Sticky associations: Quasi-experimental study of algorithmic memory and drug recommendations on snapchat

Background Social media platforms host markets for illicit drugs, facilitating access and shaping buyer and seller behaviour. Recommendation algorithms have been shown to suggest illicit drug sellers to at-risk users. This study documents whether users with prior connections to drug sellers on Snapchat can exit drug networks through voluntary disengagement and avoid further exposure to drug advertisements. Methods A quasi-experiment conditioned 40 research profiles by entering nicotine and digital sex work markets and accepting drug sellers suggested by Snapchat's recommendation algorithm. 30 profiles received and engaged with illicit drug seller suggestions during conditioning; 10 did not and remained in grey market networks. After four days of conditioning, all Snapchat connections were removed from the research profiles. The algorithm was given 48 h to adjust recommendations before suggestions to connect with drug sellers were recorded. Results 70% of profiles with prior drug connections received drug seller recommendations after removing all contacts. The remaining 30% lost exposure to drug sellers. Conclusion Recommendations after removal indicate an algorithmic memory, reflecting associations from prior exposure. This limits agency by adding friction for people attempting to leave drug networks. The findings contradict digital drift theory’s assumption that disengagement from deviant online spaces is relatively easy and suggest platform recommendations can serve as drug triggers for individuals attempting to limit or cease drug use.