Mapping the Landscape of AI-Driven Substance Use Services: A Scoping Review Organized By the Service Delivery Lifecycle

Background: Substance Use Disorders (SUDs) represent a critical global health burden, yet access to care is frequently hindered by provider shortages and systemic barriers. While Artificial Intelligence (AI) has emerged as a promising tool to expand service reach, the literature remains fragmented, often focusing on narrow technical innovations rather than integrated clinical workflows. This scoping review addresses this gap by synthesizing current evidence on client-facing AI innovations through the lens of the SUD service lifecycle.

Methods: Following PRISMA-ScR guidelines, a systematic search of the PubMed database initially identified 619 empirical studies containing keywords “artificial intelligence” and “substance use,” or similar terms, in the title or abstract. By screening, we narrowed the final sample to nine studies, including six empirical studies and three protocols. The review categorized interventions into specific clinical stages: screening/recruitment, assessment/diagnosis, treatment, and long-term recovery monitoring. Methodological rigor was documented by analyzing evidence levels, ranging from pilot usability tests to Randomized Controlled Trials, and sample sizes.

Results: Among six empirical studies, four were conducted in the United States and two were conducted in Spain and Sweden, and all three protocols were plans for the USA. In general, the number of studies, and the proportion of protocols in included studies, and the diversity of authors show that the field is still developing. The evidence base revealed a distinct maturity across the service continuum. In treatment stage, conversational agents were supported by robust RCTs with sample sizes ranging from 180 to 513, demonstrating high user satisfaction and non-inferiority to human therapeutic bonds. These studies showed that the participants reported significant improvements in smoking abstinence and reductions substance use occasions. However, screening, assessment, and monitoring stages are primarily characterized by early-stage pilots and protocols with small sample. One study validated a screening system provided safe and hallucination-free responses and another confirmed that AI screening achieves high sensitivity and empathy rating. Critical null findings were identified, including high attrition rates and performance ceiling where AI failed to significantly improve long-term self-efficacy engagement compared to controls.

Implications for Social Work: The adoption of AI innovation in SUD service delivery may become increasingly inevitable as the field responds to persistent workforce shortages, growing service demands, and the need for scalable form of support. The findings of this review suggest that AI is more likely to function as a complement to, rather than a replacement for, human providers, particularly in areas requiring relational judgment, trust-building, and complex clinical decision-making. This means the social work profession must engage proactively in shaping how AI is implemented in practice rather than treating it as an external technical development. At the same time, equitable access to AI-based services must remain a central concern. Unequal access to digital tools, as well as limited representation of marginalized populations in training data, may reproduce or deepen existing disparities in SUD care. Future research must shift from standalone tool validation toward trials that test AI integration within real-world social work practice setting to ensure continuity of care across the longitudinal patient journey.