Status
The full academic paper is currently being written. It will cover the survey methodology, latent class analysis, segment profiles, and experimental results in detail. We expect to make a preprint available in 2026.
In the meantime, the interactive data and segment profiles on this site represent the complete findings. For a summary, see the Key Findings page. For the full dataset with all charts, see The Full Picture. For individual segment deep-dives, see the Five Americas.
Abstract (draft)
Using a nationally representative survey of 2,735 U.S. adults, we identify five distinct segments of American opinion on artificial intelligence risk. Employing latent class analysis on items measuring concern intensity, concern type, institutional trust, and policy preferences, we find that the American public's relationship with AI risk is far more textured than a simple worried-to-unconcerned spectrum. We further test nine policy argument framings across these segments, finding that perceived convincingness and tone perception vary dramatically by audience. These results have implications for AI governance and public understanding of emerging technology risk.
Access
If you would like to review a preprint draft or request access to the anonymized survey data, please reach out via the contact page. We welcome peer review and are happy to share materials.
A topline methodology document with sample demographics, fielding details, and key frequency tables will be available here shortly.