Artificial intelligence in science communication: emerging practices, credibility, and public trust

Generative artificial intelligence has spread rapidly through the infrastructure of science communication. However, its implications for how the public accesses, evaluates, and trusts science remain uncertain. This mini-review brings together current conceptual, experimental, and survey-based research. The aim is to develop an integrated framework that positions generative AI as an epistemic intermediary. The framework connects three areas: emerging communication practices, the credibility of AI-enhanced messages, and public trust in science. The literature on practices shows that large language models can enhance understanding through simplification, translation, and dialogue, and may even reduce skepticism toward science. However, these tools carry risks of inaccuracy and bias. Credibility studies attribute a small but consistent penalty to visible AI authorship. However, this effect is strongly context-dependent; findings regarding both pro-human bias and the lack of impact from labeling remain controversial. Trust studies, meanwhile, reveal that the public's trust in generative AI remains limited. Furthermore, the disclosure of AI use, even by researchers, can reduce perceived legitimacy and thereby diminish trust. Across these three areas, transparency design, AI literacy, prior trust in science, and the controversial nature of the topic emerge as key moderating factors.