Application and effectiveness of artificial intelligence in adolescent mental health support and therapy: a scoping review Authors: Lu Liang, Xiaoran Lu, Dan Jiao, Bing Xiao, Fang Shan, Ruojun Zhang, Congcong Zhao, Jian Song Published: 2026 ABSTRACT BackgroundAdolescents are highly vulnerable to mental health conditions, yet traditional counseling models may not fully address their evolving needs. The rise of artificial intelligence (AI) has created new opportunities and challenges for supporting adolescent mental health.ObjectiveThis scoping review evaluates the current applications, outcomes, strengths, drawbacks, and technical and ethical challenges of AI in adolescent mental health.MethodsThe review adopted the PCC framework and was conducted in accordance with JBI guidelines, with reporting following the PRISMA-ScR framework. A systematic search was performed across PubMed, Embase, Web of Science, and ScienceDirect for peer-reviewed studies and grey literature published in English between 2015 and July 24, 2025. Title, abstract, and full-text screenings were independently conducted by two reviewers, with disagreements resolved by consensus or consultation with a third reviewer.ResultsOf the screened records, 24 studies met the eligibility criteria, all of which were published between 2020 and 2025. USA and China were the leading contributors. AI was primarily applied to address depression, anxiety, stress, suicidal ideation, and other mental health conditions. AI demonstrated promise in accurate risk prediction, enhanced clinical and self management, personalized care, expanded service access, and stigma reduction. However, persistent technical challenges included potential algorithmic diagnostic errors, reliance on self-reported and screen-based data, and complex interface design. Ethical concerns included data security, privacy, limited transparency, attribution of responsibility, and deficiencies in informed consent processes.ConclusionsAI can accurately identify and predict mental health risks, overcome geographical and temporal barriers in adolescent mental healthcare, enable efficient and personalized support, facilitate clinical management, and expand access for emergency crises, marginalized populations, and underserved areas. However, complex study protocols and severe mental health conditions may reduce adolescent engagement. The long-term sustainability of AI’s therapeutic effects requires further validation. Future efforts should focus on expanding the evidence base, validating interventions in real-world settings, strengthening regulatory frameworks, and enhancing the digital competencies of healthcare providers.Review registrationOpen Science Framework https://osf.io/79wya.Trial registrationNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12888-026-08190-y. Introduction Adolescence is a transitional phase marked by significant physical, psychological, social, and moral development, bridging childhood and adulthood. The World Health Organization (WHO) defines adolescence as the age range between 10 and 19 years [1, 2]. However, during adolescence, a period of independent development, the mental health of many individuals is at risk [3], with anxiety, depression, and stress-related psychological disorders affecting a significant portion of this age group [4–7]. According to the WHO, up to 20% of adolescents experience mental health issues, underscoring the urgent need for effective and accessible intervention strategies [8–10]. The psychological and emotional needs of adolescents encompass functional impairments, social isolation, emotional loneliness [11], and the need for effective management strategies [12]. There is robust evidence indicating that social isolation and loneliness are associated with increased all-cause mortality, cardiovascular disease, as well as depression and anxiety [13]. Providing adolescents with the necessary and effective mental health support and treatment has become a critical public health issue. Adolescents are in a developmental stage that requires more specialized approaches to mental health intervention. Due to their introverted nature and heightened sensitivity, adolescents often find it difficult to access or trust traditional mental health services [14]. Moreover, the generational gap between adolescents and adults, coupled with the fear of criticism and misunderstanding, makes adolescents more inclined to seek out new forms of psychotherapy that they perceive as more accessible and less judgmental [15]. Research indicates that, with nearly 95% of adolescents aged 13 to 17 using social media [16], this population is particularly receptive to emerging technologies such as artificial intelligence (AI) [17]. Adolescents with mental health concerns reported receiving diverse forms of social support from chatbots, namely informational, emotional, practical, and evaluative support [18]. As a result, AI applications and chat platforms have emerged as promising tools for addressing adolescent mental health. AI refers to any computer program that exhibits intelligent behavior—such as problem-solving, learning, perception, and reasoning—to emulate, augment, and expand human intelligence across a broad spectrum of fields, including knowledge discovery, data mining, natural language processing (NLP) [19], computer vision, speech recognition, expert systems [20], conversational AI (CAI) [21], and generative AI [22]. The application of information technology in mental health dates back to 1966, when the ELIZA program was developed as the first computer program capable of simulating a psychotherapist and conducting simple dialogues [23]. Subsequently, the emergence of applications such as Woebot, Wysa, and Replika highlighted the increasing focus on the application and regulation of AI in adolescent mental health. However, there is ongoing debate regarding the use of AI in adolescent mental health. Supporters argue that AI holds significant potential in diagnosing disorders, predicting mental health crises, and alerting healthcare professionals [24]. AI has been shown to significantly impact the prediction and treatment of childhood obsessive-compulsive disorder [25], attention deficit hyperactivity disorder [26], and autism across all age groups [27]. Beyond managing or alleviating adolescent depression and anxiety [28, 29], AI also provides mental health support in underserved areas [30] low- and middle-income countries [31] and crisis situations, such as natural disasters and conflict [32, 33]. Opponents highlight legal gaps and risks concerning trust in AI decision-making for mental health [34]. Some scholars express concerns that using AI to support and treat adolescent mental health could raise issues related to privacy and security [35], algorithmic bias, informed consent, accountability, and other potential risks [36–39]. For example, a boy with autism insisted that his mother disliked him simply because ChatGPT had stated so, and this conversation with ChatGPT during a clinical session reshaped his emotions, interpersonal relationships, and risk status[40]. Although recent reviews have begun to outline the landscape of AI in adolescent mental health, research gaps still remain. Sharma et al. [41] found that AI applications in adolescent mental health have focused predominantly on diagnosis, monitoring, and assessment, with limited attention to treatment and prognosis. This imbalance motivated our focus on AI-driven interventions for mental health support and therapy. This review aims to deepen the understanding of AI applications in adolescent mental health support and therapy. By “support and therapy”, we mean AI applications aimed at improving adolescents’ mental health, such as enhancing their engagement in therapeutic processes, increasing their satisfaction with therapy, and lowering their stress levels, among other positive outcomes. We chose 2015 as the starting year for the literature search, as it marked a period when researchers began exploring AI tools for supporting children with autism [42] and youths at clinical high-risk (CHR) for psychosis [43], investigating their potential as objective assessment and prediction tools in psychological interventions. That same year, Tsinghua University launched Teenchat, a virtual companion app designed for adolescents, which aims to help them cope with pressures related to academic work, self-perception, interpersonal relationships, and emotions [44]. Thus, initiating the literature search from 2015 ensures a more comprehensive review. Finally, we analyzed the 24 selected studies, focusing on the practical utility, outcomes, strengths and drawbacks, technical issues, and ethical considerations of AI tools. Materials and methods A comprehensive scoping review was conducted, encompassing both peer-reviewed and grey literature, to explore the use of AI in the support and therapy of adolescent mental health. The review followed the methodological framework provided by Joanna Briggs Institute (JBI) [45]. To ensure transparency and adherence to reporting standards, the preferred reporting items for systematic reviews and meta-analyses extension for scoping reviews (PRISMA-ScR) served as a guide for reporting the review [46]. Protocol and registration The study protocol was registered with the Open Science Framework on August 12, 2025, an international platform for the registration of reviews (https://osf.io/79wya). To ensure the quality of the review, we followed the PRISMA-ScRs guidelines and used the PRISMA flowchart [45]. The complete list of included studies can be found in Supplementary Material S1. Identifying the research question Based on the objectives of this scoping review, we have formulated the following questions. What specific types of mental health are AI tools predominantly applied to?What are the different types or names of AI tools that are currently available for assisting in optimizing the adolescent mental health support and therapy process in the studies reviewed?What are the outcomes, strengths, and drawbacks associated with the use of AI tools in assisting in adolescent mental health support and therapy?What are the ethical and technical issues encountered in the utilization of AI tools to assist in adolescent mental health support and therapy? What specific types of mental health are AI tools predominantly applied to? What specific types of mental health are AI tools predominantly applied to? What are the different types or names of AI tools that are currently available for assisting in optimizing the adolescent mental health support and therapy process in the studies reviewed? What are the different types or names of AI tools that are currently available for assisting in optimizing the adolescent mental health support and therapy process in the studies reviewed? What are the outcomes, strengths, and drawbacks associated with the use of AI tools in assisting in adolescent mental health support and therapy? What are the outcomes, strengths, and drawbacks associated with the use of AI tools in assisting in adolescent mental health support and therapy? What are the ethical and technical issues encountered in the utilization of AI tools to assist in adolescent mental health support and therapy? What are the ethical and technical issues encountered in the utilization of AI tools to assist in adolescent mental health support and therapy? Identifying relevant studies An electronic search was conducted in leading medical and computer