Abstract
INTRODUCTION: Developing mass meda campaigns to address rising youth vaping rates in Australia is timely and resource-intensive. Generative AI offers scalable content production, but little is known about youth perceptions of AI-generated multimedia materials or how their feedback can inform co-design processes. METHODS: We conducted a two-phase qualitative study in Queensland, Australia. Phase 1 explored adolescent (n = 10, ages 13-20) responses to 120 vaping awareness materials produced using an automated-AI framework. Focus group participants sorted materials into 'effective' and 'ineffective' piles and provided feedback. Based on feedback and quality criteria, 25 revised materials were created using an AI co-design framework incorporating iterative, few-shot prompting and manual text-image integration. Phase 2 explored young adult (n = 9, ages 18-25) perceptions of revised materials via semi-structured interviews. Inductive thematic analysis was conducted. RESULTS: Phase 1 participants rejected automated-AI-generated materials due to misaligned text-image combinations, artificial imagery, unrealistic vaping devices, and inauthentic language. Phase 2 identified six key characteristics of effective AI-co-designed materials that aligned with established health communication principles including visual appeal; focus on immediate consequences; relevance to youth; provision of practical advice; avoidance of ambiguity and fearmongering; and integration of multiple themes to reach diverse youth audiences. DISCUSSION AND CONCLUSIONS: AI tools can rapidly generate messages but an AI-co-design framework incorporating expert input and audience feedback is required to produce materials that are relevant, authentic, and evidence-based. This framework offers a promising pathway for developing timely, scalable responses to public health challenges such as youth vaping; though continued research is needed for effective and ethical implementation across diverse contexts.