Generative Artificial Intelligence With Youth Codesign to Create Vaping Awareness Advertisements

Importance Traditional approaches to developing youth vaping awareness campaigns are time-consuming and can create critical delays in public health response. Although generative artificial intelligence (AI) offers promising capabilities for health communication, research has been limited to text-only messages. Objective To evaluate (1) the perceived message effectiveness (PME) of AI-generated, youth-codesigned vaping awareness social media advertisements (ads) compared with existing ads from official health agencies and (2) how different source labeling is associated with PME. Design, Setting, and Participants This randomized clinical trial used a 2 (ad source) by 4 (source labeling) design and was conducted online from September 2 to September 19, 2024. Participants were individuals aged 16 to 25 years. Exposure All participants evaluated 50 ads from 2 sources (within-participants; 25 AI-generated, 25 existing) in random order. Main Outcomes and Measures The primary outcome, PME, was measured using the validated PME Scale for Youth, which assessed 2 effects perceptions (vaping perception and behavioral intent) and 3 ad perceptions (attention, information, and convincingness) on 7-point scales, with lower scores indicating better effectiveness for effects perceptions and higher scores indicating better effectiveness for ad perceptions. Results Six hundred fourteen individuals (mean [SD] age, 20.5 [2.9] years; 300 female [48.9%]; 300 male [48.9%]; 14 other [2.3%]) provided 30 700 observations. Participants were randomly allocated to 1 of 4 experimentally manipulated labeling conditions (between-participants): (1) no source label (147 participants), (2) made with AI (158 participants), (3) made by the World Health Organization (WHO) (151 participants), or (4) made with AI by the WHO (158 participants). AI-generated ads demonstrated noninferiority to existing ads across all measures. AI-generated ads received better ratings for discouraging vaping (b = 0.09; 95% CI, 0.01 to 0.17), attention-grabbing qualities (b = −0.15; 95% CI, −0.26 to −0.03), and convincingness (b = −0.18; 95% CI, −0.30 to −0.07) (all P for noninferiority tests <.001). Source labeling showed no significant association with PME scores (χ 2 values ranging from 0.10 to 4.19; all P > .20). Conclusions and Relevance In this randomized clinical trial of vaping awareness social media ads, AI-generated, youth-codesigned ads achieved superior effectiveness ratings compared with existing ads. These findings support the potential for leveraging generative AI technology in public health campaigns, while indicating the need for appropriate governance frameworks as AI-generated health materials become increasingly prevalent. Trial Registration ClinicalTrials.gov Identifier: NCT07042789