>we conducted a preregistered survey experiment on 8,221 US respondents comparing the persuasiveness of English-language foreign covert propaganda articles sour

>we conducted a preregistered survey experiment on 8,221 US respondents comparing the persuasiveness of English-language foreign covert propaganda articles sourced from real-world campaigns to text generated by a large language model, which is a form of AI. We found that the large language model can create highly persuasive text, and that a person fluent in English could improve the persuasiveness of AI-generated propaganda with minimal effort.