Artificial intelligence is turning news consumption into a highly social process, in which people can naturally interact with AI agents — via smart speakers or robots — to obtain news. Yet the influence of socially interactive AI on news use remains understudied. Building on a pilot showing that perceived device intelligence affects trust in news, this 12-month project asks whether AI agents’ communication styles and physical resemblance affect trust in, and learning from, news. It combines two methods: a pre-registered lab experiment with a representative sample of Dutch news users, using a 2×2 design that crosses agent type (humanoid robot vs. non-humanoid smart speaker) with interaction style (social vs. transactional); All news stimuli are drawn from mainstream, ideologically centred Dutch outlets, isolating the mechanisms through which AI’s social capabilities shape news users.

Exposure to opposing views is a condition for the healthy deliberation that sustains democracy — but only if people are willing to hear from the other side. Recent evidence suggests people perceive AI agents as less biased and less persuasive than humans, and are therefore more open to counter-attitudinal information from AI. But that evidence comes from generic persuasion contexts, whereas news activates motivated reasoning, selective exposure, and source-trust mechanisms that lower-stakes messages do not. This study asks whether AI agents make citizens more open to counter-attitudinal news, and whether effects differ by communication style (social vs. transactional) and agent type (chatbot, smart speaker, robot). Building on two pilot studies, we use a pre-registered between-subjects experiment with a 2×3 design plus a conventional news-site baseline (seven conditions), exposing a non-representative sample of Dutch adults (300) to counter-attitudinal news to test whether socially engineered agents help or hinder openness across political divides.