Doximity 2026 Report Shows 54% of US Physicians Using AI in Practice

The Doximity 2026 State of AI in Medicine Report shows that 54% of US physicians are using AI in clinical practice. This statistic comes from a survey of 3,151 physicians and marks an increase from 47% in the March-April 2025 period to 63% in the November 2025-January 2026 period.

94% of the surveyed physicians are either using AI or interested in using it, pointing to widespread acceptance with specific hurdles remaining in accuracy and institutional support that could influence the pace of further adoption.

Key Adoption Statistics

The 54% adoption rate establishes the current baseline for AI use in US clinical practice according to the report. This figure combines responses from two distinct survey periods, revealing a clear upward trend as more physicians incorporate the tools into their daily routines. The data also indicate that 37% of all respondents use AI at least daily, representing 69% of adopters, which suggests strong integration once the technology is introduced.

This growth pattern reflects the increasing availability of AI applications that address real workflow needs. Physicians who start with basic tasks often expand their usage, leading to higher frequency rates among users. To apply this information, healthcare administrators can benchmark their organization's adoption against the 54% mark to identify gaps in rollout strategies.

The self-reported nature of the survey introduces some uncertainty, as responses depend on individual perceptions rather than audited records. The sample from Doximity members covers a broad cross-section but may not capture views from physicians less active on the platform. One typical error is to treat the overall 54% as uniform without examining the variations by specialty that the report provides in detail.

Additionally, the 94% interest level shows that resistance is low, with only a small percentage expressing no interest at all. This high interest combined with rising use points to a maturing market for AI in medicine. The mechanics involve physicians identifying tools that fit their specific needs, such as documentation or research support, which then drive repeated use.

The increase between survey periods can be linked to greater familiarity with available options and demonstrated value in reducing routine tasks. Physicians evaluating their own progress might track usage over similar intervals to detect comparable shifts. Data limitations include the possibility that later respondents had access to more refined versions of the tools, which could contribute to the observed rise independently of broader attitude changes.

Another consideration is that daily engagement at 69% of users implies the tools are not merely supplementary but embedded in standard processes for many. This level of frequency supports planning for sustained training rather than one-time introductions. A common pitfall when reviewing these statistics is overlooking how the combined periods smooth out short-term fluctuations that might appear in single snapshots.

Adoption by Specialty, Age, and Demographics

Neurology shows the highest adoption at 64%, with gastroenterology at 61% and internal medicine at 60%, indicating stronger uptake in certain fields. Several other specialties, including family medicine and cardiology, also surpass the 50% threshold, demonstrating that AI use is spreading across both primary care and specialized areas.

Gender differences appear modest, with men at 57% adoption compared to 49% for women. The report does not provide detailed age breakdowns, but the consistency across groups suggests that age is not a major dividing factor in this dataset. These patterns can help organizations target training resources to specialties with lower rates to balance implementation.

A limitation here is that the survey covers only 15 specialties, so results may not extend to less common fields. The data are self-reported, which means they reflect stated behavior rather than observed usage. A common pitfall is assuming that higher adoption in neurology means the tools are more effective there without considering the specific use cases prevalent in that specialty.

The widespread nature across specialties points to AI's versatility in handling tasks like literature review and documentation that apply broadly. In practice, this means institutions can draw from successful implementations in high-adoption areas to inform strategies elsewhere. The slight gender difference may relate to differences in practice patterns or tool accessibility, though the report does not specify the exact reasons.

Specialty variations likely stem from the alignment between AI capabilities and the dominant workflows in each field, such as diagnostic support in neurology. Organizations reviewing these figures should examine whether their own specialty mix matches the surveyed distribution before setting targets. The absence of strong age-based splits suggests that experience level plays a smaller role than access to suitable tools and institutional encouragement.

Demographic consistency also implies that outreach efforts can focus on workflow fit rather than tailoring messages by age or gender. This approach avoids unnecessary segmentation while still addressing the modest differences observed. One error in application is generalizing the neurology rate as a universal goal without adjusting for the distinct demands of other specialties like pediatrics, where accuracy concerns run higher.

