AI-Driven Mental Health Chatbots and Their Impact on Anxiety and Depression Among College Students

1.1. Anxiety and Depression Among College Students

Anxiety and depression are among the most prevalent mental health conditions affecting college students and represent a growing concern within higher education (Luo et al., 2025; Minaya et al., 2025). Anxiety is associated with persistent worry and heightened physiological activation that interferes with cognitive functioning, whereas depression is defined by enduring low mood, diminished interest or pleasure, and functional impairment (Puente-Torre et al., 2025). These conditions frequently emerge during young adulthood, a developmental period that overlaps with the college years, making university students particularly vulnerable to psychological distress (Nemesure et al., 2021). Anxiety and depressive disorders represent the most common mental health conditions among U.S. college students, with prevalence rates increasing substantially over the past decade. These trends highlight anxiety and depression as primary drivers of psychological risk and growing mental health concerns across U.S. college campuses (Reyes-Portillo et al., 2025). In addition, anxiety and depressive symptoms among college students are associated with poorer academic performance, including difficulties with concentration, reduced academic engagement, and impaired learning outcomes (Luo et al., 2025; Minaya et al., 2025).

AI-based mental health chatbots are automated conversational systems designed to support psychological well-being through interactive dialogue, with demonstrated effectiveness in reducing depression and anxiety among college students (Reyes-Portillo et al., 2025; Xu & Ma, 2025). Empirical evidence further indicates that these interventions enhance adherence, user satisfaction, and self-efficacy, highlighting their practical value in higher education settings.Across the broader literature, anxiety and depression consistently emerge as the most prevalent mental health conditions among university students and the primary targets of AI-based interventions (Nemesure et al., 2021; Schaab et al., 2024; Zhai et al., 2025). Machine learning approaches reinforce this focus by enabling early detection and accurate prediction of these conditions, underscoring their central role in AI-driven mental health screening and intervention systems (Juárez-Santiago et al., 2025; Siddiqua et al., 2023; Zhai et al., 2025).

1.2. Limitations of Traditional Mental Health Support in Higher Education

Despite the high prevalence of anxiety and depressive disorders among university students, access to traditional mental health services remains limited due to stigma, privacy concerns, and resource constraints, leading to underutilization of available support (Wang et al., 2025; Xu & Ma, 2025; Zhai et al., 2025). Untreated mental health conditions continue to negatively affect academic performance, emotional functioning, and overall quality of life among students (Wang et al., 2025; Yang & Rui, 2025).

Conventional mental health interventions are often reactive and rely on self-initiated help-seeking, which delays detection and limits opportunities for early intervention (Zhai et al., 2025; X. Zhang & Shi, 2025). Machine learning studies further indicate that anxiety and depression frequently remain undetected until symptoms escalate, highlighting the limitations of traditional monitoring approaches in higher education contexts (Schaab et al., 2024; Zhai et al., 2025).

Consequently, growing evidence supports the use of digital and AI-based mental health interventions as scalable and accessible alternatives that can reduce stigma and improve early detection and support for university students (Wang et al., 2025; Xu & Ma, 2025; X. Zhang & Shi, 2025).

1.3. Emergence of AI-Driven Mental Health Chatbots in Higher Education

Recent studies emphasize that AI-based chatbots can deliver standardized, evidence-based interventions at scale while reducing reliance on clinician availability, addressing a key constraint in campus mental health systems (Wang et al., 2025). AI-powered mental health chatbots use conversational AI to engage students in human-like dialogue while delivering psychoeducation, emotional validation, symptom monitoring, and cognitive-behavioral therapy–based strategies to support students with anxiety and depression (Reyes-Portillo et al., 2025).

AI-powered mental health chatbots typically employ conversational AI, incorporating natural language processing and machine learning techniques to simulate interactive dialogue with users (Reyes-Portillo et al., 2025). AI-powered mental health chatbots are conversational systems designed to deliver psychoeducation, emotional support, and structured cognitive–behavioral therapy–based strategies through interactive dialogue (Klos et al., 2021). Studies involving university students show that these chatbots adapt responses to users’ emotional states and symptom reports, providing individualized support for anxiety and depressive symptoms (Klos et al., 2021; Reyes-Portillo et al., 2025). As such, chatbot-based interventions represent a structured yet flexible digital modality for supporting student mental health outside traditional clinical settings.

Emerging intervention studies indicate that chatbot-based systems support ongoing, asynchronous engagement, allowing students to access mental health support through repeated interactions delivered outside traditional counseling settings (Klos et al., 2021; Wang et al., 2025).This capability is particularly salient in higher education settings, where delayed help-seeking behaviors and limited counseling capacity frequently restrict timely access to mental health care (Wang et al., 2025). Machine-learning–based screening research indicates that anxiety and depression are often identified after college students seek clinical support, which may limit opportunities for early detection and prevention (Zhai et al., 2025).