science databases, including PubMed, Embase, Web of Science, and ScienceDirect. To minimize potential publication bias, grey literature sources were also searched. Website searches were conducted in Google Scholar, PAIS Index, and the United Nations iLibrary. In Google Scholar, the first 200 results sorted by relevance were screened. In addition, reports published by major international organizations relevant to adolescent mental health were manually searched, including those from the World Health Organization (WHO), the National Institutes of Health (NIH), the Centers for Disease Control and Prevention (CDC), and the American Psychological Association (APA). Relevant preprint servers (medRxiv and PsyArXiv) were also screened where applicable. Furthermore, backward and forward citation tracking was performed for all included studies to identify additional relevant publications. A search was performed on August 14, 2025 by LL. English papers published from 2015 through July 24, 2025 were included in this study. Our study methods, including the inclusion/exclusion criteria and the data extraction framework, have been refined since the initial registration on OSF. A comprehensive comparison and detailed justifications for these modifications are provided in Supplementary Material S2. The search terms included artificial intelligence, adolescent emotional support/therapy, adolescent mental health support/therapy (see Supplementary Material S3 for full search strategy). Search terms were different according to different databases. Boolean terms were used to separate the key words. To compile all relevant evidence sources, identify, remove duplicate studies, and screen literature, Endnote 21 was employed. Study selection To gain a clearer understanding of the application and existing challenges of AI in adolescent mental health, this review primarily used the PCC (Population, Concept, Context) framework to define the research boundaries and core content. These criteria were developed based on the JBI approach, which emphasizes the formulation of scoping review questions that describe the population, concept and context of the study [44]. The details of this framework are presented below. The scoping review sourced all relevant peer-reviewed and grey literature that describes the application of AI in adolescent mental health support and therapy. The population sample for the review were individuals who use AI technology in mental health disease, specifically focusing on the therapy processes, participants of aged 10–19 years with sub-clinical or clinical levels of mental health conditions were included.The context of the study focused on Al-assisted adolescent mental health support and therapy. The review encompassed studies and grey literature, with no geographical restrictions, to ensure a comprehensive evaluation of the development, implementation and assessment of AI technology in the field. By conducting an extensive search, the review aimed to gather a wide range of references to effectively categorize the nature and types of AI technology used for adolescent mental health support and therapy. This approach provided a more comprehensive understanding of the topic.AI application in adolescent mental health support and therapy is defined as the use of diverse AI techniques to facilitate treatment or support interventions. This definition encompasses all major sub-fields of AI, including machine learning (ML), deep learning (DL), natural language processing (NLP), computer vision, AI-driven tools, and digital applications, etc. The scoping review sourced all relevant peer-reviewed and grey literature that describes the application of AI in adolescent mental health support and therapy. The population sample for the review were individuals who use AI technology in mental health disease, specifically focusing on the therapy processes, participants of aged 10–19 years with sub-clinical or clinical levels of mental health conditions were included. The scoping review sourced all relevant peer-reviewed and grey literature that describes the application of AI in adolescent mental health support and therapy. The population sample for the review were individuals who use AI technology in mental health disease, specifically focusing on the therapy processes, participants of aged 10–19 years with sub-clinical or clinical levels of mental health conditions were included. The context of the study focused on Al-assisted adolescent mental health support and therapy. The review encompassed studies and grey literature, with no geographical restrictions, to ensure a comprehensive evaluation of the development, implementation and assessment of AI technology in the field. By conducting an extensive search, the review aimed to gather a wide range of references to effectively categorize the nature and types of AI technology used for adolescent mental health support and therapy. This approach provided a more comprehensive understanding of the topic. The context of the study focused on Al-assisted adolescent mental health support and therapy. The review encompassed studies and grey literature, with no geographical restrictions, to ensure a comprehensive evaluation of the development, implementation and assessment of AI technology in the field. By conducting an extensive search, the review aimed to gather a wide range of references to effectively categorize the nature and types of AI technology used for adolescent mental health support and therapy. This approach provided a more comprehensive understanding of the topic. AI application in adolescent mental health support and therapy is defined as the use of diverse AI techniques to facilitate treatment or support interventions. This definition encompasses all major sub-fields of AI, including machine learning (ML), deep learning (DL), natural language processing (NLP), computer vision, AI-driven tools, and digital applications, etc. AI application in adolescent mental health support and therapy is defined as the use of diverse AI techniques to facilitate treatment or support interventions. This definition encompasses all major sub-fields of AI, including machine learning (ML), deep learning (DL), natural language processing (NLP), computer vision, AI-driven tools, and digital applications, etc. Inclusion criteria To be included in this scoping review, studies must meet the following criteria: The study must be available in full text.The study must focus on the application of AI technology.The study should be original research articles, including observational study and experimental study, published in peer-reviewed journals.The study must specifically address the use of AI technology to assist in adolescent mental health support and therapy. The study must be available in full text. The study must be available in full text. The study must focus on the application of AI technology. The study must focus on the application of AI technology. The study should be original research articles, including observational study and experimental study, published in peer-reviewed journals. The study should be original research articles, including observational study and experimental study, published in peer-reviewed journals. The study must specifically address the use of AI technology to assist in adolescent mental health support and therapy. The study must specifically address the use of AI technology to assist in adolescent mental health support and therapy. Exclusion criteria Studies were excluded if they: Not be available in full text.Not include the details of AI technology used.Not focus on the AI technology used with adolescents, but adults and infants (aged over 19 or under 10 years).Focused on population comprised of adolescents without clear mental health issues, such as symptoms of depression or anxiety.Not original articles, but reviews, editorials, comments, abstracts, conference proceedings, dissertations, posters, books, book chapters, reports, and letters.Focused solely on the technical descriptions of AI as a technique rather than its exact application in treating or supporting adolescent mental health. Not be available in full text. Not be available in full text. Not include the details of AI technology used. Not include the details of AI technology used. Not focus on the AI technology used with adolescents, but adults and infants (aged over 19 or under 10 years). Not focus on the AI technology used with adolescents, but adults and infants (aged over 19 or under 10 years). Focused on population comprised of adolescents without clear mental health issues, such as symptoms of depression or anxiety. Focused on population comprised of adolescents without clear mental health issues, such as symptoms of depression or anxiety. Not original articles, but reviews, editorials, comments, abstracts, conference proceedings, dissertations, posters, books, book chapters, reports, and letters. Not original articles, but reviews, editorials, comments, abstracts, conference proceedings, dissertations, posters, books, book chapters, reports, and letters. Focused solely on the technical descriptions of AI as a technique rather than its exact application in treating or supporting adolescent mental health. Focused solely on the technical descriptions of AI as a technique rather than its exact application in treating or supporting adolescent mental health. Charting the evidence All search results were managed centrally in EndNote 21. After duplicate removal, titles and abstracts were screened against predefined inclusion criteria. Studies meeting scope evaluation criteria were uploaded to Covidence and assessed independently by two evaluators (LL and ZCC). Studies not meeting inclusion criteria were excluded based on their assessment. Disagreements were resolved through discussion, with involvement of a third evaluator (LXR) if needed. The literature search flowchart is shown in Fig. 1. Fig. 1PRISMA diagram of studies inclusion and exclusion PRISMA diagram of studies inclusion and exclusion Data extraction and synthesis Research data were transferred to an Excel-based extraction form for systematic organization and consistency. A data extraction form was developed to systematically gather relevant information from the included studies. ZCC and LL independently extracted the data from the selected studies using a pilot-tested form. In cases of discrepancies, LXR was consulted to reach a consensus. For grey literature, limitations related to reliable referencing and platform constraints, such as the absence of unique identifiers and the inability to export complete search result lists from certain websites, prevented the use of digital tools (e.g., Covidence or Excel) for duplicate removal. Instead, duplicates were identified and removed manually during the screening and data extraction process, following the same procedure applied to peer-reviewed articles. The final data extraction records for both peer-reviewed and grey literature are presented in an Excel file included in the Supplementary Material S4. Extracted details included publication information (title, authors, year, country), sample characteristics (participant number, age, gender, mental health disease), research design (AI tools used, technical issues, outcomes, strengths, drawbacks), and ethical considerations, where available. Results Study selection The database search yielded 849 studies. After removing 78 duplicates, 771 studies remained. Of these, 562 were excluded based on title and