The AI Penalty and Disclosure Paradox: Trust, Authenticity and Knowledge Uptake in AI-Mediated Communication

As Artificial Intelligence is increasingly employed to mediate human interactions, there is uncertainty around how these technologies impact human behaviour and how such mediated interactions are perceived. One such case is the expanding use of Generative Artificial Intelligence (GenAI) to create and disseminate content across a range of media contexts, known as AI-Mediated Communication (AI-MC). Such cases raise important questions about how those on the receiving end of such outputs respond to these new forms of interpersonal communications. Based on evaluating a piece of written text, this study explores how recipients' perceptions of the content creator's trustworthiness and authenticity, alongside their willingness to use the content (via knowledge uptake) are impacted by author type (human, AI-assisted, or fully AI) and media context (workplace email vs. social media post). We conducted a pre-registered experimental survey study with a 3 (author type) x 2 (media context) between-subjects factorial design (N = 547) where participants evaluated a piece of written text. Our findings demonstrate an “AI penalty”, with communication involving AI being perceived as less trustworthy, less authentic, and less useful for knowledge uptake. We also identify a “disclosure paradox”, as while participants believed it is important to disclose AI use, they also penalise the communication when such disclosure is made, which risks creating perverse incentives for non-disclosure of GenAI use for written communications.