Clinician readiness for artificial intelligence (AI) in health care is critical to realizing AI’s potential in practice. Clinicians are the interface between AI applications and patient care. They are responsible for interpreting and integrating AI outputs within the complex realities of individual patients, incomplete information, and clinical uncertainty. Even when AI systems are “trustworthy,” with rigorous risk assessment and mitigation built in, vulnerabilities remain at the point of clinical application where judgment, workflow integration, and accountability ultimately reside. To maximize benefits and minimize risks, clinicians must be “AI-ready”—that is, willing and able to understand, evaluate, and appropriately use AI tools in practice, as reflected by their knowledge and attitudes related to clinical AI.

AI introduces a unique set of concerns for clinicians, including accountability for errors, algorithmic bias, and potential erosion of the patient-clinician relationship [1,2]. Without sufficient knowledge or supportive attitudes, clinicians may fail to realize AI’s benefits. But of greater concern are threats to clinical quality, safety, ethics, patient trust, and clinician well-being. Recognizing these risks, health systems and professional organizations have identified an AI-ready clinical workforce as a top strategic priority [3,4].

AI use in clinical care is expanding rapidly, including ambient documentation [5], image analysis, and point-of-care clinical guidance. However, existing literature suggests clinicians have limited understanding of AI fundamentals, data and algorithmic bias, and ethical and legal implications, alongside concerns about workflow integration and clinician-patient relationships [6-9]. These gaps may be particularly problematic in primary care. Primary care providers (PCPs) are the first point of contact for most patients and interact with a wide range of clinical, operational, and administrative technologies. Limited readiness among PCPs could therefore have broad implications for patient safety and quality of care.

We conducted a national survey of PCPs in the Veterans Health Administration (VA) to describe current AI use, characterize knowledge and attitudes regarding AI in clinical care, and identify priority areas for education and implementation support. We also explored variations in knowledge and attitudes across PCP subgroups. Findings on readiness components are intended to inform health system efforts to ready PCPs for safe, effective AI use.

This national survey of VA PCPs reveals a promising foundation for successful AI implementation in primary care. Respondents expressed strong enthusiasm for tools that could improve workflow efficiency and enhance decision-making, especially in high-value use cases such as automated documentation, data extraction, and point-of-care guidance. Many respondents had already adopted AI chat or decision support tools. These findings highlight meaningful openness to AI innovation in primary care and suggest that health systems could capitalize on existing clinician interest and experimentation with AI at the front lines.

At the same time, respondents reported uneven knowledge about AI fundamentals and limited confidence addressing bias and ethical issues, which are domains central to safe and responsible AI use. Qualitative responses underscore the limited technical understanding of AI among many clinicians. Apprehension regarding liability, accuracy, and transparency suggests that clinicians recognize key risks and feel ill-equipped to evaluate or mitigate them. Training experience was limited and heterogeneous despite nearly universal interest. Together, these findings illuminate a tension, which is that 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 results align with prior studies showing clinicians’ rapid AI uptake alongside uncertainty around the implications of use, including technical evaluation, risk mitigation, and professional responsibilities [6-9]. Our study extends existing research by focusing on primary care—a setting with broad clinical reach and substantial implications for quality and safety—and by examining a single, integrated health system actively pursuing AI readiness. By identifying specific gaps in knowledge, attitudes, and training, our findings point to levers that health system leaders directly influence through implementation support and governance.

The VA and other health systems have prioritized developing an “AI-ready” workforce [21]. Our findings suggest concrete steps toward this goal. Standardized, clinician-focused training on foundational AI concepts, paired with policies requiring baseline AI literacy and tool-specific training, could help establish consistent expectations and competency across a health system. Many states mandate continuing medical education in high-impact areas such as end-of-life care, risk management, and electronic health record usage. Given that AI carries comparable implications for patient safety, quality of care, and workflow, similar mandatory educational requirements may be warranted as AI’s role in care delivery expands. Validated measures of AI readiness could further support needs assessment, targeted training and implementation efforts, and evaluation over time. Beyond education, governance structures can establish expectations for transparency, accountability, and workflow integration.

Concerns that were lower ranked on the survey remain important for comprehensive AI implementation, such as communicating about AI with patients, considering impact on patient-clinician relationships, and using AI for outreach support and population health management. Respondents may have prioritized these concerns lower because they are less immediate than daily workflow demands. Nonetheless, focused educational content can raise awareness of these important issues without overwhelming core training needs. A tiered approach to training could address immediate workflow and safety concerns first while introducing relational and population health considerations as AI use matures in practice.

Although this study focuses on VA primary care, several findings likely generalize beyond this setting. VA PCPs practice within an integrated health system that offers relatively consistent access to new technologies, training resources, and technical support. These features may facilitate earlier exposure to AI tools compared with some non-VA settings where access to these resources may be more variable. At the same time, our findings suggest that even in a system with substantial organizational capacity, clinicians still report uneven foundational knowledge and limited preparedness to evaluate potential risks, suggesting that these challenges may be at least as notable in more resource-constrained environments. Similar findings have been reported outside of the United States, showing that clinicians across diverse health systems express enthusiasm for AI but face gaps in readiness [17,22,23]. The parallels within the VA context and outside of it emphasize a broader need for health systems to invest in developing AI readiness among their clinicians in the interest of supporting safe and effective AI use.

This study has limitations. While our response is consistent with other VA clinician surveys [24-26] and respondents resembled nonrespondents demographically, technologically enthusiastic respondents may be overrepresented. Knowledge gaps and apprehensions in the population of nonrespondents may be even more pronounced, which would only emphasize the importance of readiness efforts to optimize clinicians’ AI use. Also, our subgroup analyses are intended to be hypothesis-generating rather than definitive; future studies should examine subgroup differences with finer grain.

VA PCPs demonstrate strong enthusiasm for AI tools promising workflow efficiencies and clinical support and report current use of generative AI and decision support technologies. However, they also report limited training and uneven knowledge in foundational and evaluative domains. These 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 of progress toward an AI-ready workforce. By providing one of the first system-wide assessments of PCP’s AI readiness within an integrated health system that is itself relatively early in its AI implementation journey, this study offers novel insights into how frontline clinicians are engaging with AI, where critical gaps in readiness exist, and which levers health systems can use to enable safe and effective AI adoption that optimizes patient care while minimizing risks. Addressing these gaps will enable health systems to capitalize on clinician openness while strengthening the quality and safety of primary care delivery at scale.