abstract evaluation. Following full-text review of the remaining 209 studies, 185 studies were screened out. 24 were included in this review. In addition, we conducted a grey literature search, which identified 22 studies. After removing three duplicates, the remaining 19 studies were assessed for eligibility. However, none met the predefined inclusion criteria and were therefore excluded (see Supplementary Material S5 for key data extracted for each of the 24 studies and funding information, see Table 1 for key data extracted from each of the 24 studies). Table 1Characteristics of the studies included in the reviewFirst AuthorYearCountryStudy DesignType of AISpecific AI Technology/ToolMental Health DiseaseBiernesser C [47]2023USAQualitative studyAI-enabled digital platformsFlourishSuicidal ideation (LGBTQ+ youth)Bogdanski E [48]2023USAExploratory case studyAI-powered toolTheraVR (VR system)PTSDHill ED [49]2025USALongitudinal cohort studyML modelSymptom-driven model (CBCL scales)Mental health riskNicol G [50]2022USAPilot RCTAI-powered toolW-GenZ appDepression and anxietyPaschali M [51]2022USALongitudinal studyML modelDLSuicidal ideation and depressionHuang Y [52]2022ChinaCross-sectional studyML modelRF, SVM, DTDepression and suicidal ideationJiang J [53]2025ChinaQuasi-experimental studyAI-powered toolGymBuddy, Elomia (chatbots)Psychological distressKuang Y [54]2025ChinaCross-sectional studyML modelFL, LR, RF, XGBoostDepressionZheng Y [29]2022ChinaSimulation experimentML modelLR, DLSocial anxietyBeaudequin D [55]2021AustraliaCohort studyML modelBNPsychological distressSu R [56]2023AustraliaLongitudinal studyML modelRFSelf-harm and suicide attemptsFujita J [57]2025JapanRCTAI-powered toolemol (chatbot)DepressionGuo Y [58]2024JapanRCTAI-enabled digital platformsAgent technology family education systemComprehensiveJohnson E [59]2020SpainQuantitative researchAI-powered toolMAmIoTie (affective cognitive toy)Mood disordersMartín-Ávila J [60]2025SpainRCTAI-enabled digital platformsemoTICareDepression and anxiety (Type 1 diabetes)Prakash GH [61]2025IndiaQuasi-experimental studyAI-enabled digital platformsAikyaMind mobile appDepression, anxiety, stressRappai S [62]2024IndiaQuantitative researchML modelCNN-LSTM modelSuicidal ideationGarcia D [63]2025SwedenComparative studyML modelMultiple linear regressions (NLP)Anxiety and depression comorbidityHaghish EF [64]2025NorwayCross-sectional studyML modelLogistic regression, HMDA, AutoEnsembleSuicidal and non-suicidal self-harmBoggiss A [65]2023New ZealandQualitative researchAI-powered toolCOMPASS (self-compassion chatbot)Psychological distress (Type 1 diabetes)Ibrahim RH [66]2025IraqQualitative studyAI-enabled digital platformsmHealth app (CBT-based)Anxiety and depressionKim KW [67]2021KoreaRetrospective chart reviewML modelLR, RF, ANN, SVM, XGBSuicide riskKuhlmeier FO [68]2025GermanyQualitative studyAI-powered toolCady (chatbot prototype)DepressionMatheson EL [69]2023BrazilRCTAI-enabled digital platformsTopity (chatbot platform)Body image and mental health Table 1Characteristics of the studies included in the reviewFirst AuthorYearCountryStudy DesignType of AISpecific AI Technology/ToolMental Health DiseaseBiernesser C [47]2023USAQualitative studyAI-enabled digital platformsFlourishSuicidal ideation (LGBTQ+ youth)Bogdanski E [48]2023USAExploratory case studyAI-powered toolTheraVR (VR system)PTSDHill ED [49]2025USALongitudinal cohort studyML modelSymptom-driven model (CBCL scales)Mental health riskNicol G [50]2022USAPilot RCTAI-powered toolW-GenZ appDepression and anxietyPaschali M [51]2022USALongitudinal studyML modelDLSuicidal ideation and depressionHuang Y [52]2022ChinaCross-sectional studyML modelRF, SVM, DTDepression and suicidal ideationJiang J [53]2025ChinaQuasi-experimental studyAI-powered toolGymBuddy, Elomia (chatbots)Psychological distressKuang Y [54]2025ChinaCross-sectional studyML modelFL, LR, RF, XGBoostDepressionZheng Y [29]2022ChinaSimulation experimentML modelLR, DLSocial anxietyBeaudequin D [55]2021AustraliaCohort studyML modelBNPsychological distressSu R [56]2023AustraliaLongitudinal studyML modelRFSelf-harm and suicide attemptsFujita J [57]2025JapanRCTAI-powered toolemol (chatbot)DepressionGuo Y [58]2024JapanRCTAI-enabled digital platformsAgent technology family education systemComprehensiveJohnson E [59]2020SpainQuantitative researchAI-powered toolMAmIoTie (affective cognitive toy)Mood disordersMartín-Ávila J [60]2025SpainRCTAI-enabled digital platformsemoTICareDepression and anxiety (Type 1 diabetes)Prakash GH [61]2025IndiaQuasi-experimental studyAI-enabled digital platformsAikyaMind mobile appDepression, anxiety, stressRappai S [62]2024IndiaQuantitative researchML modelCNN-LSTM modelSuicidal ideationGarcia D [63]2025SwedenComparative studyML modelMultiple linear regressions (NLP)Anxiety and depression comorbidityHaghish EF [64]2025NorwayCross-sectional studyML modelLogistic regression, HMDA, AutoEnsembleSuicidal and non-suicidal self-harmBoggiss A [65]2023New ZealandQualitative researchAI-powered toolCOMPASS (self-compassion chatbot)Psychological distress (Type 1 diabetes)Ibrahim RH [66]2025IraqQualitative studyAI-enabled digital platformsmHealth app (CBT-based)Anxiety and depressionKim KW [67]2021KoreaRetrospective chart reviewML modelLR, RF, ANN, SVM, XGBSuicide riskKuhlmeier FO [68]2025GermanyQualitative studyAI-powered toolCady (chatbot prototype)DepressionMatheson EL [69]2023BrazilRCTAI-enabled digital platformsTopity (chatbot platform)Body image and mental health Characteristics of the studies included in the review Characteristics of the studies included in the review The vast majority of research comes from the United States (N = 5) and China (N = 4), Other studies include Australia (N = 2), Japan (N = 2), Spain (N = 2), India (N = 2), Sweden (N = 1), Norway (N = 1) New Zealand (N = 1) Iraq (N = 1), Korea (N = 1) Germany (N = 1), Brazil (N = 1). All 24 included studies were published between 2020 and 2025, with no publications meeting the criteria from 2015 to 2019. The distribution by year is as follows: 2020 (4.2%), 2021 (8.3%), 2022 (12.5%), 2023 (29.2%), 2024 (4.2%), 2025 (41.6%). In terms of research design, the included studies primarily used the following methodologies: randomized controlled trials (20.8%), qualitative studies (16.7%), cross-sectional studies (12.5%), quantitative researches (8.3%), quasi-experimental studies (8.3%), cohort studies (8.3%), longitudinal studies (8.3%), comparative study (4.2%), exploratory case study (4.2%), retrospective chart review (4.2%), and simulation experiment (4.2%). According to the survey conducted in this scoping review, adolescents primarily face mental health issues such as depression (31.2%), anxiety (25%), suicidal ideation (18.7%), stress (6.3%), distress (6.3%), and other issues (12.5%) (the proportions of adolescent mental health diseases are shown in Fig. 2). Fig. 2Proportions of adolescent mental health diseases Proportions of adolescent mental health diseases Classification of AI applied to adolescent mental health Before presenting the findings, it is essential to clarify the conceptual distinctions among three categories of AI applications in adolescent mental health. Machine learning (ML) models refer to statistical or computational algorithms that learn patterns from data to perform predictions or classifications. These models are typically developed and evaluated in research settings and are not directly accessible to adolescent users. Examples from the included studies include random forest (RF) models for suicide risk prediction, logistic regression models for depression classification, and deep learning models (e.g., CNN–LSTM) for analyzing suicidal ideation. These models are generally evaluated using performance metrics such as accuracy (ACC), sensitivity, and area under the curve (AUC). Their primary function is risk prediction or detection rather than direct intervention and they are backend research algorithms with no direct access or interaction for adolescent users.AI-powered tools refer to user-facing systems that apply AI techniques to deliver mental health support or therapeutic assistance directly to adolescents. These systems typically integrate AI algorithms with interactive interfaces such as chatbots, mobile applications, or virtual environments, enabling real-time interaction with users. Examples include Woebot, Elomia, Cady, and COMPASS. These tools are commonly evaluated in terms of usability, acceptability, engagement, and potential clinical outcomes and they are standalone, user-interactive intervention tools, not comprehensive multi-service platforms.AI-enabled digital platforms refer to comprehensive software systems that incorporate AI components alongside other digital health features, such as psychoeducation modules, self-assessment tools, symptom monitoring, and professional counseling services. These platforms often represent multi-component interventions implemented in real-world settings and they integrate AI as a component within a full mental health service ecosystem, distinct from single-function AI tools. Examples include AikyaMind, emoTICare, and Flourish, which integrate AI-based functions with broader mental health support services. Machine learning (ML) models refer to statistical or computational algorithms that learn patterns from data to perform predictions or classifications. These models are typically developed and evaluated in research settings and are not directly accessible to adolescent users. Examples from the included studies include random forest (RF) models for suicide risk prediction, logistic regression models for depression classification, and deep learning models (e.g., CNN–LSTM) for analyzing suicidal ideation. These models are generally evaluated using performance metrics such as accuracy (ACC), sensitivity, and area under the curve (AUC). Their primary function is risk prediction or detection rather than direct intervention and they are backend research algorithms with no direct access or interaction for adolescent users. Machine learning (ML) models refer to statistical or computational algorithms that learn patterns from data to perform predictions or classifications. These models are typically developed and evaluated in research settings and are not directly accessible to adolescent users. Examples from the included studies include random forest (RF) models for suicide risk prediction, logistic regression models for depression classification, and deep learning models (e.g., CNN–LSTM) for analyzing suicidal ideation. These models are generally evaluated using performance metrics such as accuracy (ACC), sensitivity, and area under the curve (AUC). Their primary function is risk prediction or detection rather than direct intervention and they are backend research algorithms with no direct access or interaction for adolescent users. AI-powered tools refer to user-facing systems that apply AI techniques to deliver mental health support or therapeutic assistance directly to adolescents. These systems typically integrate AI algorithms with interactive interfaces such as chatbots, mobile applications, or virtual environments, enabling real-time interaction with users. Examples include Woebot, Elomia, Cady, and COMPASS. These tools are commonly evaluated in terms of usability, acceptability, engagement, and potential clinical outcomes and they are standalone, user-interactive intervention tools, not comprehensive multi-service platforms. AI-powered tools refer to