflects the current state of the literature rather than a limitation of our search strategy or review process. These were the only studies that met our pre-defined inclusion criteria based on study quality, relevance, and methodological rigor. A full systematic review would have yielded the same set of studies, but would have required considerably more time and resources without changing the conclusions.

By evaluating chatbots’ potential to bridge mental health care gaps, this review informs universities, developers, and policymakers seeking cost-effective solutions. Findings will highlight best practices for integrating chatbots into campus wellness programs while addressing limitations (e.g., ethical concerns, cultural responsiveness) to ensure equitable access.

This systematic rapid review followed established guidelines for accelerated evidence synthesis (16), retaining core systematic review principles (17) to minimize bias while streamlining processes to meet time constraints. Key adaptations included focused search strategies, predefined PICO criteria, and single-reviewer title/abstract screening with dual verification. A health science librarian at the University of Florida collaborated on search term development and database selection to optimize precision and recall.

The systematic search yielded 442 articles across four databases: ACM Digital Library (208), Applied Science & Technology Source (208), PsycINFO (13), and PubMed (13). After removing 14 duplicates, 428 records underwent title/abstract screening, excluding 406 irrelevant studies. Full-text review of 21 articles yielded 9 eligible studies for final analysis ( Figure 1 : PRISMA flowchart).

The nine included studies, encompassing 1,082 participants (sample sizes: 42–250) and eight distinct chatbots. Interventions ranged from single-use sessions (12 minutes) to 6-week programs, with most focusing on anxiety, depression, or well-being.

Eight studies (89%) reported statistically significant improvements in at least one mental health outcome ( Table 2 ): (2, 6, 8, 9, 11, 22–24).

One study (Mind Tutor; 21) showed no significant changes in well-being (SWEMWBS: Δ = +0.04, p = 0.62), potentially due to brief intervention duration (6 weeks) or lack of personalized feedback.

As for the Level of Evidence, there were five Level I RCTs and four Level II quasi-experimental studies were included. There were two studies with good quality (PEDro scores 6–8): 2, 8. Two studies were rated fair (PEDro scores 4-5): 11, 24. The remaining studies were rated poor (PEDro scores <5), limited by small samples or lack of control groups. Higher-quality studies (PEDro ≥6) consistently supported chatbot efficacy, whereas lower-scoring studies (e.g., 6; PEDro = 1) showed smaller effect sizes.

For the Chatbot design and intervention delivery ( Table 3 ), we noticed that effective interventions shared those key features: 1) CBT integration: Woebot and XiaoNan used structured CBT modules (e.g., mood tracking, cognitive restructuring). 2) Personalization: ARU incorporated cultural adaptation for Indian students, improving adherence. 3) Optimal delivery frequency: Daily interactions (e.g., Tess; 2) correlated with greater engagement (usage rate: 78% vs. 52% in biweekly groups). In contrast, passive apps (Mind Tutor) with static content showed minimal impact, and brief interventions (<2 weeks; e.g., Gloomy; 9) had transient effects.

A total of 14 different outcome measures were used across the nine studies as shown in Table 4 , with three predominating ( Figure 2 ): GAD-7 (5 studies): Detected anxiety reductions (e.g., 6: Δ = −1.20, p = 0.04). PHQ-9 (4 studies): Tracked depression improvements (e.g., 11: chatbot Δ = −5.25 vs. control Δ = −2.98, p < 0.001). PANAS (4 studies): Captured mood shifts (e.g., 8: negative affect Δ = −1.26 vs. control Δ = +1.21).

The most used instruments were the Generalized Anxiety Disorder Scale-7 (GAD-7), Patient Health Questionnaire-9 (PHQ-9), and Positive and Negative Affect Schedule (PANAS) as shown in Figure 2 .

Five studies used the GAD-7 (2, 6, 8, 11, 23). This is a seven-item self-report scale to assess anxiety symptoms over the past two weeks, it uses a four-point Likert scale, ranging from 0 (not at all) to 3 (nearly every day). Four studies used the PHQ-9 (2, 8, 11, 23). This is a nine-item self-report questionnaire that assesses the frequency and severity of depressive symptoms within the previous two weeks. Four studies used the PANAS (2, 8, 11, 21). This is a 20-item self-report measure of current positive and negative effects. All other instruments reported in Figure 2 were only used by one study.

Chatbots represent a promising, scalable solution to address the mental health crisis among college students, particularly when integrating evidence-based therapies like CBT and prioritizing frequent, personalized engagement. While this review underscores their potential to reduce anxiety, depression, and academic stress, critical gaps remain: 1) Consensus on core outcome measures (e.g., PHQ-9, GAD-7) and intervention duration (≥2 weeks) is needed. 2) Future chatbots must incorporate emergency responses to students with severe symptoms and referrals to human providers. Our findings support piloting chatbots as adjuncts, not replacements to traditional counseling, particularly during peak stress periods (e.g.,exams).