Think Twice: Improving Privacy Awareness with Tailored LLM-Powered Interventions

With the widespread sharing of photos on social media, increasing users’ awareness and encouraging privacy consideration of such sharing is critical. This study investigates the potential of large language models (LLMs) to support users in identifying possible interpersonal privacy violations prior to posting images on social media. We introduce two LLM-powered privacy interventions: categorical and granular, which vary in the level of detail about the image. We compare the privacy and cognitive implications of these nudges to generic privacy intervention (universal) and no-intervention conditions. Both categorical and granular interventions significantly reduced participants’ likelihood to share images, and the categorical intervention achieved this reduction while maintaining a lower cognitive load. Participants indicated privacy intervention as an educational tool, complementing their own judgment during the decision-making process. Overall, our findings suggest that tailored privacy insights can enable more informed and autonomous sharing decisions on social media, supporting both privacy protection and user agency.