If your model of patient acceptance of AI in healthcare assumes that acceptance follows a single trajectory, where patients become more comfortable as AI demonstrates capability and acceptance grows accordingly, the 2020-2025 research on patient perspectives shows the pattern is more structurally specific. Publication volume on patient AI acceptance grew tenfold between 2020 and 2025 (9 papers in 2020 to 102 in 2025). The dominant pattern across the literature: conditional acceptance. Patients endorse AI in assistive roles, not autonomous decision-making, particularly in high-stakes clinical contexts. Trust in AI is clinician-mediated rather than technology-evaluated. Provider endorsement functions as the key determinant of patient acceptance.

Publications on patient perceptions of AI in healthcare, 2020–25:

The methodological observation matters: patient AI acceptance flows through the clinical relationship, not around it. The implication is that AI deployment strategies that treat patient acceptance as a function of the technology itself will under-engage with what actually shapes acceptance.

The literature consolidates around five consistent findings:

The first finding: conditional acceptance is the default pattern. Patients tended to endorse AI in assistive roles rather than autonomous decision-making, particularly in high-stakes clinical contexts (Fee et al., 2025; Allen et al., 2025; Hmido et al., 2025). The condition: AI as augmentation, not replacement. Patients support AI deployment that helps their clinician do better work; they resist AI deployment that replaces clinician decision-making.

The second finding: provider endorsement determines patient acceptance. Berger et al. 2025, Machado et al. 2025, and Nong et al. 2025 (three independent studies) converge on this. When a trusted clinician explains that AI is being used and endorses its use, patient acceptance is high. When AI deployment happens without clinician communication or endorsement, patient acceptance is materially lower. The clinical relationship is the trust transfer mechanism.

The third finding: demographic disparities in acceptance are documented across multiple dimensions. Patients show variation in AI acceptance by age, gender, education, and race (Labinsky et al., 2025; Ogu et al., 2025; Li et al., 2025). The disparities are not uniform: different studies find different patterns depending on context and AI use case. But the consistent observation is that patient AI acceptance varies systematically across demographic groups, with implications for equitable AI deployment.

The fourth finding: preservation of the human relationship is a consistent patient concern. Patients identified the potential loss of empathic care as a primary concern (Carl et al., 2025; Davis et al., 2025). The concern is not about AI capability, since patients may accept that AI is accurate. The concern is about what AI deployment does to the experience of being cared for as a person. Empathic communication, presence, and emotional support are valued; AI deployment that diminishes these is resisted.

The fifth finding: transparency and disclosure of AI use are consistently prioritised. Patients across multiple studies prefer to know when AI is being used in their care, what it is being used for, and how clinical decisions are being made. Emerging disclosure frameworks (Mello et al. 2025, a two-question framework for determining when consent, notification, or neither is required) provide practical guidance.

The methodological observation about the literature itself: medical specialties most studied for patient AI perspectives are general healthcare (70 publications), internal medicine (24), and radiology (16). The geographic distribution is concentrated in the United States, United Kingdom, and Germany. Studies from sub-Saharan Africa, Latin America, and Southeast Asia remain underrepresented. Studies that include children and adolescents as participants rather than drawing solely on parent or caregiver perspectives remain rare. The patient perspective literature is structurally weighted toward Western, adult, hospital-based contexts.

Medical specialties represented in patient-perception studies, 2020–25:

The data observation is methodologically clear: patient AI acceptance is conditional (assistive roles preferred over autonomous), clinician-mediated (provider endorsement is decisive), demographically variable, relationship-preserving (empathic care valued), and transparency-requiring (disclosure prioritised). The patterns are consistent across many independent studies.

Three implications follow for health systems setting AI deployment strategy with patient engagement in 2026.

The first implication: AI deployment communication strategies need to engage with clinical relationships as the primary channel. Patient education materials, hospital webpages, and broad-audience communication can establish awareness, but the operational acceptance happens through individual clinical encounters where clinicians explain AI use. Training clinicians to discuss AI deployment with patients, providing them with talking points and patient-facing materials they can share, and supporting them in handling patient questions are the operationally consequential investments.

The second implication: disclosure and transparency practices need to match the disclosure framework patients expect. The Mello et al. 2025 two-question framework (does AI use carry harm risk? does the patient have an opportunity to express agency in response to disclosure?) provides operational guidance. AI deployments that disclose proactively, even where regulation does not require it, align with patient preference. AI deployments that disclose only when required operate against patient preference and bear the trust cost.

The third implication: AI deployment design should preserve and ideally enhance the human relationship rather than diminish it. The ambient AI scribe deployments documented in a companion analysis illustrate this principle in operation: physicians reported a 58% increase in undivided patient attention at UChicago Medicine, with note-writing handled by AI rather than during the encounter. The AI deployment created more space for human relationship, not less. Other AI deployments that reduce clinician-patient time, replace human communication with AI communication, or otherwise diminish the relationship will produce more patient resistance.

The methodological observation about the limits of current research: the patient perspective literature is dominated by Western, adult, hospital-based studies. Patient perspectives from sub-Saharan Africa, Latin America, and Southeast Asia are underrepresented. Paediatric and adolescent patient perspectives are rare. The findings about conditional acceptance, clinician mediation, and relationship preservation may generalise globally but are not yet verified across the full diversity of healthcare contexts. The methodological gap should be addressed through deliberate research investment in underrepresented populations.

For health systems setting global AI deployment strategy in 2026, the methodological observation recommends a structured approach: treat clinical relationships as the primary AI acceptance channel and invest in clinician training and patient-communication infrastructure accordingly; build proactive disclosure and transparency as standard practice, aligning with patient preference rather than minimum regulation; design AI deployments to preserve or enhance the human relationship, treating relationship preservation as a primary deployment criterion rather than an afterthought; engage with demographic variation through context-aware deployment design; and recognise that the current research is concentrated in Western adult hospital contexts, verifying generalisation elsewhere through deliberate research investment.

The trajectory: through 2026-2028, patient AI acceptance research will continue to expand (102 publications in 2025 likely to exceed 200 by 2027). The conditional, clinician-mediated, transparency-requiring pattern is likely to persist as the dominant finding. Strategic plans that engage with this pattern will produce AI deployments aligned with patient preference. Plans that treat patient acceptance as a function of AI capability alone will be misaligned with the methodological evidence.

The methodological observation has direct operational consequence: AI deployment that operates through the clinical relationship will produce different patient outcomes than AI deployment that operates around it. The data is consistent enough to make this a planning premise rather than an empirical question.