1.4. Research Gap and Purpose of the Study

Although scholarship on AI-driven mental health chatbots in higher education has grown, current evidence remains insufficient to draw firm conclusions about their effectiveness in reducing anxiety and depressive symptoms among college students, with existing studies highlighting methodological limitations and the need for more rigorous outcome-focused research (Albikawi et al., 2025; Reyes-Portillo et al., 2025). Existing studies demonstrate substantial variability in the performance outcomes of machine learning–based mental health tools, largely attributable to differences in data sources, model types, validation strategies, and study quality. Most studies rely on internal validation and exhibit very low certainty of evidence (Schaab et al., 2024). Many investigations rely on short-term or pilot study designs, which constrain conclusions about the sustained effects of chatbot-based interventions on anxiety and depressive symptoms among college students (Reyes-Portillo et al., 2025).

Schaab et al. (2024) explain that much of the current evidence is derived from observational and correlational studies that rely on retrospective data and internal validation rather than randomized or longitudinal designs. As a result, these studies can identify associations between AI-based mental health tools and outcomes but cannot establish whether observed changes in anxiety or depression are directly caused by the interventions, leading to low certainty of evidence regarding causal effectiveness (Schaab et al., 2024). Prior research often examines AI-based mental health technologies within broader AI-enhanced environments, combining chatbots with tools such as predictive models and intelligent systems, which limits the ability to isolate chatbot-specific effects (Yang & Rui, 2025; X. Zhang & Shi, 2025). Many studies integrate multiple AI approaches for screening, intervention, and support rather than evaluating chatbots independently (Zhai et al., 2025; X. Zhang & Shi, 2025). Consequently, this aggregated approach reduces clarity regarding the unique contribution of chatbot-based interventions to anxiety and depression outcomes in higher education (Yang & Rui, 2025; Yeh et al., 2025).

2. Methods

This study uses a literature review as its methodological approach. A literature review integrates and interprets existing research to consolidate fragmented evidence and advance conceptual understanding beyond individual studies (Snyder, 2019). This approach is appropriate because research on AI-driven mental health chatbots addressing anxiety and depression spans diverse study designs, intervention models, and institutional contexts, limiting the suitability of quantitative synthesis. A literature-based method, therefore, enables integrative examination of therapeutic impact, application characteristics, and implementation patterns relevant to higher education mental health systems (Snyder, 2019).Literature searches were conducted across multiple academic databases, including PubMed, PsycINFO, ScienceDirect, ERIC, and with Google Scholar used as a supplementary source to identify additional relevant studies. The literature search employed a structured set of keywords organized into core concept domains relevant to the study focus.

2.1. Search Strategy

Terms related to artificial intelligence included “artificial intelligence,” “AI,” “machine learning,” and “deep learning.” To capture chatbot-based technologies, keywords such as “chatbot,” “conversational agent,” “conversational AI,” and “virtual assistant” were used. Mental health outcomes were represented using terms including “anxiety,” “depression,” “depressive symptoms,” and “anxiety disorders.” Population-specific terms used in higher education contexts included “college students,” “university students,” “higher education,” and “postsecondary students.” Within each concept domain, keywords were combined using the Boolean OR to capture terminological variation, while AND was applied across domains to retrieve studies addressing artificial intelligence–based conversational systems for anxiety and depression among college students.

2.2. Inclusion and Exclusion Criteria

To ensure relevance, rigor, and alignment with the study objectives, explicit inclusion and exclusion criteria were applied during the literature selection process.

Inclusion Criteria

Studies were included if they (a) examined artificial intelligence–driven mental health chatbots or conversational agents designed to support, monitor, screen, or intervene in anxiety and/or depression; (b) focused on college or university student populations within higher education settings; (c) reported empirical findings, including randomized controlled trials, quasi-experimental studies, longitudinal studies, pilot interventions, or observational analyses; (d) were published in peer-reviewed academic journals; and (e) were available in English. Both intervention-based chatbot studies and machine-learning–based screening or monitoring studies were included when anxiety or depression constituted primary outcomes.

Exclusion Criteria

Studies were excluded if they (a) focused exclusively on non-chatbot AI tools (e.g., intelligent tutoring systems, learning analytics platforms, or purely administrative AI applications) without a conversational mental health component; (b) examined mental health conditions other than anxiety or depression without reporting disaggregated findings for these outcomes; (c) targeted populations outside higher education, such as primary or secondary school students, clinical psychiatric populations without a student focus, or the general public; (d) consisted solely of conceptual papers, opinion pieces, editorials, protocols, or non-peer-reviewed sources; or (e) lacked sufficient methodological detail or outcome reporting to support interpretation of therapeutic impact or implementation relevance.

Table 1. Summary of Included Studies.

