A systematic review published on 11 February 2026 examines how artificial intelligence, machine learning and mental health chatbots are reshaping access to care, while identifying critical gaps in clinical evidence, equity, privacy and governance.
A new systematic review published in DIGITAL HEALTH brings together evidence from 37 studies to examine the growing role of artificial intelligence (AI) in mental healthcare, highlighting opportunities to improve accessibility, personalisation and service delivery through interconnected healthcare networks.
The research, titled Network-based artificial intelligence in mental healthcare: A systematic review of chatbots, artificial intelligence/machine learning models and ethical considerations in global healthcare networks, was conducted by Zahra Rezaei and colleagues and published online on 11 February 2026.
Drawing on research published between 2017 and 2024, the review explores three interconnected areas: AI-powered chatbots for mental health support, AI and machine learning (ML) models for assessment and diagnosis, and the ethical and privacy considerations essential to responsible implementation.
The findings offer important insights for policymakers, healthcare providers, researchers and digital mental health innovators seeking to integrate AI into sustainable, equitable and person-centred mental health systems.
AI-Powered Chatbots: Expanding Access to Mental Health Support
The review identifies growing evidence that AI-powered chatbots can improve access to mental health support, particularly for people facing geographical, financial or social barriers to traditional services.
Conversational tools such as Woebot, Wysa and Tess have demonstrated promising results in reducing symptoms of depression, anxiety and psychological distress. Drawing on approaches including cognitive behavioural therapy (CBT), mindfulness and emotional support, these technologies can provide accessible, on-demand assistance across a range of settings.
The researchers highlight findings from previous meta-analyses reporting improvements in depressive symptoms (Hedges’ g = 0.39) and psychological distress (Hedges’ g = 0.33), although the quality of evidence remains limited by the availability of robust randomised controlled trials.
Chatbot interventions also demonstrated potential during the COVID-19 pandemic, supporting groups including healthcare workers, students and older adults experiencing heightened psychological stress.
Beyond clinical outcomes, the review identifies user engagement as an important factor in effectiveness. Personalised interactions, accessible interfaces, empathetic communication and culturally appropriate design can influence whether individuals continue using digital mental health tools.
However, the authors emphasise that much of the existing evidence relates to short-term outcomes. Larger, longer-term studies are needed to establish whether these improvements are sustained and how chatbot interventions can be effectively integrated into ongoing care.
Machine Learning and Mental Health Assessment: Promising Results, Important Limitations
The review also examines advances in AI and ML models designed to identify patterns associated with mental health conditions, support early detection and assist clinical decision-making.
Specialised natural language processing models, including MentalBERT and MentalRoBERTa, achieved F1 scores ranging from 68% to 93% across mental health classification tasks. Another model, SR-BERT, demonstrated an F2 score of 76.2% and an ROC-AUC of 92% in suicide risk prediction research.
These findings demonstrate the technical capabilities of emerging AI models to analyse language and identify indicators of psychological distress.
However, the researchers caution that strong performance in computational testing does not necessarily translate into clinical reliability.
Many models have been trained using social media data from platforms such as Reddit and Twitter. The language, behaviours and contexts represented in these datasets can differ substantially from those encountered in clinical settings.
This creates challenges for generalisability, particularly when models are applied to different populations, languages, cultures or healthcare environments.
The review therefore recommends that AI models currently be considered tools for screening, risk identification and clinical decision support, with further validation required before they can be relied upon for clinical diagnosis.
Ethical Governance, Privacy and Equity Remain Critical Priorities
A central finding of the review is the need for stronger ethical and regulatory frameworks to guide AI implementation in mental healthcare.
The researchers identify persistent challenges involving data privacy, informed consent, algorithmic bias, transparency, accountability and the preservation of human therapeutic relationships.
Mental health information is particularly sensitive, making secure data management and clear communication about how personal information is collected, stored and used essential.
The review also identifies a significant gap in ethical reporting across the research landscape. Approximately one-third of the studies explicitly reported institutional ethical approval, informed consent procedures or data anonymisation practices.
