Artificial Intelligence and Mental Health Misinformation: A Systematic Scoping Review

Background and Purpose
Artificial intelligence (AI) is increasingly embedded in how clients, practitioners, and clients access mental health information, from search engines and social media platforms to chatbots and clinical decision-support tools. While AI has the potential to expand access to information, it also introduces significant risks related to misinformation, disinformation, and misleading or fabricated content. Mental health misinformation is especially consequential for social work practice and education, as it may shape help-seeking behaviors, treatment adherence, stigma, and trust in professional services and can have serious downstream consequences, including delaying or deterring appropriate care, promoting ineffective or harmful self-treatment, and eroding trust in licensed professionals and evidence-based interventions. Despite growing interdisciplinary scholarship in this field, the existing evidence has not been systematically examined in how AI is being used in mental health-related misinformation. This study aims to map the current literature on AI-related mental health misinformation, assess the quality of evidence, and identify implications for social work education and professional competencies.

Methods
This systematic scoping review followed JBI guidance and the Population-Concept-Context framework. Searches were conducted in PubMed, Scopus, ProQuest, PsycINFO, Web of Science, IEEE Xplore, Cochrane, and EBSCO All Databases. Eligible studies examined AI-generated, AI-mediated, or AI-detected mental health misinformation in digital contexts. Data extraction captured study characteristics, AI modality, misinformation type, and outcomes. Studies were dual-coded and quality-appraised using a modified GRADE-informed framework.

Results
Nineteen studies met inclusion criteria. Nearly all conducted in non-clinical, public-facing digital contexts (n = 18, 94.7%). General mental health was the most common focus (n = 9, 47.4%), followed by substance use treatment or related questions (n = 5, 26.3%), suicide and self-harm (n = 2, 10.5%), neurodevelopmental disorders (n = 2, 10.5%), and depression (n = 1, 5.3%). AI was most frequently examined as a misinformation detection or classification tool (n = 8, 42.1%), particularly for social media surveillance and identification. However, nearly one-third of studies documented AI systems generating, amplifying, or failing to correct misleading content (n = 6, 31.6%), including hallucinated citations, guideline non-adherence, and unstable responses under prompt variation. Experimental work demonstrated that large language models could be intentionally manipulated to produce persuasive disinformation (n = 2, 10.5%), while correction-focused interventions remained limited (n = 2, 10.5%). Evidence quality was primarily moderate to low (n = 16, 84.2%), while methodological limitations and limited evaluation of downstream behavioral or clinical impact were common.

Conclusions and Implications
The rapid integration of AI into mental health information ecosystems has outpaced the evidence needed to safeguard accuracy, trust, and client well-being. This misalignment poses tangible risks, yet it also represents a defining moment for the profession. Resilience in this context is not passive adaptation but an active capacity to critically engage, ethically govern, and strategically shape AI-mediated environments. Our findings suggest that AI-related mental health misinformation is not merely a technical flaw but a systemic challenge that intersects with vulnerability, trust, and professional responsibility. Building resilience therefore requires strengthening AI literacy, reinforcing human judgment, and embedding regulation, accountability, and transparency into technology use.