user-facing systems that apply AI techniques to deliver mental health support or therapeutic assistance directly to adolescents. These systems typically integrate AI algorithms with interactive interfaces such as chatbots, mobile applications, or virtual environments, enabling real-time interaction with users. Examples include Woebot, Elomia, Cady, and COMPASS. These tools are commonly evaluated in terms of usability, acceptability, engagement, and potential clinical outcomes and they are standalone, user-interactive intervention tools, not comprehensive multi-service platforms. AI-enabled digital platforms refer to comprehensive software systems that incorporate AI components alongside other digital health features, such as psychoeducation modules, self-assessment tools, symptom monitoring, and professional counseling services. These platforms often represent multi-component interventions implemented in real-world settings and they integrate AI as a component within a full mental health service ecosystem, distinct from single-function AI tools. Examples include AikyaMind, emoTICare, and Flourish, which integrate AI-based functions with broader mental health support services. AI-enabled digital platforms refer to comprehensive software systems that incorporate AI components alongside other digital health features, such as psychoeducation modules, self-assessment tools, symptom monitoring, and professional counseling services. These platforms often represent multi-component interventions implemented in real-world settings and they integrate AI as a component within a full mental health service ecosystem, distinct from single-function AI tools. Examples include AikyaMind, emoTICare, and Flourish, which integrate AI-based functions with broader mental health support services. Together, these three categories represent different levels of AI implementation in adolescent mental health research, ranging from algorithm-level predictive models to interactive tools and integrated digital intervention platforms. These AI-based approaches aim to provide scalable support and therapeutic resources for improving adolescent mental health outcomes. Research objectives of using AI tools to provide support and therapy for adolescent mental health Based on Supplementary Material S5, we summarized the 24 included studies and found that the main purposes of applying AI tools include the following: Developing ML models and AI tools to support and treat adolescent mental health. Through the validation of AI tools and comparison of models, the usability, acceptability, interpretability, and accuracy of AI tools or applications in adolescent mental health are identified [47, 53, 61].Using AI tools to feedback on and improve symptoms of mental health disorders in adolescents. Examples include using telemedicine-based VR technology to improve patient symptoms and engagement [48], accurately predicting mental health risks [29], and enhancing adolescents’ understanding of the risk impacts of experiencing negative symptoms [55, 60].Determining the optimal approach to integrating AI tools into adolescent mental health management. While AI technologies support the improvement of initial mental health disorder symptoms, they still need to be used cautiously for adolescents with severe mental health issues [57]. For example, the research findings by Nicol G et al. [50] initially support the feasibility of using CBT-based chatbot technology to supplement mental health treatment received by adolescents with mild to moderate depression in primary care settings. Ibrahim RH [66] found that the mobile CBT app shows potential for improving adolescent mental health, with initial improvements in mood and anxiety.Focusing on user experience and mental health outcomes, and evaluating adolescents’ attitudes, expectations, and design preferences regarding AI tools used for treating adolescent mental health. Studies have found that adolescents generally hold a positive attitude toward the use of AI tools, participants reported less anxiety when using chatbots than when seeing human therapists, and they believed chatbots could better protect patients’ privacy. Additionally, adolescents put forward more personalized preferences for AI tools, such as natural daily conversations and advice for emergency issues [68]. Developing ML models and AI tools to support and treat adolescent mental health. Through the validation of AI tools and comparison of models, the usability, acceptability, interpretability, and accuracy of AI tools or applications in adolescent mental health are identified [47, 53, 61]. Developing ML models and AI tools to support and treat adolescent mental health. Through the validation of AI tools and comparison of models, the usability, acceptability, interpretability, and accuracy of AI tools or applications in adolescent mental health are identified [47, 53, 61]. Using AI tools to feedback on and improve symptoms of mental health disorders in adolescents. Examples include using telemedicine-based VR technology to improve patient symptoms and engagement [48], accurately predicting mental health risks [29], and enhancing adolescents’ understanding of the risk impacts of experiencing negative symptoms [55, 60]. Using AI tools to feedback on and improve symptoms of mental health disorders in adolescents. Examples include using telemedicine-based VR technology to improve patient symptoms and engagement [48], accurately predicting mental health risks [29], and enhancing adolescents’ understanding of the risk impacts of experiencing negative symptoms [55, 60]. Determining the optimal approach to integrating AI tools into adolescent mental health management. While AI technologies support the improvement of initial mental health disorder symptoms, they still need to be used cautiously for adolescents with severe mental health issues [57]. For example, the research findings by Nicol G et al. [50] initially support the feasibility of using CBT-based chatbot technology to supplement mental health treatment received by adolescents with mild to moderate depression in primary care settings. Ibrahim RH [66] found that the mobile CBT app shows potential for improving adolescent mental health, with initial improvements in mood and anxiety. Determining the optimal approach to integrating AI tools into adolescent mental health management. While AI technologies support the improvement of initial mental health disorder symptoms, they still need to be used cautiously for adolescents with severe mental health issues [57]. For example, the research findings by Nicol G et al. [50] initially support the feasibility of using CBT-based chatbot technology to supplement mental health treatment received by adolescents with mild to moderate depression in primary care settings. Ibrahim RH [66] found that the mobile CBT app shows potential for improving adolescent mental health, with initial improvements in mood and anxiety. Focusing on user experience and mental health outcomes, and evaluating adolescents’ attitudes, expectations, and design preferences regarding AI tools used for treating adolescent mental health. Studies have found that adolescents generally hold a positive attitude toward the use of AI tools, participants reported less anxiety when using chatbots than when seeing human therapists, and they believed chatbots could better protect patients’ privacy. Additionally, adolescents put forward more personalized preferences for AI tools, such as natural daily conversations and advice for emergency issues [68]. Focusing on user experience and mental health outcomes, and evaluating adolescents’ attitudes, expectations, and design preferences regarding AI tools used for treating adolescent mental health. Studies have found that adolescents generally hold a positive attitude toward the use of AI tools, participants reported less anxiety when using chatbots than when seeing human therapists, and they believed chatbots could better protect patients’ privacy. Additionally, adolescents put forward more personalized preferences for AI tools, such as natural daily conversations and advice for emergency issues [68]. The outcomes of AI application in adolescent mental health support Drawing on the analysis and synthesis of the 24 studies included in this review, the key findings regarding AI-driven support and treatment for adolescent mental health are presented below(See Supplementary Material S6 for specific details): Machine learning (ML) models have demonstrated high accuracy in predicting identifying and classifying adolescent mental health conditions, enabling large-scale screening, providing clinical decision support, facilitating timely targeted interventions, and potentially reducing the risk of suicide and depression among adolescents. 11 studies support the effectiveness of AI in predicting, identifying, and classifying, such as Huang et al. [52] compared three ML models—random forest (RF), support vector machine (SVM), and decision tree (DT)—and found that the RF model achieved higher accuracy in predicting suicidal ideation (accuracy[ACC] = 87.3%, SD = 3.2%; area under the curve [AUC] = 92.4%) and depressive symptoms (ACC = 84.0%, SD = 2.8%; AUC = 90.1%). Kuang et al. [54] demonstrated that federated learning (FL) can enhance the accuracy of locally trained ML models, achieving a maximum increase of 20% (AUC from 0.7544 to 0.9064). Furthermore, FL models can provide clinicians with valuable insights to help identify and support adolescents at risk for depression. In mental health risk classification, Haghish [64] demonstrated that holistic multimodel domain analysis (HMDA) can effectively distinguish suicidal self-harm (SSH) from non-suicidal self-harm (NSSH) without relying on intention or suicide-related information.AI-based mental health interventions can improve or alleviate adolescent mental health issues. Multiple RCTs demonstrated that AI-based interventions can alleviate adolescent mental health symptoms. Nicol et al. [50] reported that adolescents using a CBT-based chatbot showed greater PHQ-9(Patient Health Questionnaire-9) reduction compared to waitlist controls (3.3 vs. 2.0 points). Guo [58] found that agent technology-based family education significantly improved multiple mental health domains compared to traditional education (p < 0.05). Matheson et al. [69] reported small but significant improvements in body satisfaction and positive affect among adolescents using the Topity chatbot. However, concerns were identified regarding applicability for severe cases and long-term effectiveness. Although Jiang et al. [53] reported sustained mental health improvements following AI intervention, three other studies [66, 68, 69] found that the effects of AI-driven tools and digital application platforms were immediate but short-term. Finally, Fujita et al. [57] found that AI interventions may increase psychological burden in adolescents with severe depression.The application of AI improves the clinical management of adolescent mental health, including compliance, participation, and acceptability. Five studies mentioned that AI tools improved participants’ engagement, satisfaction, and acceptance, such as Biernesser et al. [47] reporting that 19 of 20 LGBTQ+ youth preferred fully or partially automated interventions over non-automated alternatives. Bogdanski et al. [48] found that head-mounted display-based VR with avatar and behavioral activation significantly enhanced therapeutic alliance and engagement compared