Generative AI Adoption is More Widespread than Predictive AI at U.S. Hospitals
Chelsea Richwine, Jordan Everson, Paige NongAbstract
Introduction
Artificial intelligence (AI) adoption among U.S. hospitals has accelerated rapidly, yet national estimates of predictive and generative AI implementation across clinical and administrative settings remain limited.
Methods
Using data from 2,203 nonfederal acute care hospitals in the 2025 American Hospital Association Information Technology Survey, we described AI adoption, model source (electronic health record [EHR] developer, third-party developer, or self-developed), and specific clinical and administrative use cases. We assessed associations between hospital and area characteristics and AI adoption, model source, and number of AI use cases using weighted multivariable regression.
Results
In 2025, 81% of hospitals reported using AI, more than double the rate reported in 2024. The most common applications were ambient documentation (73%) and predicting health trajectories or risks (70%). System affiliation and use of the market-leading EHR were the strongest predictors of AI adoption. Hospitals using EHR-developed AI reported 2.48 additional AI use cases (95% CI, 1.93–3.03).
Conclusion
Generative AI adoption surpassed predictive AI for the first time, driven by ambient documentation. While use of EHR-developed AI was associated with greater use, there was a greater influence of third-party vendors for generative AI compared to the predictive. Continued monitoring will be essential to assess AI's impact across healthcare.