Abstract:  As AI increasingly participates in science communication, it is unclear how people evaluate AI as a source of scientific information. This study examines how message framing and identity cues shape public evaluations of communicative AI and whether these effects differ when AI is encountered through reading or direct interaction. Two preregistered online experiments in Germany contrasted science communication about AI (reading a news-style article) with science communication with AI (interacting with a chatbot), manipulating risk versus progress framing and human-like versus machine-like cues. In an article-based context (Experiment 1, N = 862), progress framing increased trust in AI, while machine-like wording further improved trust. In an interactive context (Experiment 2, N = 868), framing shaped evaluations indirectly by reducing fear, while human-like cues increased social presence and parasocial connection, producing indirect gains across key outcomes. Across both experiments, higher AI competence was associated with more positive evaluations. Overall, the findings show that framing and design cues exert modest but systematic effects that depend on the communicative format.

Keywords:  artificial intelligence; framing; identity cues; machine heuristic; science communication; social presence; trust

Published:   21 April 2026

Issue:   Vol 14 (2026): Exploring Engagement With Complex Information: Perspectives on Generative AI as an Information Intermediary (In Progress)