Feeling Iffy About Generative AI: Investigating Audiences’ Trustworthiness Perceptions of Task-Specific AI Disclosures

Abstract

News organisations are experimenting with how to best integrate generative AI into their journalistic workflows. This raises questions about how AI use should be disclosed and how such disclosures affect readers. Prior research shows predominantly negative effects on perceived trustworthiness and credibility, but says little about how AI disclosures for different use cases compare to each other. In this study, we report the results of a conjoint experiment (N = 683) on the effects of task-specific AI disclosures on the perceived trustworthiness of news. While the magnitude of effects varies, we consistently find negative effects across all task-specific AI disclosures. However, moderation and cluster analyses suggest that these effects are not universal, but depend on individual-level characteristics that co-determine AI disclosure effects. By (1) highlighting important individual-level moderators such as respondents’ political position and their knowledge of journalistic AI, and (2) describing five distinctive preference profiles and their predictors, our results inform future research and help practitioners cater AI disclosures to particular groups of readers. To this end, we also situate our work within broader debates about what meaningful transparency should look like.