No.	APA 7th Author List (In-Text)	Study Location	Target Group	Research Objective	Research Approach	Principal Outcomes
1	(Albikawi et al., 2025)	Northern Jordan	University students	Examine perceptions of AI mental health support and links with anxiety, depression, and help-seeking.	Quantitative survey.	Positive perceptions associated with distress, prior AI use, and help-seeking; AI viewed as adjunct, not replacement.
2	(Cheng, 2024)	China	University students	Develop and evaluate AI models to recognize depression from facial expressions and actions.	Quantitative study	AI models accurately identified depression in university students, with the fused model showing the best performance.
4	(Daza et al., 2023)	Multi-country (literature-based)	University students	Review machine-learning methods used to predict anxiety and stress.	Systematic review (PRISMA)	SVM and logistic regression were most effective for predicting anxiety and stress.
5	(Klos et al., 2021)	Argentina	University students	Evaluate the feasibility and impact of an AI chatbot on anxiety and depression.	Quantitative	Chatbot use reduced anxiety within the intervention group and showed good acceptability; no significant effect on depression.
6	(Kumar et al., 2024)	India	University students	Detect early risk of anxiety and depression using machine learning.	Quantitative	Random forest accurately identified students at mental health risk (~81% performance)
7	(Lai et al., 2025)	China	University students	Examine links between depression and use of conversational AI for companionship.	Quantitative	Depression increased AI companionship use, mediated by loneliness and moderated by gender and mind perception.
8	(Liu et al., 2022)	China	University students	CCompare effectiveness of an AI chatbot with bibliotherapy for depression.	Quantitative	Chatbot significantly reduced depression and anxiety and showed a stronger therapeutic alliance than bibliotherapy.
9	(Luo et al., 2025)	China	University students	Identify predictors of depression using machine learning.	Quantitative	Psychological factors were the strongest predictors; the model achieved high accuracy, with gender-specific risk patterns.
10	(Minaya et al., 2025)	Peru	University students	Develop and validate an AI system for early detection of depression, anxiety, and stress.	Quantitative	The AI system achieved high detection accuracy, significantly reduced assessment time, and achieved high user satisfaction.
11	(Nemesure et al., 2021)	USA	University students	Predict depression and anxiety using EHR-based machine learning.	Quantitative	Machine-learning models showed moderate accuracy in detecting depression and anxiety using non-psychiatric EHR data.
12	(Reyes-Portillo et al., 2025)	USA	University students	Assess the feasibility, acceptability, and preliminary effects of a generative AI mental wellness chatbot.	Quantitative	Chatbot use reduced anxiety, depression, and hopelessness and improved well-being, with high user satisfaction.
14	(Juárez-Santiago et al., 2025)	Mexico	University students	To evaluate deep learning models for early screening of anxiety, depression, and stress in higher education	Quantitative	Optimized deep learning models achieved >95% accuracy, demonstrating strong potential for AI-based mental health screening in universities
15	(Schaab et al., 2024)	Not location-specific	University students	To evaluate the performance of machine learning models in detecting depression, anxiety, and stress among undergraduates	Systematic review	Most models showed acceptable accuracy (>70%), but evidence quality was very low and largely based on internal validation, limiting real-world applicability
16	(Siddiqua et al., 2023)	Bangladesh	University students	To develop and validate an AI-based system for early detection and severity classification of depression among students	Quantitative	Optimized models achieved very high accuracy (up to 98.08%), with Random Forest showing the best performance and interpretability, supporting AI-based depression screening in higher education settings
17	(Puente-Torre et al., 2025)	Spain	University students	To analyze anxiety levels in university students and examine the role of technology in anxiety management	Quantitative	Technology-based interventions, particularly mindfulness applications, are associated with reduced anxiety levels among students, though findings indicate correlation rather than causation
18	(Wang et al., 2025)	China	University students	To evaluate the effectiveness of a CBT-based AI chatbot on depression and loneliness and examine the moderating role of financial stress	Quantitative	The AI chatbot significantly reduced depression and loneliness, particularly among students with high financial stress, with no significant effect on anxiety
19	(Xu & Ma, 2025)	China	College students	To compare the effectiveness of high-social-cue versus low-social-cue AI chatbots in reducing depression	Quantitative	Chatbots with high social cues produced greater reductions in depression and anxiety and higher adherence, satisfaction, and therapeutic alliance than text-only chatbots
20	(Yang & Rui, 2025)	China	University EFL students	To examine how AI-enhanced learning environments relate to students’ emotional health (depression and anxiety) and engagement	Quantitative study	Emotional health significantly mediates the relationship between AI-enhanced environments and student engagement, with depression and anxiety negatively affecting engagement while supportive AI features enhance it
21	(Yeh et al., 2025)	Taiwan	University students	To examine whether using the Woebot AI chatbot reduces depression and anxiety, and to assess student acceptance of chatbot-assisted learning	Mixed-methods	Woebot use did not produce statistically significant reductions in depression or anxiety, and student acceptance declined over time due to usability, language, and interaction limitations, despite some perceived cognitive and motivational benefits
22	(Zhai et al., 2025)	USA	College students	To develop machine learning models to predict anxiety and depressive disorders and guide prevention and intervention efforts	Quantitative study	Machine learning models showed good predictive accuracy (AUC ≈ 0.74–0.77), identifying key risk factors such as financial stress and low campus belonging for anxiety and depression among college students
23	(X. Zhang & Shi, 2025)	China	College students	To develop and validate an AI-driven framework for predicting psychological health risk and delivering interventions	Quantitative	The model accurately identified high-risk students and AI interventions improved academic performance
24	(T. Zhang et al., 2025)	China	College students	To model academic anxiety and examine the influence of professional commitment and achievement goal orientation	Quantitative study	The RF-SHAP model achieved high prediction accuracy (~97%), identifying mastery-avoidance goal orientation as the strongest contributor to academic anxiety

Table 1. Summary of Included Studies.

