No Easy Fix to Countering AI-Generated Visual Disinformation: The (in)Effectiveness of AI-Labels, Fact-Check Labels and Community Notes

As generative AI makes it easier to create synthetic visuals at scale and speed in digital information environments, AI-driven visual disinformation is becoming more common on social media. However, while much research highlights its potential harm, less is known about how to reduce its potential to mislead. In this study, we therefore conducted a preregistered online experiment in the Netherlands (N = 1018) to test the effectiveness of various platform interventions: (1) AI labels or “watermarks,” (2) fact-check labels, and (3) community notes. We tested how effective these sources are in lowering credibility of the false visual and belief in the false claim it portrays across two topics: climate change and immigration. When pooling both topics together and for climate-change related disinformation in isolation, the interventions showed no significant differences in effectiveness, suggesting a null-finding. However, when zooming in on visual disinformation about immigration, community notes were most effective, especially among participants with strong anti-migrant views. Our finding suggests that while labelling has limited impact overall, its effectiveness might vary by context, and no one-size-fits-all solution exists for combating AI-generated visual disinformation.