to standard telehealth. However, three studies mentioned difficulties with low participation. In one study [57], all three patients who ultimately agreed to participate either withdrew or were excluded, mainly due to increased psychological burden among severely mentally ill patients, concerns about diary recording requirements, and medication initiation. In another study [68], two participants (14%) expressed concerns about low engagement caused by depressive symptoms such as lack of motivation or forgetfulness, as well as a lack of social pressure during use, which hindered effective engagement. The third study [69] was primarily affected by COVID-19, resulting in a high intervention attrition rate (531/858, 61.9%). Furthermore, two studies [57, 61] reported AI tools support adolescent self-management, as demonstrated by the emol chatbot’s mood logging and sleep tracking features, which enable users to actively monitor their mental health. AI tools can also facilitate collaboration among clinicians, guardians, and adolescents, for instance, when AI detects suicidal ideation, alerts can notify healthcare providers for clinical management [50].AI-powered remote technologies and digital applications have expanded access to mental health services for underserved populations, low- and middle-income countries, emergency situations, and adolescents with special medical conditions. Among the included studies, several AI tools supported specific adolescent populations: Flourish [47] provided suicide prevention interventions for LGBTQ+ youth who had experienced online victimization; virtual reality telemedicine [48] helped alleviate symptoms in patients with post-traumatic stress disorder (PTSD); and emoTICare [60] and COMPASS [65] improved psychological distress in adolescents with Type 1 diabetes mellitus. AI also expands opportunities in low- and middle-income countries [69], resource-scarce areas, and war-affected or crisis-stricken regions. In Iraq, for instance, resource scarcity, post-war instability, socio-economic challenges, and an overstretched healthcare system have restricted young people’s access to traditional treatments. AI-powered services, however, have offered new avenues for improving youth mental health [66]. Additionally, chatbots demonstrated particular utility during the COVID-19 pandemic [50, 65], providing online counseling and companionship to adolescents experiencing social isolation, thereby helping to alleviate symptoms of anxiety and depression. Machine learning (ML) models have demonstrated high accuracy in predicting identifying and classifying adolescent mental health conditions, enabling large-scale screening, providing clinical decision support, facilitating timely targeted interventions, and potentially reducing the risk of suicide and depression among adolescents. 11 studies support the effectiveness of AI in predicting, identifying, and classifying, such as Huang et al. [52] compared three ML models—random forest (RF), support vector machine (SVM), and decision tree (DT)—and found that the RF model achieved higher accuracy in predicting suicidal ideation (accuracy[ACC] = 87.3%, SD = 3.2%; area under the curve [AUC] = 92.4%) and depressive symptoms (ACC = 84.0%, SD = 2.8%; AUC = 90.1%). Kuang et al. [54] demonstrated that federated learning (FL) can enhance the accuracy of locally trained ML models, achieving a maximum increase of 20% (AUC from 0.7544 to 0.9064). Furthermore, FL models can provide clinicians with valuable insights to help identify and support adolescents at risk for depression. In mental health risk classification, Haghish [64] demonstrated that holistic multimodel domain analysis (HMDA) can effectively distinguish suicidal self-harm (SSH) from non-suicidal self-harm (NSSH) without relying on intention or suicide-related information. Machine learning (ML) models have demonstrated high accuracy in predicting identifying and classifying adolescent mental health conditions, enabling large-scale screening, providing clinical decision support, facilitating timely targeted interventions, and potentially reducing the risk of suicide and depression among adolescents. 11 studies support the effectiveness of AI in predicting, identifying, and classifying, such as Huang et al. [52] compared three ML models—random forest (RF), support vector machine (SVM), and decision tree (DT)—and found that the RF model achieved higher accuracy in predicting suicidal ideation (accuracy[ACC] = 87.3%, SD = 3.2%; area under the curve [AUC] = 92.4%) and depressive symptoms (ACC = 84.0%, SD = 2.8%; AUC = 90.1%). Kuang et al. [54] demonstrated that federated learning (FL) can enhance the accuracy of locally trained ML models, achieving a maximum increase of 20% (AUC from 0.7544 to 0.9064). Furthermore, FL models can provide clinicians with valuable insights to help identify and support adolescents at risk for depression. In mental health risk classification, Haghish [64] demonstrated that holistic multimodel domain analysis (HMDA) can effectively distinguish suicidal self-harm (SSH) from non-suicidal self-harm (NSSH) without relying on intention or suicide-related information. AI-based mental health interventions can improve or alleviate adolescent mental health issues. Multiple RCTs demonstrated that AI-based interventions can alleviate adolescent mental health symptoms. Nicol et al. [50] reported that adolescents using a CBT-based chatbot showed greater PHQ-9(Patient Health Questionnaire-9) reduction compared to waitlist controls (3.3 vs. 2.0 points). Guo [58] found that agent technology-based family education significantly improved multiple mental health domains compared to traditional education (p < 0.05). Matheson et al. [69] reported small but significant improvements in body satisfaction and positive affect among adolescents using the Topity chatbot. However, concerns were identified regarding applicability for severe cases and long-term effectiveness. Although Jiang et al. [53] reported sustained mental health improvements following AI intervention, three other studies [66, 68, 69] found that the effects of AI-driven tools and digital application platforms were immediate but short-term. Finally, Fujita et al. [57] found that AI interventions may increase psychological burden in adolescents with severe depression. AI-based mental health interventions can improve or alleviate adolescent mental health issues. Multiple RCTs demonstrated that AI-based interventions can alleviate adolescent mental health symptoms. Nicol et al. [50] reported that adolescents using a CBT-based chatbot showed greater PHQ-9(Patient Health Questionnaire-9) reduction compared to waitlist controls (3.3 vs. 2.0 points). Guo [58] found that agent technology-based family education significantly improved multiple mental health domains compared to traditional education (p < 0.05). Matheson et al. [69] reported small but significant improvements in body satisfaction and positive affect among adolescents using the Topity chatbot. However, concerns were identified regarding applicability for severe cases and long-term effectiveness. Although Jiang et al. [53] reported sustained mental health improvements following AI intervention, three other studies [66, 68, 69] found that the effects of AI-driven tools and digital application platforms were immediate but short-term. Finally, Fujita et al. [57] found that AI interventions may increase psychological burden in adolescents with severe depression. The application of AI improves the clinical management of adolescent mental health, including compliance, participation, and acceptability. Five studies mentioned that AI tools improved participants’ engagement, satisfaction, and acceptance, such as Biernesser et al. [47] reporting that 19 of 20 LGBTQ+ youth preferred fully or partially automated interventions over non-automated alternatives. Bogdanski et al. [48] found that head-mounted display-based VR with avatar and behavioral activation significantly enhanced therapeutic alliance and engagement compared to standard telehealth. However, three studies mentioned difficulties with low participation. In one study [57], all three patients who ultimately agreed to participate either withdrew or were excluded, mainly due to increased psychological burden among severely mentally ill patients, concerns about diary recording requirements, and medication initiation. In another study [68], two participants (14%) expressed concerns about low engagement caused by depressive symptoms such as lack of motivation or forgetfulness, as well as a lack of social pressure during use, which hindered effective engagement. The third study [69] was primarily affected by COVID-19, resulting in a high intervention attrition rate (531/858, 61.9%). Furthermore, two studies [57, 61] reported AI tools support adolescent self-management, as demonstrated by the emol chatbot’s mood logging and sleep tracking features, which enable users to actively monitor their mental health. AI tools can also facilitate collaboration among clinicians, guardians, and adolescents, for instance, when AI detects suicidal ideation, alerts can notify healthcare providers for clinical management [50]. The application of AI improves the clinical management of adolescent mental health, including compliance, participation, and acceptability. Five studies mentioned that AI tools improved participants’ engagement, satisfaction, and acceptance, such as Biernesser et al. [47] reporting that 19 of 20 LGBTQ+ youth preferred fully or partially automated interventions over non-automated alternatives. Bogdanski et al. [48] found that head-mounted display-based VR with avatar and behavioral activation significantly enhanced therapeutic alliance and engagement compared to standard telehealth. However, three studies mentioned difficulties with low participation. In one study [57], all three patients who ultimately agreed to participate either withdrew or were excluded, mainly due to increased psychological burden among severely mentally ill patients, concerns about diary recording requirements, and medication initiation. In another study [68], two participants (14%) expressed concerns about low engagement caused by depressive symptoms such as lack of motivation or forgetfulness, as well as a lack of social pressure during use, which hindered effective engagement. The third study [69] was primarily affected by COVID-19, resulting in a high intervention attrition rate (531/858, 61.9%). Furthermore, two studies [57, 61] reported AI tools support adolescent self-management, as demonstrated by the emol chatbot’s mood logging and sleep tracking features, which enable users to actively monitor their mental health. AI tools can also facilitate collaboration among clinicians, guardians, and adolescents, for instance, when AI detects suicidal ideation, alerts can notify healthcare providers for clinical management [50]. AI-powered remote technologies and digital applications have expanded access to mental health services for underserved populations, low- and middle-income countries, emergency situations, and adolescents with special medical conditions. Among the included studies, several AI tools supported specific adolescent populations: Flourish [47] provided suicide prevention interventions for LGBTQ+ youth who had experienced online victimization; virtual reality telemedicine [48] helped alleviate symptoms in patients with post-traumatic stress disorder (PTSD); and emoTICare [60] and COMPASS [65] improved psychological distress in adolescents with Type 1 diabetes mellitus. AI also expands opportunities in low- and middle-income countries [69], resource-scarce areas, and war-affected or crisis-stricken regions. In Iraq, for instance, resource scarcity, post-war instability, socio-economic challenges, and an overstretched healthcare system have restricted young people’s access to traditional treatments. AI-powered services, however, have offered new avenues for improving youth mental health [66]. Additionally, chatbots demonstrated particular utility during the COVID-19 pandemic [50, 65], providing online counseling and companionship to adolescents experiencing social isolation, thereby helping to alleviate symptoms of anxiety and depression. AI-powered remote technologies and digital applications have expanded access to mental health services for underserved populations, low- and middle-income countries, emergency situations, and adolescents with special medical conditions. Among the included studies, several AI tools supported specific adolescent populations: Flourish [47] provided suicide prevention interventions for LGBTQ+ youth who had experienced online victimization; virtual reality telemedicine [48] helped alleviate symptoms in patients with post-traumatic stress disorder (PTSD); and emoTICare [60] and COMPASS [65] improved psychological distress in adolescents with Type 1 diabetes mellitus. AI also expands opportunities in low- and middle-income countries [69], resource-scarce areas, and war-affected or crisis-stricken regions. In Iraq, for instance, resource scarcity, post-war instability, socio-economic challenges, and an overstretched healthcare system have restricted young people’s access to traditional treatments. AI-powered services, however, have offered new avenues for improving youth mental health [66]. Additionally, chatbots demonstrated particular utility during the COVID-19 pandemic [50, 65], providing online counseling and companionship to adolescents experiencing social isolation, thereby helping to alleviate symptoms of anxiety and depression. Strengths and drawbacks of AI application in adolescent mental health This review noted AI generally benefits adolescent mental health, yet also entails drawbacks (see Table 2). Compared to traditional interventions, AI shows improved well-being, treatment acceptability, clinical efficacy, and cost-effectiveness. Table 2The strengths and drawbacks of AI in the application of adolescent mental healthItemStrengthsDrawbacksCategoryHigh accuracy[29, 49, 51–53, 67]Algorithmic incompleteness and high false positive rates[56]High efficiency[49, 52, 62, 66]Insights into the risk, improve risk assessment and protective factor[51, 55, 56, 64]Differences in measurement tools[63]Dynamic analysis and strong sensitivity[50, 55, 57]Usability and Personalization[53, 65, 66, 68]Limit the generalizability of the findings[47, 50, 51, 54, 55, 57, 60, 61, 63, 65, 69]High acceptability and satisfaction[48, 55, 61, 69]Reduce stigma[47, 54]Excessive intervention intensity[57]Promoting holistic well-being[47, 48, 54, 58, 63, 66]Restrict the participation of family members[60]Collaborate with professionals[55]The benefits of multidisciplinary teams have not yet been realized[55]Increase access to supportive care[29, 47, 69]Economic benefits[50, 54, 59, 60, 66, 69]The effect is not noticeable in the long term[58, 61, 66] Table 2The strengths and drawbacks of AI in the application of adolescent mental healthItemStrengthsDrawbacksCategoryHigh accuracy[29, 49, 51–53, 67]Algorithmic incompleteness and high false positive rates[56]High efficiency[49, 52, 62, 66]Insights into the risk, improve risk assessment and protective factor[51, 55, 56, 64]Differences in measurement tools[63]Dynamic analysis and strong sensitivity[50, 55, 57]Usability and Personalization[53, 65, 66, 68]Limit the generalizability of the findings[47, 50, 51, 54, 55, 57, 60, 61, 63, 65, 69]High acceptability and satisfaction[48, 55, 61, 69]Reduce stigma[47, 54]Excessive intervention intensity[57]Promoting holistic well-being[47, 48, 54, 58, 63, 66]Restrict the participation of family members[60]Collaborate with professionals[55]The benefits of multidisciplinary teams have not yet been realized[55]Increase access to supportive care[29, 47, 69]Economic benefits[50, 54, 59, 60, 66, 69]The effect is not noticeable in the long term[58, 61, 66] The strengths and drawbacks of AI in the application of adolescent mental health High accuracy [29, 49, 51–53, 67] Algorithmic incompleteness and high false positive rates [56] High efficiency [49, 52, 62, 66] Insights into the risk, improve risk assessment and protective factor [51, 55, 56, 64] Differences in measurement tools [63] Dynamic analysis and strong sensitivity [50, 55, 57] Usability and Personalization [53, 65, 66, 68] Limit the generalizability of the findings [47, 50, 51, 54, 55, 57, 60, 61, 63, 65, 69] High acceptability and satisfaction [48, 55, 61, 69] Reduce stigma [47, 54] Excessive intervention intensity [57] Promoting holistic well-being [47, 48, 54, 58, 63, 66] Restrict the participation of family members [60] Collaborate with professionals [55] The benefits of multidisciplinary teams have not yet been realized [55] Increase access to supportive care [29, 47, 69] Economic benefits [50, 54, 59, 60, 66, 69] The effect is not noticeable in the long term [58, 61, 66] Firstly, compared to traditional mental health support, AI technology offers greater efficiency [49, 52] and accuracy in predicting [29, 53], classifying, and evaluating risk factors for adolescent mental health [55]. The dynamic analysis capabilities of AI enable rapid data updates and multivariate scenario assessments [55], providing personalized mental health treatment plans [53, 65]. Secondly, the application of AI technology has significantly improved participation, satisfaction, and acceptability among adolescents [61, 69]. Participation is a key predictor of intervention effectiveness, and AI’s user-centered design better aligns with adolescents’ needs and preferences, enhancing engagement and satisfaction [61]. During identity exploration, adolescents are sensitive to external evaluation and increasingly value privacy. The anonymity provided by AI reduces mental health stigma, offering a non-judgmental alternative to in-person counseling and fostering greater trust and participation [47, 54]. Thirdly, AI helps address the lack of resources in certain areas or during emergencies, expanding access to mental health support for adolescents [29, 47, 69]. The widespread use of AI in mental health can enhance existing care systems in schools and healthcare services, offering broader support to marginalized and under-served populations [47]. With remote medical technology and AI-driven dialogue systems, AI improves coverage and accessibility, providing convenient channels for interaction with mental health professionals [48]. This ensures that adolescents’ mental health needs are met in emergency situations or low-resource settings like Iraq [66], preventing crises and minimizing the impact of a shortage or absence of professional staff. Fourthly, empowering adolescent mental health with AI can enhance the well-being of patients [63], schools, and society at large. For adolescents, AI can help alleviate anxiety, depression, and stress [29], provide a better understanding of their physical changes, and improve their ability to cope with mental health challenges [54, 63]. AI could augment existing care in schools and mental health services [47]. Moreover, AI-based mental health technologies offer economic benefits by reducing the need for short- and long-term care [60, 66], eliminating travel to therapy appointments, and lowering out-of-pocket expenses [50]. It can also promote collaboration among experts in the field of mental health [55]. The integration of AI will increase the accessibility, acceptability, and scalability of mental health resources, driving lasting, positive changes that contribute to overall social welfare [69]. While AI technology has a positive impact on adolescent mental health, it still faces numerous challenges and limitations. Firstly, the application of AI may increase additional risks. Algorithmic incompleteness and high false positive rates can affect adolescent mental health treatment [56]. Differences in AI measurement tools can also impact the identification of correlation factors, while uncertainties and data biases in interventions may worsen adolescents’ psychological health, introducing unnecessary experimental risks [63]. Secondly, the intensity of the AI intervention may have been overwhelming for participants. Fujita [57] has found that emol can trigger a significant psychological burden in adolescents with severe mental health challenges. AI intervention in adolescent mental health may restrict the participation of family members. The use of AI technology limits direct parental involvement, which may hinder their ability to provide necessary protection and support for adolescents [60]. Thirdly, the benefits of multidisciplinary teams brought by telemedicine have not yet been realized. While AI-driven remote healthcare enhances communication between patients and mental health experts, adolescent mental health requires collaboration across multiple disciplines, such as pediatrics and psychiatry. However, the full potential of multidisciplinary teams in integrating AI technology has not yet been realized [55]. Fourthly, long-term effectiveness is not significant. Most reviewed studies had short research periods and lacked long-term follow-up data [58]. A study reported no significant mental health improvements after a one-month follow-up, raising concerns about long-term effectiveness [61]. Although participants experienced short-term improvements in mood and anxiety, these benefits were not sustained without frequent app use [66]. Finally, the development and implementation of AI systems face challenges related to population, race, gender, and environmental factors. Small sample sizes limit the generalizability of findings and hinder widespread adoption, reducing applicability across diverse groups [50, 57, 60, 69]. Additionally, reliance on screen-based data prevents AI tools from capturing nonverbal cues, such as facial expressions and body language, limiting feedback richness and diagnostic sensitivity [65]. Technical and ethical issues While AI technology is becoming increasingly significant in adolescent mental health, it presents a dual nature. On one hand, it offers convenience and efficiency, but on the other, it raises several unavoidable technical and ethical issues. There are several issues related to the application of AI tools in adolescent mental health. Firstly, since the use of AI in this field is still in the exploratory phase, its full potential has not yet been realized [55], and there are no significant differences in the predictability of mechanism-driven models [29]. Secondly, there is a lack of feasibility validation, as some AI tools have not been practically implemented in real-world settings and lack support from actual data [50, 65]. Thirdly, for adolescents with severe mental health issues, using conversational AI tools may inadvertently increase their psychological burden [57]. Finally, other technical challenges, for instance, complex navigation design were commonly noted by participants, who often struggled to locate tools or navigate the app’s interface [66]. Users