No.	APA 7th Author List (In-Text)	Study Location	Target Group	Research Objective	Research Approach	Principal Outcomes
1	(Albikawi et al., 2025)	Northern Jordan	University students	Examine perceptions of AI mental health support and links with anxiety, depression, and help-seeking.	Quantitative survey.	Positive perceptions associated with distress, prior AI use, and help-seeking; AI viewed as adjunct, not replacement.
2	(Cheng, 2024)	China	University students	Develop and evaluate AI models to recognize depression from facial expressions and actions.	Quantitative study	AI models accurately identified depression in university students, with the fused model showing the best performance.
4	(Daza et al., 2023)	Multi-country (literature-based)	University students	Review machine-learning methods used to predict anxiety and stress.	Systematic review (PRISMA)	SVM and logistic regression were most effective for predicting anxiety and stress.
5	(Klos et al., 2021)	Argentina	University students	Evaluate the feasibility and impact of an AI chatbot on anxiety and depression.	Quantitative	Chatbot use reduced anxiety within the intervention group and showed good acceptability; no significant effect on depression.
6	(Kumar et al., 2024)	India	University students	Detect early risk of anxiety and depression using machine learning.	Quantitative	Random forest accurately identified students at mental health risk (~81% performance)
7	(Lai et al., 2025)	China	University students	Examine links between depression and use of conversational AI for companionship.	Quantitative	Depression increased AI companionship use, mediated by loneliness and moderated by gender and mind perception.
8	(Liu et al., 2022)	China	University students	CCompare effectiveness of an AI chatbot with bibliotherapy for depression.	Quantitative	Chatbot significantly reduced depression and anxiety and showed a stronger therapeutic alliance than bibliotherapy.
9	(Luo et al., 2025)	China	University students	Identify predictors of depression using machine learning.	Quantitative	Psychological factors were the strongest predictors; the model achieved high accuracy, with gender-specific risk patterns.
10	(Minaya et al., 2025)	Peru	University students	Develop and validate an AI system for early detection of depression, anxiety, and stress.	Quantitative	The AI system achieved high detection accuracy, significantly reduced assessment time, and achieved high user satisfaction.
11	(Nemesure et al., 2021)	USA	University students	Predict depression and anxiety using EHR-based machine learning.	Quantitative	Machine-learning models showed moderate accuracy in detecting depression and anxiety using non-psychiatric EHR data.
12	(Reyes-Portillo et al., 2025)	USA	University students	Assess the feasibility, acceptability, and preliminary effects of a generative AI mental wellness chatbot.	Quantitative	Chatbot use reduced anxiety, depression, and hopelessness and improved well-being, with high user satisfaction.
14	(Juárez-Santiago et al., 2025)	Mexico	University students	To evaluate deep learning models for early screening of anxiety, depression, and stress in higher education	Quantitative	Optimized deep learning models achieved >95% accuracy, demonstrating strong potential for AI-based mental health screening in universities
15	(Schaab et al., 2024)	Not location-specific	University students	To evaluate the performance of machine learning models in detecting depression, anxiety, and stress among undergraduates	Systematic review	Most models showed acceptable accuracy (>70%), but evidence quality was very low and largely based on internal validation, limiting real-world applicability
16	(Siddiqua et al., 2023)	Bangladesh	University students	To develop and validate an AI-based system for early detection and severity classification of depression among students	Quantitative	Optimized models achieved very high accuracy (up to 98.08%), with Random Forest showing the best performance and interpretability, supporting AI-based depression screening in higher education settings
17	(Puente-Torre et al., 2025)	Spain	University students	To analyze anxiety levels in university students and examine the role of technology in anxiety management	Quantitative	Technology-based interventions, particularly mindfulness applications, are associated with reduced anxiety levels among students, though findings indicate correlation rather than causation
18	(Wang et al., 2025)	China	University students	To evaluate the effectiveness of a CBT-based AI chatbot on depression and loneliness and examine the moderating role of financial stress	Quantitative	The AI chatbot significantly reduced depression and loneliness, particularly among students with high financial stress, with no significant effect on anxiety
19	(Xu & Ma, 2025)	China	College students	To compare the effectiveness of high-social-cue versus low-social-cue AI chatbots in reducing depression	Quantitative	Chatbots with high social cues produced greater reductions in depression and anxiety and higher adherence, satisfaction, and therapeutic alliance than text-only chatbots
20	(Yang & Rui, 2025)	China	University EFL students	To examine how AI-enhanced learning environments relate to students’ emotional health (depression and anxiety) and engagement	Quantitative study	Emotional health significantly mediates the relationship between AI-enhanced environments and student engagement, with depression and anxiety negatively affecting engagement while supportive AI features enhance it
21	(Yeh et al., 2025)	Taiwan	University students	To examine whether using the Woebot AI chatbot reduces depression and anxiety, and to assess student acceptance of chatbot-assisted learning	Mixed-methods	Woebot use did not produce statistically significant reductions in depression or anxiety, and student acceptance declined over time due to usability, language, and interaction limitations, despite some perceived cognitive and motivational benefits
22	(Zhai et al., 2025)	USA	College students	To develop machine learning models to predict anxiety and depressive disorders and guide prevention and intervention efforts	Quantitative study	Machine learning models showed good predictive accuracy (AUC ≈ 0.74–0.77), identifying key risk factors such as financial stress and low campus belonging for anxiety and depression among college students
23	(X. Zhang & Shi, 2025)	China	College students	To develop and validate an AI-driven framework for predicting psychological health risk and delivering interventions	Quantitative	The model accurately identified high-risk students and AI interventions improved academic performance
24	(T. Zhang et al., 2025)	China	College students	To model academic anxiety and examine the influence of professional commitment and achievement goal orientation	Quantitative study	The RF-SHAP model achieved high prediction accuracy (~97%), identifying mastery-avoidance goal orientation as the strongest contributor to academic anxiety

