This whitepaper assesses which AI support tasks in primary care and internal medicine have enough evidence for careful consideration, which remain experimental, and what governance Latin American teams would need. It is a rapid narrative synthesis based on PubMed and targeted searches run in June and July 2026, covering longitudinal synthesis, documentation, EHR search, messaging, risk prediction and diagnostic support. It is non-peer-reviewed gray literature published by Arkangel AI, and it does not evaluate any Arkangel product.
A systematic review of 519 studies of large language models in health care found that only 5% used real patient-care data. In a study integrated into the EHR, AI-drafted patient-message replies increased physicians' reading time by 21.8%, did not significantly reduce response time and made replies 17.9% longer, and a randomized trial with 50 physicians found no significant improvement in diagnostic reasoning with access to a language model. A review of predictive algorithms in primary care identified 43 implemented tools with limited public evidence across their life cycle. Direct regional evidence is narrow, led by a Brazilian scoping review of 27 studies, and the authors position AI as reviewable support for synthesis, documentation and search rather than an independent diagnostic or therapeutic tool.
