Knowledge, Attitudes, and Training Needs for AI in Primary Care: National Survey Study of Clinicians in the Veterans Health Administration

BACKGROUND: Clinicians are the interface between artificial intelligence (AI) applications and patient care. To maximize benefits and minimize risks of AI, clinicians must be "AI-ready"-that is, willing and able to understand, evaluate, and appropriately use AI tools in practice. Prior literature suggests that clinicians lack fundamental competencies in the use of AI. These gaps could be especially problematic in primary care, given its broad reach into patient care. We surveyed primary care providers (PCPs) in the United States' largest integrated health care system, relatively early in its widespread implementation of clinical AI for frontline use, in order to identify readiness gaps that may warrant particular attention as part of a comprehensive AI implementation strategy. OBJECTIVE: The aim of this study was to characterize PCPs' use, knowledge, attitudes, and training priorities related to AI in order to inform health system AI implementation efforts. METHODS: We conducted a national cross-sectional survey of United States Veterans Health Administration (VA) PCPs in October 2025, assessing AI use, self-reported knowledge, attitudes, and training experience. Descriptive analyses summarized responses with exploratory bivariate comparisons across clinician subgroups. Conventional content analysis with inductive coding was used to characterize open-ended responses providing a definition of AI. RESULTS: Among 170 respondents (170/989, 17.2% response rate), 66.5% (113/170) reported current AI use, most commonly generative AI and decision support tools. Overall attitudes toward AI were positive, with 70.6% (120/170) mostly enthusiastic or more enthusiastic than apprehensive. Confidence in understanding sources of AI bias (62/170, 36.5%) and ethical issues (81/170, 47.6%) was limited. When asked to define AI, very few respondents provided an accurate technical definition. Key concerns about use of AI included accountability, accuracy, and transparency. Though 88.2% (150/170) identified AI training as a priority, only 26.5% (45/170) had any training. Training experiences ranged widely in source, focus, and structure. CONCLUSIONS: PCPs are eager to harness AI's practical advantages but lack foundational competencies to do so in ways that maximize benefit and minimize risk. Our findings highlight a need for targeted education that prioritizes critical appraisal, workflow integration, and risk mitigation, supported by governance that addresses clinicians' concerns and validated measures to evaluate progress toward an AI-ready workforce. These steps can empower PCPs to leverage AI safely and effectively and strengthen the quality and safety of primary care delivery at scale.