A Representative Study on Human Detection of Artificially Generated Media Across Countries

The study examines human ability to detect AI-generated media through an online survey conducted in the USA, Germany, and China, involving 3,002 participants evaluating audio, image, and text samples. Using a preregistered methodology, participants rated whether media was human- or machine-generated. Results show that AI-generated media is nearly indistinguishable from human-created content, with participants often guessing. Detection accuracy was below 50% for images and slightly higher for text and audio, influenced by factors such as trust, cognitive reflection, and familiarity with deepfakes. The study underscores the risks posed by AI-generated media in spreading disinformation, emphasizing the need for improved detection tools to maintain trust and accuracy in digital communications.