Many computational studies relying on publicly available datasets provided limited information about consent, data provenance or user awareness.
The authors call for more consistent ethical reporting, stronger data governance and clear accountability mechanisms throughout the development and deployment of AI-based mental health technologies.
Addressing Bias and Cultural Representation
The review raises particular concerns about the geographical and cultural limitations of existing AI datasets.
Many models are developed using data predominantly drawn from Western, English-speaking or high-income populations. This creates risks when technologies are deployed in communities with different languages, cultural understandings of mental health or healthcare needs.
Without representative datasets and culturally responsive design, AI systems may produce inaccurate assessments or recommendations that fail to reflect users’ lived experiences.
The researchers recommend greater involvement of diverse populations, local mental health experts and clinicians in the design, evaluation and ongoing monitoring of AI interventions.
These considerations are especially important for efforts to expand digital mental health services across low- and middle-income countries, where infrastructure, workforce capacity and access to care may differ significantly.
Building Interconnected, Human-Centred Mental Health Systems
A distinctive contribution of the review is its examination of AI through the lens of interconnected healthcare networks.
The authors explore how AI technologies can connect individuals, healthcare professionals, digital platforms and service providers to support more coordinated and responsive mental healthcare.
These network-based approaches offer opportunities to improve service accessibility, support personalised interventions and strengthen continuity of care, including during periods of increased demand or disruption.
However, successful integration requires attention to how technologies interact with existing clinical services, governance arrangements and human relationships.
The researchers emphasise the importance of human oversight, with AI supporting clinical expertise and decision-making while preserving the trust, empathy and accountability essential to mental healthcare.
The review also highlights the potential of privacy-preserving approaches such as federated learning, which enables collaborative AI model training across institutions without requiring the sharing of raw personal data.
Priorities for Future Research and Implementation
To support the responsible development of AI-enabled mental healthcare, the researchers identify several priorities:
- Long-term clinical evidence: Larger, multicentre trials and longitudinal studies to establish sustained effectiveness, safety and real-world outcomes.
- Inclusive and representative datasets: Greater cultural, linguistic and demographic diversity to improve fairness and generalisability.
- Transparent ethical governance: Stronger standards for informed consent, data protection, algorithmic accountability and explainability.
- Human oversight and clinical integration: Clear roles for healthcare professionals in supervising AI-assisted interventions and decision-making.
- Interdisciplinary collaboration: Closer cooperation between clinicians, researchers, technologists, policymakers and communities to ensure technologies respond to real-world needs.
The authors also advocate for ethics-by-design approaches, embedding privacy, fairness, accountability and user autonomy throughout the AI development process.
Implications for the Future of Digital Mental Health
For the international digital mental health community, this research reinforces the importance of considering AI within the broader systems of care in which it operates.
The findings highlight opportunities to expand access, support earlier intervention and strengthen the responsiveness of mental health services, while drawing attention to the clinical, ethical and implementation challenges that must be addressed for these benefits to be realised.
These priorities closely align with eMHIC’s commitment to advancing safe, equitable, evidence-informed and person-centred digital mental health through international collaboration, knowledge exchange and system-wide integration.
As AI technologies continue to evolve, the review provides a timely contribution to discussions about how innovation can be translated into meaningful improvements in mental health outcomes.
Its overarching message is clear: realising the potential of AI in mental healthcare will require robust clinical evidence, transparent governance, cultural responsiveness and continued human oversight, supported by collaboration across the wider mental health ecosystem.
Research Source
Title: Network-based artificial intelligence in mental healthcare: A systematic review of chatbots, artificial intelligence/machine learning models and ethical considerations in global healthcare networks
Authors: Zahra Rezaei, Atieh Khorraminia and colleagues
Journal: DIGITAL HEALTH
Publication date: 11 February 2026
Research type: Systematic review of 37 studies, covering research published between 2017 and 2024
Read the original research: https://doi.org/10.1177/20552076261421688