may also encounter slow or crashing interface loading [66]. In addition to the potential technological risks of AI, AI tools applied to adolescent mental health also bring many ethical risks, such as the complexity of informed consent, data security and confidentiality, possible bias issues, and transparency and interpretability.Informed consent. A key ethical issue discussed in the included literature is informed consent [48, 50, 55, 56, 60, 61, 63, 66, 69]. Since the research subjects are adolescents, many of whom are minors, obtaining consent from both guardians and the adolescents themselves becomes challenging [57], complicating the process of ensuring true informed consent.Data security and privacy protection. As AI increasingly functions as a therapist in adolescent mental health care, significant concerns arise regarding data security and privacy [64, 69]. The effectiveness of AI technology relies on large-scale ML, which requires vast amounts of data. This necessitates the transmission and storage of sensitive information, heightening the risk of privacy breaches or misuse of personal data [29]. On the other hand, the data and processes involved in AI therapy may be shared with parents or guardians to help them better support and care for their children [47]. However, this raises concerns about the privacy rights of adolescents, potentially eroding their trust and willingness to cooperate with the treatment.Safety and bias issues. Given the incomplete state of rational development during adolescence, over-reliance on AI for mental health treatment may compromise adolescents’ ability to differentiate between real and virtual environments. This over-dependence may also diminish their self-efficacy in proactively seeking help from appropriate sources [50]. Such reliance could potentially result in AI-induced self-harm or suicide thereby introducing additional safety risks. Furthermore, the application of AI in adolescent mental health treatment frequently depends on self-reported measures, which are vulnerable to recall bias, social desirability bias, and underreporting. These data inaccuracies may ultimately hinder therapeutic effectiveness [60, 67].Transparency and accountability. AI technology itself carries potential risks related to transparency, interpretability, and accountability [58, 60]. When AI tools are deployed to support adolescent mental health, questions arise regarding who bears responsibility in the event of adverse outcomes—such as missed suicide risk, inappropriate therapeutic advice, or algorithmic failure. In traditional therapeutic settings, responsibility is clearly assigned to licensed clinicians. However, when AI operates autonomously or as a black-box system, accountability becomes fragmented among developers, healthcare institutions, and clinicians who may rely on AI outputs without fully understanding their limitations. This ambiguity is particularly concerning for adolescent users, whose capacity to recognize and respond to AI errors is limited. As noted by Opel et al. [3], the use of AI as a mental health therapist for adolescents raises fundamental questions about whether machines can genuinely assume professional responsibility, especially when empathy, judgment, and legal accountability are inherently human attributes. Without clear frameworks for assigning liability, the integration of AI into adolescent mental health care risks creating a responsibility gap that may undermine patient safety and erode trust. Informed consent. A key ethical issue discussed in the included literature is informed consent [48, 50, 55, 56, 60, 61, 63, 66, 69]. Since the research subjects are adolescents, many of whom are minors, obtaining consent from both guardians and the adolescents themselves becomes challenging [57], complicating the process of ensuring true informed consent. Informed consent. A key ethical issue discussed in the included literature is informed consent [48, 50, 55, 56, 60, 61, 63, 66, 69]. Since the research subjects are adolescents, many of whom are minors, obtaining consent from both guardians and the adolescents themselves becomes challenging [57], complicating the process of ensuring true informed consent. Data security and privacy protection. As AI increasingly functions as a therapist in adolescent mental health care, significant concerns arise regarding data security and privacy [64, 69]. The effectiveness of AI technology relies on large-scale ML, which requires vast amounts of data. This necessitates the transmission and storage of sensitive information, heightening the risk of privacy breaches or misuse of personal data [29]. On the other hand, the data and processes involved in AI therapy may be shared with parents or guardians to help them better support and care for their children [47]. However, this raises concerns about the privacy rights of adolescents, potentially eroding their trust and willingness to cooperate with the treatment. Data security and privacy protection. As AI increasingly functions as a therapist in adolescent mental health care, significant concerns arise regarding data security and privacy [64, 69]. The effectiveness of AI technology relies on large-scale ML, which requires vast amounts of data. This necessitates the transmission and storage of sensitive information, heightening the risk of privacy breaches or misuse of personal data [29]. On the other hand, the data and processes involved in AI therapy may be shared with parents or guardians to help them better support and care for their children [47]. However, this raises concerns about the privacy rights of adolescents, potentially eroding their trust and willingness to cooperate with the treatment. Safety and bias issues. Given the incomplete state of rational development during adolescence, over-reliance on AI for mental health treatment may compromise adolescents’ ability to differentiate between real and virtual environments. This over-dependence may also diminish their self-efficacy in proactively seeking help from appropriate sources [50]. Such reliance could potentially result in AI-induced self-harm or suicide thereby introducing additional safety risks. Furthermore, the application of AI in adolescent mental health treatment frequently depends on self-reported measures, which are vulnerable to recall bias, social desirability bias, and underreporting. These data inaccuracies may ultimately hinder therapeutic effectiveness [60, 67]. Safety and bias issues. Given the incomplete state of rational development during adolescence, over-reliance on AI for mental health treatment may compromise adolescents’ ability to differentiate between real and virtual environments. This over-dependence may also diminish their self-efficacy in proactively seeking help from appropriate sources [50]. Such reliance could potentially result in AI-induced self-harm or suicide thereby introducing additional safety risks. Furthermore, the application of AI in adolescent mental health treatment frequently depends on self-reported measures, which are vulnerable to recall bias, social desirability bias, and underreporting. These data inaccuracies may ultimately hinder therapeutic effectiveness [60, 67]. Transparency and accountability. AI technology itself carries potential risks related to transparency, interpretability, and accountability [58, 60]. When AI tools are deployed to support adolescent mental health, questions arise regarding who bears responsibility in the event of adverse outcomes—such as missed suicide risk, inappropriate therapeutic advice, or algorithmic failure. In traditional therapeutic settings, responsibility is clearly assigned to licensed clinicians. However, when AI operates autonomously or as a black-box system, accountability becomes fragmented among developers, healthcare institutions, and clinicians who may rely on AI outputs without fully understanding their limitations. This ambiguity is particularly concerning for adolescent users, whose capacity to recognize and respond to AI errors is limited. As noted by Opel et al. [3], the use of AI as a mental health therapist for adolescents raises fundamental questions about whether machines can genuinely assume professional responsibility, especially when empathy, judgment, and legal accountability are inherently human attributes. Without clear frameworks for assigning liability, the integration of AI into adolescent mental health care risks creating a responsibility gap that may undermine patient safety and erode trust. Transparency and accountability. AI technology itself carries potential risks related to transparency, interpretability, and accountability [58, 60]. When AI tools are deployed to support adolescent mental health, questions arise regarding who bears responsibility in the event of adverse outcomes—such as missed suicide risk, inappropriate therapeutic advice, or algorithmic failure. In traditional therapeutic settings, responsibility is clearly assigned to licensed clinicians. However, when AI operates autonomously or as a black-box system, accountability becomes fragmented among developers, healthcare institutions, and clinicians who may rely on AI outputs without fully understanding their limitations. This ambiguity is particularly concerning for adolescent users, whose capacity to recognize and respond to AI errors is limited. As noted by Opel et al. [3], the use of AI as a mental health therapist for adolescents raises fundamental questions about whether machines can genuinely assume professional responsibility, especially when empathy, judgment, and legal accountability are inherently human attributes. Without clear frameworks for assigning liability, the integration of AI into adolescent mental health care risks creating a responsibility gap that may undermine patient safety and erode trust. Discussion With the rapid advancement of AI technology, it has increasingly been integrated into mental health intervention, support, and treatment, garnering growing support from the public [70–72]. Adolescence, a critical developmental period involving physical, emotional, and social changes, often brings overlooked mental health challenges. AI offers an accessible and cost-effective solution that transcends geographic and temporal barriers, providing timely and innovative support essential for adolescent well-being. AI interventions may offer a more accessible entry point to mental health services for adolescents who avoid traditional care, helping to fill the treatment gap during critical developmental windows [57]. This review highlights the growing application of AI in psychological support and therapy, driven by adolescents’ increased awareness of privacy and security, as well as their demand for personalized mental health care. AI is primarily applied to adolescent mental health support in the areas of mental health screening and early identification, risk prediction, individualized therapy, mental health intervention, telemedicine, and continuous monitoring. For instance, youth-focused chatbots can detect suicidal cues in language, enabling early risk identification and timely intervention [73, 74]. Adolescents are increasingly seeking AI-driven services that cater to their specific psychological needs. The review underscores the positive impact of AI-based interventions in alleviating symptoms of depression, anxiety, and stress in adolescents. This is especially evident in impoverished and war-torn regions, during emergency situations like