Common Use Cases and Frequency

Literature search leads the use cases at 35% in early 2026, up from 22% in earlier data, while voice-based and ambient documentation tools stand at 29%, up from 20%. These increases show that both research support and administrative relief are driving adoption as tools improve.

Among users, 69% engage with AI daily or more, indicating that once adopted, the tools become habitual rather than occasional. This frequency supports the idea that AI addresses ongoing needs in clinical workflows. Physicians evaluating tools should prioritize those with proven growth in these areas to maximize relevance.

The report's data on use cases are limited to the most common ones, without exhaustive lists of all possible applications. Self-reporting may inflate perceived frequency if respondents recall positive experiences more readily. An error to avoid is selecting tools based solely on popularity without assessing fit for the specific clinical environment.

Voice tools, for instance, appear to gain traction because they reduce typing time during patient interactions. This mechanic allows for real-time documentation that integrates into existing processes. Consider a hypothetical case where a physician uses ambient documentation to capture notes during an exam, freeing attention for the patient. The growth in these use cases suggests that future tools will likely build on these foundations for even higher engagement.

The shift toward daily use among adopters reflects successful matching of tool functions to repeated tasks, such as searching current literature for treatment updates. This pattern can guide procurement decisions by highlighting which applications deliver sustained value. Limitations arise because the survey periods capture snapshots rather than continuous tracking, so seasonal variations in workload might influence reported frequencies.

Physicians can use the growth rates in literature search and documentation tools as indicators of maturing technology that warrants investment in training. A typical mistake involves adopting multiple tools simultaneously without prioritizing the ones showing the strongest upward trends in the data. The 69% daily usage rate among users further suggests that integration succeeds when tools reduce friction in high-volume activities rather than adding new steps.

Reported Benefits and Early Impacts

Among AI users, 75% report reduced administrative burden and improved job satisfaction, while 69% note better patient care and outcomes. These benefits align with the excitement factors where 69% cite reduced administrative workload, 67% improved work-life balance, and 50% greater job satisfaction.

The data also show 49% of users experiencing greater capacity for new patients, with 40% already seeing increased time for patient care. Estimates suggest 1-5 hours per week reallocated from admin tasks. Organizations can use these reported gains to build business cases for AI investment by focusing on workload reduction metrics.

These benefits are self-reported perceptions from the past year, so they may not capture long-term effects or objective measures like error rates. The survey does not link specific tools to particular benefits, leaving room for variation based on implementation quality. A typical mistake is expecting immediate results without allowing time for physicians to adapt to the new tools.

The connection between reduced admin burden and higher satisfaction appears direct, as physicians gain time for core clinical work. This can lead to better retention and reduced burnout over time. The 88% who believe AI can reduce burnout further supports the potential for these early impacts to scale with wider use.

Reallocation of 1-5 hours weekly can translate into meaningful capacity gains when aggregated across a practice group, though individual results depend on baseline admin loads. The alignment between reported benefits and excitement factors indicates that physicians value efficiency improvements that directly affect daily experience. Limitations include the lack of control groups, meaning some perceived gains might stem from concurrent workflow changes unrelated to AI.

Administrators assessing these impacts should pair the percentages with internal metrics on patient throughput to validate the self-reported capacity increases. A common error is focusing only on satisfaction scores while neglecting to measure actual time shifts, which could lead to overinvestment in tools that deliver less tangible workload relief. The 69% reporting better outcomes suggests that freed time often redirects toward clinical activities rather than remaining unused.

Barriers, Concerns, and Institutional Context

Accuracy and reliability of AI outputs tops the concerns at 71%, consistent across all age groups and all 15 specialties, with pediatricians at 78%. Legal and regulatory uncertainty follows as another significant barrier that affects confidence in deployment.

Nearly half of physicians describe their institution's AI decision-making as still evolving, with only 8% finding it clear and understood. This gap between individual adoption and institutional clarity can create uncertainty in how to proceed with new tools. Administrators should prioritize developing transparent policies to align with the high individual interest levels.

