Foodie traps within facebook cannabis promotional posts: Deploying multimodal deep learning AIs to monitor audience engagement.

BACKGROUND: Cannabis marketing has flourished due to the legalization of recreational cannabis across multiple U.S. states, particularly on social media platforms that effectively reach adolescents and young adults. Understanding how cannabis promotional content shape audience engagement is important for monitoring exposure and informing public health and regulatory discussions. METHODS: Leveraging advanced computational analysis, including the multimodal large language model LLaVA and GPT-4o, we analyzed a comprehensive dataset of Facebook posts from 2021 to 2022 to explore associations between visual tactics and online audience engagement (e.g., likes, shares, and comments). RESULTS: Unsupervised clustering identified ten distinct visual themes, with human and animal portrayals, food cues being the most prevalent. High-level visual features such as food cues were consistently associated with increased shares and comments. Also, low-level color characteristics, particularly red, yellow, and orange, predicted higher audience engagement. Nearly half of posts contained health benefit-oriented messaging, whereas explicit health risk or safety information was rare. DISCUSSION: Visually engaging strategies in cannabis advertising on social media are associated with higher audience engagement and may increase the visibility of cannabis-related content, raising public health concerns about widespread exposure and the normalization of cannabis use.