Trusting the Machine: Separating Trust in GenAI Authors from Trust in Their Content

As Generative Artificial Intelligence (GenAI) becomes more sophisticated and widespread, it matters how its perceived humanness affects trust. Previous work shows that anthropomorphism increases an agent’s perceived trustworthiness as an advisor. We asked whether an author’s presentation also affects trust in the information it produces in a non-advisory role. In two experiments, participants read a science communication blog post attributed to an author with varying anthropomorphic cues. They rated the trustworthiness of the author (author trust) and the content (content trust). In Experiment 1 (N = 270), the post was attributed to a GenAI author described with low or high anthropomorphic cues, or to a human expert. The human author, who also differed in implied professional expertise, received higher author trust than either GenAI author. This effect was reliably larger than the effect on content trust. The low- and high-cue GenAI authors did not reliably differ, including in anthropomorphism ratings. In Experiment 2 (N = 148, preregistered), participants chatted with a GenAI chatbot, presented as the author, that used fewer (low condition) or more (high condition) anthropomorphic cues. The high-cue chatbot was rated more anthropomorphic and received modestly higher author trust. No reliable content trust effect was detected, and evidence that the two effects differed was inconclusive. Author and content trust were strongly correlated in both experiments. These results suggest that cues to an author’s humanness may shape trust in the author more than in what it communicates, although the clearest evidence came from comparing human and GenAI authors.