The survey data on barriers are based on multiple-choice selections, which may not capture nuanced views or emerging issues. The focus on 15 specialties means barriers in other areas might differ. One common error is to implement AI without addressing accuracy concerns through validation steps, leading to underutilization despite initial interest.

The evolving institutional processes highlight the need for clear guidelines on data privacy and liability. In a conditional example, a hospital might form a committee to review AI tools before widespread rollout to mitigate regulatory risks. This approach can help bridge the gap where only 8% currently see clear policies.

The 71% accuracy concern remains the dominant issue because it directly affects clinical decision-making and patient safety perceptions. This consistency across demographics suggests that technical improvements in model reliability would address a broad segment of hesitation. Institutional clarity at only 8% indicates that many organizations are still defining governance structures, which slows coordinated adoption even as individual use rises.

Physicians navigating these barriers can start by documenting specific accuracy issues encountered in their workflows to inform institutional discussions. A pitfall to avoid is assuming that legal uncertainty will resolve without active advocacy for clearer regulations at the organizational level. The high concern rate in pediatrics may reflect greater caution around vulnerable populations, requiring tailored validation protocols in those settings.

Future Interest and Potential Workload Impacts

88% of physicians believe AI can reduce burnout and improve job satisfaction, while 91% see it freeing time for patient care by cutting administrative workload. Among current users, 40% have already experienced increased patient care time, with many estimating 1-5 hours weekly reallocated.

This expressed interest extends to expanded applications in documentation, literature review, and decision support. The high percentages indicate strong demand for tools that build on current use cases. Healthcare leaders can leverage this to advocate for expanded AI access within their organizations.

These beliefs are forward-looking and based on current experiences, so actual outcomes may vary with tool quality and training. The report does not provide data on long-term impacts beyond the survey period. A pitfall here is overestimating time savings without accounting for the learning curve associated with new AI systems.

The potential for 1-5 hours reallocation per week could compound across a practice, leading to meaningful capacity increases. This aligns with the excitement around improved work-life balance. Physicians interested in future tools should focus on those that target the top barriers like accuracy to realize these projected benefits.

The 91% expectation of freed time for patient care ties directly to the administrative workload reductions already reported by users. This forward projection can guide development priorities toward features that extend current documentation and search capabilities. Limitations stem from the absence of longitudinal tracking, so sustained interest may depend on continued accuracy improvements that address the 71% primary concern.

Organizations planning expansions can use the 88% burnout reduction belief as a metric for evaluating new tool proposals against physician feedback. An error in this area involves launching additional applications without first resolving institutional policy gaps that leave most physicians uncertain about approval processes. The existing user benefits in job satisfaction provide a foundation for scaling interest into measurable workload shifts over time.

Methodology and Scope of the Survey

The survey involved 3,151 US physicians from 15 specialties and combined data from two periods in 2025 and 2026. Doximity membership covers over 85% of US physicians, though the sample may skew toward those using the platform regularly.

Results may not fully represent all US physicians or rarer specialties, and the ARISE report provides only secondary context on clinical AI without physician survey data. All findings are self-reported perceptions without independent verification of usage or outcomes. The core data remain relevant as of the report's release in 2026, with no alterations from later updates noted.

Physicians and administrators reviewing these results can compare their local adoption rates to the 54% benchmark and assess institutional policy clarity against the 8% figure for clear processes. This comparison offers a starting point for identifying areas where additional support or policy development could accelerate effective AI integration in clinical practice.

The two-period design allows observation of change over roughly nine months, which helps distinguish temporary spikes from sustained trends. Self-reporting introduces recall bias, particularly for frequency and benefit estimates that rely on subjective assessment. The 15-specialty scope excludes fields with potentially different adoption dynamics, such as those with lower patient volumes or unique regulatory constraints.

Secondary sources like the ARISE report add context on broader clinical AI developments but lack the direct physician input that forms the core of these findings. When applying the data, organizations should supplement with internal audits to account for sample skew toward platform users. A final consideration is that the absence of verified long-term outcome data means the reported benefits and barriers represent current perceptions that may evolve with newer tool versions.