U.S. clinicians largely see GenAI as useful, but readiness remains uneven
Generative artificial intelligence is moving into clinical work faster than many institutions can define guardrails for its use. In a nationwide survey of U.S. healthcare providers, most respondents viewed GenAI as useful in current patient care, yet far fewer had formal training or an organizational framework that could support dependable use at the bedside. The gap between enthusiasm and implementation was wide enough to matter for daily workflow, documentation quality, and the credibility of AI-assisted decision support.
Of 109 completed core surveys, 86.9% of respondents agreed that GenAI is useful in current patient care, and that proportion rose to 92.9% when asked about future usefulness. Those findings point to a workforce that is already mentally moving toward AI-assisted practice. But confidence was not matched by infrastructure: only 42.4% reported formal GenAI training, and just 23.2% said their organization had begun adopting AI. In other words, clinicians are being asked to imagine a toolset that many have not yet been taught to use safely.
The survey sampled physicians, nurses, allied professionals, and other providers across the United States. Physicians made up 38.5% of respondents, nurses 16.5%, allied professionals 12.8%, and other providers 32.2%. Women accounted for 54.2% of the sample, and nearly two-thirds of participants were 50 years or older. Those demographics matter because comfort with digital tools, perceptions of occupational risk, and attitudes toward workflow change often vary by role and career stage.
Documentation gains and error reduction drove the strongest appeal
Two perceived benefits stood out. Improved documentation and clerking topped the list at 57.0%, followed by error reduction at 49.4%. Those are not abstract efficiencies. Documentation is one of the most time-consuming parts of clinical work, and any tool that can shorten note writing, streamline intake, or organize a patient story carries immediate operational value. Error reduction carries an even heavier clinical weight, because poorly structured notes, missed details, and transcription fatigue can ripple through diagnosis, medication reconciliation, and handoffs.
The pattern suggests that clinicians are not thinking about GenAI as a replacement for clinical judgment. They are looking at it as a support layer for repetitive cognitive labor. That distinction shapes how the technology should be deployed. If AI is used where language generation, summarization, or clerical synthesis dominates, its utility may be high without pushing it into roles where misclassification or hallucinated output could expose patients to harm.
Training gaps, privacy concerns, and job anxiety shaped adoption resistance
The most common barrier was limited AI knowledge, cited by 24.7% of respondents. Fear of job loss followed at 16.9%. Those concerns are clinically relevant because technology adoption rarely fails only on performance metrics; it fails when the people expected to use it do not trust it, do not understand its limits, or see it as a threat to professional identity.
Data privacy and ethical risk also remained central concerns in the survey’s conclusions. Those worries fit the reality of GenAI, which can process sensitive clinical language at scale but still depends on institutional policies for safe handling of protected health information, verification of outputs, and accountability when the tool is wrong. A model that drafts a clean note is useful only if the clinician can audit every clinically important statement before it enters the record. Without that layer of oversight, speed can become a liability.
The survey also found that only 23.2% of respondents worked in organizations that had begun adopting AI. That low adoption rate helps explain why training remains thin. When institutions move slowly, informal learning fills the void, and clinicians are left to experiment without standard operating procedures. The result is predictable: uneven implementation, variable confidence, and a tendency to treat GenAI either as a novelty or as a hidden risk rather than as a governed clinical tool.
Support exists, but clinicians want human oversight and structured training
Despite the reservations, 72% of respondents expressed willingness to support broader GenAI adoption. Two safeguards rose to the top: human oversight at 67.1% and staff training at 60.8%. Those preferences reveal what clinicians want from implementation. They are not asking for autonomous systems; they are asking for supervised systems with clear human accountability.
This is the practical pathway for GenAI in healthcare. A model that drafts, summarizes, organizes, or flags likely inconsistencies can ease administrative load, but the final clinical act still needs a licensed professional who understands context, exceptions, and risk. Human oversight also protects the patient relationship. If AI is used to give clinicians back time, that time should be redirected toward conversation, examination, shared decision-making, and reassurance, not simply absorbed by more documentation.
The paper’s conclusion that administrators should use efficiency gains to improve provider-patient relationships and work-life balance is consistent with that logic. Burnout in clinical practice is often driven by administrative saturation, not by the core work of medicine itself. A tool that reduces clerical burden could help, but only if the time saved is actually returned to clinicians rather than consumed by additional throughput demands.
Gender, age, and qualifications influenced how GenAI was viewed
Several statistically significant associations emerged in the survey. Perceived AI usefulness varied by gender, with χ2 = 29.2 and p < 0.001. Organizational adoption of AI and views on where AI is most useful differed by qualifications, with χ2 = 31.6.2 and p = 0.047 for organizational adoption and χ2 = 101.1 and p < 0.001 for the setting in which AI was most useful. Support for AI adoption also varied by age, with χ2 = 18.0 and p = 0.02.
Those associations suggest that attitudes toward GenAI are not uniform across the workforce. Career stage, professional training, and role-specific exposure likely shape whether AI is seen as a convenience, a threat, or a tool that could help with repetitive tasks. For implementation teams, those differences matter. A one-size-fits-all training module is unlikely to address the concerns of a senior physician, a nurse managing bedside workflow, or an allied professional whose daily tasks differ from both.
The survey design cannot prove causality, and it cannot predict how attitudes will shift as institutional systems mature. Even so, the signal is clear enough for hospital leadership to act now. Clinicians are already using or anticipating use of GenAI, and many want clearer instruction, formal oversight, and transparent policies before adoption accelerates further.
Safe adoption will depend on governance, not enthusiasm alone
The strongest implication from this survey is not that GenAI should be adopted everywhere at once. It is that implementation without governance will deepen distrust, while implementation with training and supervision may convert interest into safe use. Hospitals and health systems that move first will need policies for data handling, output review, documentation standards, escalation pathways, and accountability when AI-generated content is inaccurate or incomplete.
The survey also suggests a simple operational truth: clinicians are willing to engage with GenAI when it removes friction rather than adding risk. Improved documentation, fewer errors, and better workflow efficiency are the reasons support exists. Privacy uncertainty, limited formal education, and fear of job displacement are the reasons caution persists. For healthcare leaders, the next step is not broader marketing of AI. It is building the conditions under which clinicians can use it without sacrificing safety, trust, or professional control.