Abstract
Objective
To evaluate public perspectives on artificial intelligence (AI) in healthcare in the USA and assess variation in trust, comfort and perceived fairness across demographic groups.
Methods and analysis
This cross-sectional survey of US adults was conducted between 21 November and 4 December 2025, via YouGov, a private public opinion research firm, using a non-probability-based online panel. The sample included 500 general population respondents and an oversample of 750 adults aged 60+. Survey weights were applied to approximate the US adult population. Responses were measured using Likert-type or categorical scales and analysed using survey-weighted statistical tests with false discovery rate (FDR) correction for multiple comparisons. Primary outcomes included comfort with healthcare providers using AI, attitudes towards AI-generated second opinions, prior use of AI for healthcare, relative trust in AI versus clinicians for health information, preference for AI chatbots versus providers in explaining care plans and perceived effects of AI on healthcare fairness.
Results
Among weighted respondents (n=500), 34.0% reported prior healthcare AI use, which was more common among college graduates than those without a high school diploma (46.5% vs 22.9%). Only 25.4% reported being comfortable or very comfortable with their provider using AI, while 43.0% were uncomfortable and 31.5% were neutral. Regarding information trust, 44.5% trusted AI much less than clinicians; 88.4% reported a provider would better help them understand their care plan than a chatbot. Prior AI use was associated with more favourable attitudes towards laboratory/imaging interpretation and AI-assisted referrals. Differences by political affiliation, insurance, health status and age were not statistically significant after FDR correction.
Conclusions
AI implementation appears to be outpacing public trust. Higher educational attainment was the most consistent predictor of favourable attitudes. Notably, one-third of respondents were neutral across outcomes, suggesting public attitudes are potentially modifiable. These findings underscore the need for transparent deployment and targeted trust-building strategies to support equitable AI adoption in healthcare.
Keywords: Artificial intelligence; Decision Making, Computer-Assisted; Delivery of Health Care; Health Communication; Patient-Centered Care
WHAT IS ALREADY KNOWN ON THIS TOPIC.
WHAT THIS STUDY ADDS
In a nationally weighted cross-sectional survey of 1250 US adults, fewer than one-quarter were comfortable with clinicians using AI in their care. Prior AI use and higher educational attainment were the most consistent predictors of favourable attitudes; differences by age, politics, insurance and health status were non-significant after correction for multiple comparisons.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
AI deployment in healthcare is outpacing public acceptance, underscoring the need for transparent implementation and targeted trust-building strategies to support equitable adoption across diverse patient populations.
Introduction
Artificial intelligence (AI) is rapidly reshaping medicine across clinical, administrative and patient-facing domains1,2 yet its successful integration into practice depends not only on technical performance but on public trust and acceptance.3–6 Patients consistently report concerns about depersonalisation of care, opaque decision-making, data security and erosion of autonomy,7 while expressing greater comfort when AI is framed as a clinician-directed support tool rather than a replacement.8–11 Acceptance also appears to vary across demographic groups, including by age, education, race and ethnicity, and prior experiences with healthcare and technology.12,13 If certain groups are less willing to engage with AI-enabled care, implementation may risk exacerbating existing disparities.14,15
Despite growing interest, the existing evidence base has significant limitations. Many foundational surveys predate widespread public exposure to generative AI tools such as ChatGPT, limiting their contemporary relevance.16 Prior studies have examined attitudes within narrow clinical contexts or specific patient populations, and relatively few use nationally representative samples to examine variation across demographic and social groups.10 Recent work has begun to address this gap. A 2025 nationally representative survey of US adults by Nong and Platt found that most respondents had low trust in their health systems to use AI responsibly or protect them from AI-related harm, with trust levels closely tied to patients’ broader trust in the healthcare system.17 A 2025 multinational study of attitudes towards AI in healthcare and diagnostics by Busch et al18 found significant variation by gender, health status and technological literacy, and preferences for explainable AI and physician-led decision-making. Yet Busch et al only surveyed hospital patients, limiting generalisability to the USA, while Nong and Platt’s survey was fielded in mid-2023, prior to the widespread adoption of consumer-facing AI tools.
To address these gaps, we conducted a cross-sectional, national survey of US adults to characterise perspectives on AI in healthcare and quantify variation by key demographics, including age, education, political orientation and race. Understanding where public attitudes stand today, and which populations remain most sceptical, is essential to ensuring that the rapid deployment of clinical AI advances, rather than undermines, equitable care.
Materials and methods
Participants for this cross-sectional, national survey were recruited through YouGov, an international public opinion research firm that administers opt-in, non-probability online surveys and applies sample matching and poststratification weighting to approximate the US adult population.19 YouGov’s matched sampling performs favourably compared with other online panels and, in some cases, approximates probability-based samples.20,21
Eligible participants were US residents aged 18 years or older and active members of the YouGov panel. Survey administration occurred between 21 November and 4 December 2025. A total of 500 respondents were included in the general population cohort, with an additional oversample of 750 adults aged 60 years or older to support age-stratified analyses. Adults aged 60 and older were oversampled because they experience higher chronic disease burden and frequent healthcare touchpoints, ensuring sufficient representation to characterise perspectives in this high-utilisation demographic. Overall, a total of 686 respondents from the general population and 1068 adults aged 60 years or older responded to the survey invitation; of these, 542 and 888, respectively, completed the surveys. Both samples went through data quality control checks (including automated checks on completion speed, response consistency and attention-check questions to exclude invalid responses) and were matched down to 500 respondents from the general population and 750 respondents aged 60 years or older with the following process. Respondents were matched on gender, age, race and education to sampling frames derived from the 2023 American Community Survey (ACS) 1-year sample,22 using stratified samples from the full population and subsamples of adults aged 60 years or older and women aged 18 to 40 years. Within-stratum selection used weighted sampling with replacement; matched cases were then weighted using propensity scores and post-stratified on 2024 presidential vote choice and a four-way cross of gender, age, race and education. Weights were trimmed and normalised to prevent undue influence from individual respondents. Our primary analysis was conducted in the general population cohort, with age-based comparisons incorporating the oversample; both analyses incorporated the weighting approach described above.