epidemics, and among adolescents with chronic mental health conditions. Particularly in low- and middle-income countries, where mental health services and professionals are scarce, AI has the potential to aid in diagnosis, symptom tracking, disease prediction, and treatment interventions [31]. Despite its considerable potential, significant gaps remain in the literature regarding the effectiveness of AI in addressing common adolescent mental health issues. As highlighted in a recent systematic review [75] on doctoral student mental health several research gaps are equally pertinent to adolescent populations, the lack of cross-cultural comparisons, and insufficient attention to vulnerable subgroups (e.g., LGBTQ+ youth, those with chronic illnesses), and a scarcity of large-scale, long-term evaluations of intervention effectiveness. Evidence of long-term efficacy is limited, and AI tools may be unsuitable for adolescents with severe or complex mental health conditions [57]. In addition, many young patients, especially those who have not received a formal diagnosis or treatment, may have difficulty understanding the relevance of application-based mental health support and may not fully recognize its therapeutic value, particularly when compared to the immediate effects of face-to-face interventions such as medication or in-person therapy [57]. AI technology also raises concerns related to patient legal protections, potential harm, and the erosion of professional standards, which may undermine trust in settings where human judgment is essential. At present, no universally accepted framework exists for AI-based adolescent mental health treatment. Consequently, barriers to implementation are not only technological but also rooted in governance challenges, including ethics, regulation, and oversight [38]. At the technical level, to improve acceptability and encourage long-term voluntary use, AI development should follow a user-centered, iterative approach that actively involves adolescents and their parents from initial design through ongoing refinement. Developers and designers also need to enhance technological stability to support long-term engagement. Features such as regular reminders, progress tracking, and long-term reinforcement of positive behaviors may play a key role in maintaining user engagement. In addition, individuals with more severe symptoms may require more personalized support to enhance their participation and compliance with digital interventions [57]. Stakeholder engagement is critical for increasing adoption, maximizing intervention effectiveness, and ensuring cost-effectiveness [76, 77]. For safe and effective integration, psychological and clinical professionals must apply AI tools strictly within their professional scope. They should guide adolescents and parents to use these tools in a scientific, legal, and rational manner, while preventing overreliance or addiction. Although AI systems demonstrate strong computational and analytical capabilities, responsibility for treatment ultimately rests with human therapists. For ethical, professional, and legal reasons, essential interpersonal elements of therapy—such as empathy, communication, and human connection—can not be replaced by machines [24]. Limitations and suggestions for future research This scoping review has several limitations. Firstly, no formal quality assessment of the included studies was conducted, which may introduce bias in the interpretation of findings. Secondly, the search was restricted to English-language publications, which may have omitted relevant studies published in other languages and potentially biased the evidence synthesis. Thirdly, our use of the WHO definition of “adolescents” (ages 10–19) may not align with the varying age ranges used in other studies, potentially affecting the comprehensiveness of the included literature. Fourthly, the review did not fully explore how psychological factors and demographic differences—such as regional, racial, and socioeconomic backgrounds—influence the applicability of AI tools, which may limit the generalizability of our findings. Finally, although our search strategy covered the period from 2015 to 2025, no eligible studies were identified between 2015 and 2019. This gap likely reflects the nascent stage of AI applications in adolescent mental health during those early years. While foundational technologies such as rule-based conversational agents (e.g., ELIZA) and early chatbot prototypes (e.g., Teenchat) emerged during this period, the development and empirical testing of generative AI and conversational agents specifically tailored for adolescent mental health remained limited. It was not until approximately 2020, coinciding with advances in NLP and the growing recognition of digital mental health solutions, that research in this area began to gain momentum. This temporal pattern underscores the relatively recent emergence of robust evidence in this field and is consistent with the broader trajectory of AI in mental health research. Beyond the methodological limitations discussed above, a technical constraint of current AI tools warrants consideration. A notable limitation is the predominant reliance on screen-based data (e.g., typed text, self-report questionnaires), which excludes nonverbal communication cues such as facial expressions, vocal tone, and body language that are fundamental to clinical assessment. This absence may limit diagnostic sensitivity by preventing detection of subtle distress signals observable to trained clinicians and reducing prediction accuracy for conditions like depression and anxiety. Furthermore, the lack of bidirectional multimodal interaction may compromise therapeutic engagement and perceived empathy, which are particularly important for adolescents. These constraints suggest that while screen-based AI tools may be suitable for screening and low-intensity support, their diagnostic and therapeutic capabilities remain limited compared to human-delivered care, especially for adolescents with complex mental health conditions. Future research should prioritize adolescents by expanding studies in underrepresented regions and validating the real-world effectiveness of AI to strengthen the evidence base. Involving adolescents and parents in the design and validation of AI technologies may enhance user-friendliness and long-term engagement. At the technical level, future AI development should focus on multimodal data integration (e.g., speech analysis and facial expression recognition) to enhance diagnostic sensitivity and therapeutic responsiveness. Objective indicators, such as physiological markers or activity trackers, may also be incorporated in future research to enable more comprehensive and accurate mental health assessments [53]. Concurrently, because the target of treatment is adolescents, AI design can adjust its tone and communicate in a way that is neither indifferent nor overly informal, positioning AI as a trustworthy and supportive adult figure. On the other hand, AI tools can integrate various types of content to make interactions more dynamic, such as infographics, animations, images, inspiring quotes, and video- or audio-based content links [47]. Finally, robust supervision, safety protocols, and governance frameworks for digital mental health tools require urgent development to ensure ethical use. Investing in training for both professionals and users is crucial to ensure the safe, transparent, and clinically integrated application of AI in mental health care. Conclusion This scoping review synthesized evidence from 24 studies published between 2020 and 2025 on AI applications in adolescent mental health. The findings indicate that AI interventions, including machine learning based risk prediction models, AI powered mental health tools, and AI-enabled digital therapeutic platforms, show promising benefits for adolescents with emotional disorders such as depression and anxiety. These tools demonstrate potential advantages in risk prediction accuracy, personalized treatment delivery, and user engagement, with high acceptability among adolescent users who value the anonymity and accessibility of AI mediated support. Furthermore, compared to traditional face-to-face psychological interventions, AI based tools may enhance adolescents’ self management skills, reduce stigma and discrimination, and expand access to mental health resources for underserved populations, conflict affected regions, and low and middle income countries. However, given the limited number of studies included in this review, the findings should be interpreted and applied with appropriate caution. The evidence base for AI in adolescent mental health remains preliminary and characterized by important limitations. Three studies raised concerns about the short-term effects and sustainability of AI-driven mental health interventions, reporting that 13 out of 40 participants (32%) mentioned that the improvements were short-term and did not last without regular app usage. Furthermore, although five studies reported that AI applications improved user engagement, satisfaction, and acceptability, three other studies found that factors such as the complexity of the research design, concerns about privacy and security, the severity of mental health disorders, and the impact of COVID-19 contributed to low participation and high attrition rates among adolescents. Additionally, the predominance of screen-based data collection limits the diagnostic sensitivity of current AI tools, as they cannot capture nonverbal cues essential to clinical assessment and therapeutic alliance. The absence of rigorous clinical validation in real-world settings further limits the confidence with which these tools can be recommended for widespread use, particularly for adolescents with complex or severe mental health conditions. Notably, the deployment of AI in adolescent mental health faces substantial ethical challenges that must be addressed before these tools can be considered a reliable component of routine care. These include unresolved concerns regarding safety, data security and privacy, the complexity of obtaining informed consent from minors, potential algorithmic bias, attribution of responsibility, and the lack of standardized regulatory frameworks. In summary, while AI holds considerable promise as a complement to existing mental health services—particularly in expanding screening and adolescent self-management—its future role in adolescent mental health care will depend on addressing persistent technical, ethical, and legal challenges. Researchers need to conduct continued and in-depth studies to establish long-term efficacy through well-designed longitudinal and comparative research, validate AI tools in diverse real-world settings, and develop robust governance frameworks to ensure safety, transparency, and accountability. To improve adolescents’ adherence to and engagement with AI-supported treatment, it is essential to involve adolescents, parents, and clinicians in the co-design and evaluation of AI tools. This is crucial for ensuring that the technology meets the needs of end users while upholding ethical standards. Until these evidence and safeguards are in place, AI should be regarded as a complementary tool rather than a replacement for human-delivered care, and its adoption must be accompanied by careful attention to and scrutiny of potential limitations and risks. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6