This Reuters Institute study explores how audiences in Mexico, the UK and the US perceive the growing use of generative AI in journalism. Participants tested 25 different applications, revealing that public attitudes become more nuanced when people see how AI is actually used in news production. Acceptance varies according to the task, the type of content, the medium and, above all, the degree of human oversight. Most participants considered editorial supervision essential, while transparency became especially important when AI was used to create or substantially alter news content.