ChatGPT provides original, human-like responses to user prompts based on supervised and reinforcement machine learning techniques. It has become the poster child of generative AI, which is widely diagnosed to disrupt many realms of life — including science communication. This essay reflects on this development. It discusses opportunities for the practice of science communication, such as generative AI’s translational and multimodal capacities and its capacity to provide dialogical science communication at scale, but also challenges in terms of accuracy, ‘wrongness at scale’ or job market implications. It also ponders implications for research on science communication, which has largely neglected (generative) AI so far. It argues that scholars should analyze public communication “about” AI as well as communication “with” AI, given its ‘increased agency’. Furthermore, scholars should analyze the impact of AI on science communication itself and the larger science communication ecosystem.