4.1. The Impact of AI-Driven Mental Health Chatbots on Anxiety and Depression Among College Students

Effectiveness of AI-Driven Chatbots in Reducing Anxiety and Depression

Empirical evidence indicates that AI-driven mental health chatbots can reduce symptoms of anxiety and depression among college students by offering accessible, structured psychological support beyond traditional counseling settings (Liu et al., 2022). Randomized controlled trials demonstrate that CBT-based chatbot interventions are associated with statistically significant reductions in depressive and anxiety symptoms among university students compared with bibliotherapy control conditions (Liu et al., 2022).Findings from longitudinal, open-label randomized trials indicate that sustained engagement with chatbot interventions can maintain reductions in anxiety and depressive symptoms, supporting their potential scalability as mental health resources in higher education (Xu & Ma, 2025). However, observed differences in effect sizes across studies suggest that chatbot effectiveness is not uniform and may be influenced by factors such as intervention duration, baseline symptom severity, and levels of student engagement (Liu et al., 2022; Xu & Ma, 2025). Collectively, these findings indicate that AI-driven chatbots, particularly those incorporating richer social cues or structured CBT-based interactions, hold therapeutic promise for reducing depressive and, to a lesser extent, anxiety symptoms among college students; however, variability in outcomes across chatbot designs, user engagement, and contextual factors suggests that results should be interpreted cautiously across diverse institutional and implementation settings (Wang et al., 2025; Xu & Ma, 2025).

Influence of Chatbot Design and Engagement on Mental Health Outcomes

Beyond chatbot interventions, design features, and user engagement, critical contextual factors, such as financial stress and students’ broader socioeconomic conditions, play a central role in shaping mental health outcomes among college students (Daza et al., 2023; Wang et al., 2025).Evidence indicates that AI-driven chatbots that incorporate social cues, such as voice, animation, or human-like interaction, are associated with greater reductions in anxiety and depression than text-only chatbot formats (Liu et al., 2022; Xu & Ma, 2025).Studies suggest that conversational adaptability and structured, evidence-based therapeutic dialogue are associated with higher user engagement and satisfaction, which coincide with improvements in depression and anxiety outcomes among college students (Reyes-Portillo et al., 2025). Conversely, evidence suggests that technical constraints, limited interactivity, and poor alignment between chatbot functionality and students’ academic and emotional contexts can diminish acceptance and sustained engagement, ultimately constraining therapeutic outcomes (Albikawi et al., 2025; Yeh et al., 2025).These findings highlight that chatbot effectiveness depends not only on artificial intelligence capabilities but also on human-centered design and contextual relevance within higher education environments (Albikawi et al., 2025; Yeh et al., 2025).

Contextual Limitations and Implications for Higher Education Mental Health Support

Despite generally positive findings, several studies report mixed or non-significant outcomes, underscoring important contextual and implementation-related limitations in using AI-driven chatbots to support student mental health (Wang et al., 2025; Yeh et al., 2025). In academically demanding environments, chatbot-based interventions may show limited effects on anxiety and depression, indicating that academic context and workload may constrain intervention uptake and effectiveness (Yeh et al., 2025). Additionally, reliance on self-report measures and the short follow-up period limits conclusions regarding long-term psychological benefits (Klos et al., 2021). In Klos et al. (2021), outcomes were assessed over an eight-week intervention, and the findings were described as preliminary. As a result, observed symptom changes should be interpreted as short-term effects rather than evidence of sustained therapeutic impact, underscoring the need for longer-term evaluation. From an institutional perspective, these findings suggest that AI-driven chatbots should be integrated as complementary supports within campus mental health systems, providing accessible psychoeducation and early intervention while facilitating engagement with professional care (Liu et al., 2022; Minaya et al., 2025; X. Zhang & Shi, 2025). Future research should prioritize longitudinal evaluation, cross-cultural validation, and implementation-focused approaches to clarify how AI-driven chatbots can be effectively integrated into higher education mental health ecosystems (Wang et al., 2025).

