News outlets are currently adopting AI to summarize news stories and experiment with conversational agents to convey news. In this paper, we explore this emerging practice, in which an AI-powered agent retells news in dialogue with the user rather than presenting them with a fixed narrative. In a co-speculation workshop with industry professionals and through a field trial of an LLM-based conversational news agent probe, we explore the design space of this conversational news format and examine the experiences, insights, and concerns reported by the participants. Our contribution is twofold. First, we identify five dimensions that shape how news can be retold by a conversational agent. Second, we provide a detailed empirical account of how users experience conversational news as clear and easy to follow, enabling them to probe and question stories in new ways, and address how these interactions are marked by tensions around trust, accuracy, and transparency.
ACM CHI Conference on Human Factors in Computing Systems