Artificial intelligence innovations in substance use prevention on social media: A scoping review

This scoping review examined the current application of artificial intelligence (AI)/machine learning (ML) models on social media platforms for substance use prevention, along with key challenges and recommendations for implementation.

Our scoping review followed the framework proposed by Arksey and O'Malley for literature search as well as the PRISMA-ScR guidelines for reporting.

We searched seven databases through November 2024. Studies were included if they applied AI/ML on social media platforms for substance use prevention. Data were charted and synthesized narratively, focusing on study characteristics, targeted substances, social media platforms, AI model types, application purposes, key challenges, and recommendations.

Fifteen studies were included, focusing on AI/ML for monitoring substance-related trends, predicting behavioral risk, and evaluating prevention campaigns. Twitter was the most frequently examined platform with most studies being conducted in the United States. Across studies, 22 AI/ML models were applied, mainly natural language processing (50.0%), machine learning classifiers (36.4%), and computer vision techniques (13.6%). Key challenges that limit the expansion and effective use of AI/ML models for preventing substance use on social media included language ambiguity, model interpretability, data imbalance, and ethical concerns. Recommendations emphasized the need for explainable AI, human validation, and cross-platform data integration.

The evidence has several limitations, including insufficient reporting on AI model development and validation, unclear sample characteristics, and inconsistent outcome measures, which hinder quality assessment and comparability. Current applications of AI/ML in substance use prevention on social media are largely descriptive and lack real-time, user-engaging features. Most evidence comes from high-income, English-speaking contexts. To enhance effectiveness, future research should address technical and ethical barriers, adopt longitudinal designs, and develop interactive, locally adapted interventions.