4.2. Characterization of AI-Driven Mental Health Chatbot Applications for Depression Management Among College Students

CBT-Based Therapeutic Chatbots (e.g., Woebot, Tess)

AI-driven chatbots such as Tess provide adaptive, text-based mental health support integrating multiple therapeutic approaches and dynamically tailored through user feedback (Klos et al., 2021).High engagement and a positive feedback–interaction link highlight perceived empathy as central to sustained use (Klos et al., 2021).Although between-group effects were non-significant, within-group reductions in anxiety indicate its potential as a low-intensity, accessible intervention in higher education (Klos et al., 2021).Yeh et al. (2025) show that while Woebot offers structured CBT-based support and perceived motivational benefits, it does not produce measurable reductions in anxiety or depression. The decline in acceptance over time further suggests that usability and contextual misalignment can undermine sustained engagement. These findings indicate that effectiveness depends less on therapeutic design alone and more on user experience and educational fit (Yeh et al., 2025). AI mental health chatbots rely on structured CBT protocols that support consistency but may oversimplify diverse student experiences (Liu et al., 2022). While effective for scalability, their rigidity can limit personalization and responsiveness. This tension highlights a key limitation: standardization may come at the cost of meaningful, individualized support (Liu et al., 2022).

Socially Cued and Multimodal Chatbots

Chatbots with social cues improve adherence, therapeutic alliance, and emotional engagement compared to text-only systems, despite similar CBT content (Xu & Ma, 2025)This aligns with evidence that socially responsive AI enhances perceived empathy and user connection (Lai et al., 2025). However, declining engagement over time raises concerns about long-term effectiveness (Xu & Ma, 2025) For example, although university students exchanged a substantial volume of messages and reported high acceptability of an AI-based chatbot intervention, the clinical effects on depressive symptoms remained modest, suggesting that interaction frequency alone may be insufficient to drive sustained symptom change (Klos et al., 2021).These findings raise design considerations regarding how social cues are integrated, indicating that engagement-enhancing features should complement evidence-based therapeutic frameworks rather than operate independently of them (Wang et al., 2025).

Functional Integration Within Higher Education Mental Health Ecosystems

Liu et al. (2022) indicate that AI-driven chatbots are intended to function as bounded self-help supports within established mental health care pathways, emphasizing professional oversight, safety protocols, and integration with existing services rather than autonomous treatment delivery. Building on this view, Lai et al. (2025) describe conversational AI primarily as a source of emotional companionship for students with depressive symptoms, highlighting its role in alleviating loneliness and reinforcing its function as a supportive, compensatory resource alongside, rather than in place of, formal mental health services. Depressive experiences among students are heterogeneous, limiting the effectiveness of any single chatbot model (Albikawi et al., 2025). Chatbots mainly expand access and reduce stigma but function as entry-level supports rather than replacements for professional care (Liu et al., 2022; Yeh et al., 2025). Overall, evidence supports a layered care model where chatbots complement broader institutional mental health systems (Albikawi et al., 2025).

4.3. Implementation of AI-Driven Mental Health Chatbots in Higher Education Settings

Institutional Embedding and Care Pathway Integration

Implementation studies consistently position AI-driven mental health chatbots as adjunctive digital tools that extend access and support early engagement, while formal assessment, clinical decision-making, and sustained care remain anchored within institutional counseling and professional mental health services (Albikawi et al., 2025; Juárez-Santiago et al., 2025; Puente-Torre et al., 2025). Albikawi et al. (2025) acknowledge ethical and acceptability constraints but stop short of specifying how chatbot outputs should be operationalized within counseling workflows, limiting their translational impact. Puente-Torre et al. (2025) demonstrate the potential of digital tools to reduce anxiety and stigma; however, the absence of institutional coordination or referral mechanisms positions these interventions as isolated supports that depend on sustained student self-motivation. Juárez-Santiago et al. (2026) advance technical feasibility for early screening, yet their model remains diagnostically oriented, with little consideration of escalation pathways or continuity of care following risk identification. While this positioning expands initial access to support, it simultaneously constrains clinical continuity, particularly for students whose depressive symptoms exceed the scope of self-guided intervention and require professional escalation (Albikawi et al., 2025). The evidence suggests that implementation effectiveness is determined more by the extent to which institutions intentionally align chatbot services with existing mental health infrastructure and clearly define their functions within coordinated care pathways (Albikawi et al., 2025; Juárez-Santiago et al., 2025).

