Echoes of Doubt: Exposure to Information About Generative AI Decreases Believability of News

The emergence of generative artificial intelligence (GenAI) has sparked a debate about its potential misuse for creating political disinformation. However, the effects of providing information about generative AI on disinformation perceptions remain unclear. We fill this gap by testing the impact of GenAI literacy interventions on truth discrimination (i.e., the ability to accurately distinguish between genuine and false online news) and deception bias (i.e., the tendency to believe that online news is false) in an online experiment among 897 Canadian adults. Respondents were randomly assigned to a GenAI literacy intervention (explainer videos), showing how ChatGPT and Midjourney can be used to create political disinformation vs. art. The GenAI interventions increased participants’ propensity to classify online news as false, yet signal detection analyses showed no improvement in truth discrimination. In addition, we find evidence for a deception bias where participants have a slight tendency to judge online news as false rather than true. We conclude that GenAI literacy interventions need to be carefully crafted to avoid further undermining the believability of genuine news.