Exposure to (AI-Generated) Untrustworthy Websites in the 2024 U.S. Presidential Election

Generative AI’s ability to produce untrustworthy content at scale has heightened concerns about its influence on elections. We provide preregistered evidence of exposure to AI-generated untrustworthy websites during the 2024 U.S. Presidential Election, combining web-browsing logs and surveys from a national sample of American adults (N = 1,069; 6M visits) with expert databases of websites. The share of Americans visiting untrustworthy websites continued the decline witnessed from 2020 and 2016 to reach even lower levels. However, those exposed in 2024 visited more frequently, indicating concentrated engagement. Older adults and Republicans remained more likely to encounter such sites, though gaps narrowed, and Twitter/X’s role as a referrer increased. AI-generated untrustworthy sites reached more people in 2024 but comprised a small fraction of overall exposure with no pronounced demographic disparities.The results highlight both the persistence of risks and the need for targeted resilience strategies as generative AI tools evolve.