Ethical Governance, Data Stewardship, and Institutional Responsibility

The lack of standardized methodological practices limits transparency and confidence in machine learning–based mental health applications (Daza et al., 2023). This concern is reinforced by evidence of weak validation and low-certainty findings across studies, which further constrain the reliability and real-world applicability of these models (Schaab et al., 2024). Although machine learning approaches can accurately identify students at elevated risk for depression, current predictive studies do not articulate how institutions should act upon high-risk classifications, highlighting a critical gap between predictive modeling and institutional responsibility for care pathways (Luo et al., 2025).

Extending this concern, Schaab et al. (2024) note that although many machine learning models demonstrate adequate predictive performance, the overall quality of evidence remains very low, and the practical applicability of these models within real-world student mental health contexts is insufficiently addressed. Whereas Cheng (2024) centers on technical accuracy in AI-based depression recognition, Shi and Zhang (2025) emphasize institution-level governance frameworks that define accountability, ethical oversight, and data stewardship in clinically sensitive AI systems.

Contextual Adaptation, Sustainability, and System-Level Viability

At the system level, the sustainability of chatbot implementation in higher education appears to depend on contextual responsiveness and institutional commitment, rather than on short-term scalability narratives alone, a position consistent with evidence that technology-based interventions demonstrate context-dependent and non-causal effects (Puente-Torre et al., 2025). Chatbots may be cost-effective, but their effectiveness depends on alignment with students’ emotional and contextual needs, as poorly adapted AI systems can increase anxiety or disengagement rather than sustain support (Yang & Rui, 2025). This is supported by evidence that student engagement with AI tools is shaped by emotional relevance, usability, and perceived usefulness, with weak alignment leading to reduced acceptance and engagement (Wang et al., 2025; Yeh et al., 2025). Evidence suggests that student engagement with mental health chatbots depends on perceived trust and emotional credibility rather than institutional deployment and may therefore remain compensatory rather than a durable component of depression support in higher education (Lai et al., 2025).

5. How This Review Contributes New Insights

This work advances the literature by moving beyond outcome-focused assessments of AI-driven mental health chatbots to provide an integrated, systems-level analysis of their role within higher education mental health ecosystems. Whereas prior scholarship has largely emphasized symptom reduction or examined chatbot interventions in isolation, this synthesis integrates evidence across effectiveness, application types, and implementation contexts to clarify how chatbot technologies operate in real-world institutional settings rather than under idealized conditions. By organizing the analysis around therapeutic impact, functional differentiation, and institutional integration, this study introduces a multidimensional perspective that remains underdeveloped in existing research.

A central contribution lies in the conceptual reframing of AI-driven mental health chatbots as sociotechnical interventions embedded within organizational, ethical, and contextual structures, rather than as standalone digital solutions. The analysis demonstrates that chatbot effectiveness is closely tied to institutional governance, care pathways, and contextual alignment, challenging technology-centric narratives that privilege scalability while overlooking accountability and continuity of care. This framing helps explain the variability and mixed outcomes observed across studies and highlights the decisive role of implementation context in shaping student engagement and therapeutic relevance.

This work further contributes by systematically distinguishing between chatbot application types and their functional roles in supporting students experiencing depression. By differentiating CBT-based, generative, and socially cued chatbots according to therapeutic orientation, engagement mechanisms, and suitability across levels of student need, the analysis provides a more precise conceptual vocabulary for future research and supports more targeted evaluation and deployment strategies in higher education contexts.

The findings situate AI-driven mental health chatbots within the broader landscape of traditional counseling services, emphasizing their function as complementary supports rather than substitutes for clinician-led care. The evidence indicates that chatbots primarily redistribute low-intensity, early-stage support while preserving professional intervention for complex and high-risk cases, thereby clarifying ongoing debates about supplementation versus replacement and offering an evidence-informed basis for institutional decision-making.

By foregrounding implementation considerations, including governance, equity, sustainability, and system-level viability, this analysis identifies critical gaps that extend beyond technical performance and short-term efficacy. These insights point toward the need for longitudinal, implementation-focused, and ethically grounded research to support the responsible and sustainable integration of AI-driven mental health chatbots within higher education mental health systems.

6. Limitations of Using AI for Screening and Diagnosis of Anxiety and Depression

The use of artificial intelligence for screening and diagnosing anxiety and depression is accompanied by substantial limitations related to data security, clinical reliability, ethical accountability, and equitable implementation. AI-based mental health systems frequently rely on distributed digital infrastructures and external data storage, which increases exposure to privacy breaches and unauthorized access to highly sensitive psychological information. Such risks are particularly consequential in mental health contexts, where confidentiality is foundational to patient trust and engagement. Concerns regarding diagnostic accuracy further constrain the clinical applicability of AI-driven screening tools. Many systems rely on indirect indicators, such as self-reported data or physiological signals captured by wearable devices, which may lack sufficient specificity to distinguish between overlapping psychiatric symptoms or contextual stress responses. As a result, AI-generated assessments may misclassify symptom severity or fail to account for situational and subjective factors critical to accurate diagnosis.

Ethical accountability represents an additional challenge in AI-assisted mental health assessment. When algorithmic errors occur, responsibility is often distributed across developers, healthcare institutions, and clinicians, creating ambiguity regarding liability and patient recourse. This lack of clearly defined accountability is especially problematic when AI-informed decisions influence diagnostic labeling or treatment pathways with potential long-term consequences for individuals. Limitations related to transparency and patient autonomy also remain salient. Many AI systems operate through opaque decision-making processes, offering limited interpretability of how screening or diagnostic conclusions are reached. This opacity undermines informed consent and restricts patients’ ability to critically engage with or challenge AI-supported assessments, which is inconsistent with principles of shared decision-making in mental health care.

The intrinsic complexity of anxiety and depressive disorders further limits the capacity of AI systems to function as autonomous diagnostic tools. Mental health assessment relies heavily on empathic understanding, contextual interpretation, and the therapeutic relationship, elements that remain beyond the capabilities of current AI technologies. While wearable devices and automated tools can capture physiological or behavioral data, they cannot adequately interpret subjective experience, emotional nuance, or interpersonal meaning, all of which are central to psychiatric diagnosis and treatment. Consequently, AI-based screening tools are best understood as complementary supports rather than replacements for trained mental health professionals.

Broader concerns regarding acceptability and equity also restrict widespread adoption. Cultural attitudes toward machine-mediated mental health care vary considerably, and resistance to AI involvement may limit engagement in certain populations. In addition, disparities in technological infrastructure, digital literacy, and professional training may confine the benefits of AI-based screening to well-resourced settings, potentially exacerbating existing inequalities in mental health service access. Bias embedded within training data further raises the risk of discriminatory outcomes, particularly for marginalized or underrepresented groups.

The absence of comprehensive regulatory frameworks, standardized clinical guidelines, and universally accepted ethical standards continues to impede the responsible integration of AI into mental health screening and diagnosis. Without clear legal and professional governance, consistency, safety, and accountability remain difficult to ensure. Together, these limitations underscore the need for cautious, ethically grounded application of AI in mental health screening, reinforcing its role as an adjunct to, rather than a substitute for, human clinical expertise in the assessment of anxiety and depression

7. Future of Artificial Intelligence in Mental Health

The future of artificial intelligence in mental health depends on its successful translation from experimental algorithms to clinically and institutionally integrated applications. Current evidence indicates that AI-based tools for screening, monitoring, and intervention remain at an early stage of implementation, with their practical value contingent on rigorous validation and contextual deployment rather than technological novelty alone. As research advances, the central challenge will be ensuring that AI systems meaningfully contribute to early identification and support for anxiety and depression without compromising clinical integrity or patient trust.

In clinical and higher education contexts, AI-powered screening systems may increasingly support routine assessment, triage, and longitudinal monitoring by identifying risk patterns that warrant professional evaluation. Similarly, AI-driven mental health chatbots may continue to function as low-threshold, scalable supports that complement existing services by providing psychoeducation, self-regulation strategies, and interim support between clinical encounters. The effectiveness of such applications will depend on their integration within established care pathways rather than their operation as standalone tools. Responsible advancement of AI in mental health requires sustained attention to ethical governance, including data privacy, transparency, bias mitigation, and equitable access. Robust external validation and continuous performance monitoring will be essential to ensure reliability across diverse populations and institutional settings. Without these safeguards, the expansion of AI-based mental health tools risks amplifying existing disparities and undermining confidence in digital care.

The future of AI in mental health lies in augmenting rather than replacing human expertise. AI systems are best positioned as decision-support tools that enhance clinicians’ capacity for early detection, prioritization, and personalized care, while preserving the relational, interpretive, and ethical dimensions central to mental health practice. Achieving this balance will require interdisciplinary collaboration among clinicians, researchers, technologists, and policymakers to align AI development with clinical workflows and patient-centered values. The trajectory of AI in mental health points toward a hybrid model in which human judgment and artificial intelligence operate synergistically. Continued research focused on implementation, governance, and long-term outcomes will be critical to realizing AI’s potential as a sustainable and ethically grounded contributor to mental health care.

8. Conclusion

As interest in digital approaches to student mental health continues to grow, this article synthesizes current evidence on the role of AI-enabled conversational tools in higher education contexts. The analysis indicates that these technologies can contribute to alleviating anxiety and depressive symptoms by broadening access to early and low-intensity forms of psychological support, although effectiveness varies according to design features and institutional conditions. Differences across chatbot architectures reveal that therapeutic focus, interaction modality, and engagement logic shape how these systems address depression rather than offering uniform solutions. Evidence further suggests that the greatest value of such tools emerges when they are incorporated into coordinated campus care structures instead of operating as independent interventions. Collectively, these insights support future theory development and interdisciplinary investigation to advance inclusive, context-aware, and equity-oriented mental health strategies in higher education.