Survey instrument and measures
Participants completed a 58-item questionnaire developed by clinician, health services and survey scientists (online supplemental file 2). Development of selected survey items was also informed by patient perspectives obtained through prior Patient and Family Advisory Councils (PFACs) activities and patient focus groups conducted by members of the study team. Key survey domains included health experiences, general comfort with AI and AI implications for diagnosis, data, communication, policy and ethics in healthcare. Survey items assessed comfort, trust, perceived usefulness, fairness and preferences regarding AI-supported healthcare, as well as demographic characteristics and prior use of AI for health-related purposes. Perceived fairness was defined as respondents’ beliefs surrounding similar quality of care for all patients regardless of things like background or location.
For this initial quantitative analysis, primary outcomes were prespecified and included: comfort with healthcare providers using AI; attitudes towards physicians using AI for second opinions; prior use of AI for healthcare; relative trust in AI versus clinicians for health answers; perceived ability of AI chatbots versus healthcare providers to support understanding of health plans; and perceived effects of AI on healthcare fairness. Responses were recorded using ordinal Likert-type scales or categorical options. Secondary outcomes evaluated trust in AI for reading laboratory and medical imaging results; comfort with AI-supported medical imaging; preferences for automatic AI-initiated referrals versus clinician review; perceived usefulness of AI chatbots for questions; chatbot support for care plan understanding; trust in AI for simple health questions; and attitudes towards AI-assisted prioritisation of patients in emergencies. Prespecified exposure variables included age, political party affiliation, self-rated general health, insurance status and educational attainment. Prior AI use for healthcare purposes was examined as an additional stratification variable for secondary analyses. Geographic variables (ZIP code and state) were collected to support exploratory visualisation of regional variation in comfort with AI. Additional survey items including exploratory use cases and rich qualitative responses were collected for separate investigations.
Data collection and management
Participants were invited to complete the survey through email or in-app notification distributed by YouGov, and the study description was intentionally neutral to minimise response priming. Respondents completed the survey in exchange for standard panel incentives (points redeemable for gift cards or cash). Survey weights were provided by YouGov to support population-representative inference.
Statistical analysis
Participant demographics were summarised using weighted counts and percentages for categorical variables and medians with IQRs for ordinal outcomes. Primary analyses compared six prespecified primary outcomes across five demographic exposures: age, political party affiliation, self-rated health, insurance status and education. For binary outcomes, group differences were assessed using a Pearson χ² (or Fisher’s exact) test when cell sizes were small. For ordinal outcomes, a Wilcoxon rank-sum (binary exposures) or Kruskal-Wallis (≥3-group exposures) test was applied. Effect sizes were quantified using standardised mean differences (SMDs) with 95% CIs. SMD reflects the magnitude of differences in characteristics between groups and the magnitude of effect is described as small, medium or large for SMD values of 0.2, 0.5 or 0.8, respectively.23 To account for multiple comparisons, false discovery rate (FDR) was controlled using the Benjamini-Hochberg procedure24 within prespecified families of tests. Both raw p values and FDR-adjusted q-values were evaluated at <0.05 for statistical significance. Secondary analyses evaluated AI ‘use case’ outcomes stratified by age group and prior AI use for healthcare. Geographic variation in comfort with providers using AI was visualised using choropleth maps by 9 US Census divisions as defined by the US Census Bureau.25
All analyses incorporated survey weights and design variables provided by YouGov to produce population-representative estimates and valid uncertainty measures. All analyses were conducted using complete case data; all outcomes had no missingness and did not require any imputation. All analyses were performed in SAS V.9.426 and the survey package in R V.4.1.1.26
Patient and public involvement
Patients and the public were not directly involved in the formal design, recruitment, conduct or dissemination of this survey. However, the survey domains and questions were informed by qualitative insights gathered from PFACs and prior patient focus groups exploring perceptions of healthcare AI and clinical chatbots.
Results
Participant characteristics
After application of survey weights, the general population cohort had a mean age of 48.0 years (SD, 19.9) and was approximately evenly split by gender (49.0% male; 51.0% female). The weighted sample was predominantly White (62.7%), followed by Hispanic (13.8%), Black (11.6%) and Asian (4.9%) respondents. Educational attainment ranged from a high school diploma or less (37.9%) to a 4-year college degree or higher (32.7%). Insurance coverage was most commonly employer-sponsored or union-sponsored (37.1%), Medicare (24.3%) or Medicaid (20.3%), while 13.2% reported being uninsured. Politically, 27.6% identified as Democrat, 27.4% as Republican, 33.4% as independent and 11.6% as other. Full demographic characteristics are presented in table 1.
Table 1. Participant demographics for general population cohort.
| Total N=500 Unweighted n (%) |
Total N=500 Weighted n (%) |
|
|---|---|---|
| Age in years, mean (SD) | 48.6 (17.7) | 48.0 (19.9) |
| Gender | ||
| Male | 247 (49.40) | 245 (48.98) |
| Female | 253 (50.60) | 255 (51.02) |
| Race | ||
| Asian | 25 (5.00) | 24 (4.87) |
| Black | 54 (10.80) | 58 (11.58) |
| Hispanic | 60 (12.00) | 69 (13.79) |
| Native American | 3 (0.60) | 6 (1.27) |
| Middle Eastern | 2 (0.40) | 1 (0.26) |
| White | 329 (65.80) | 313 (62.65) |
| Two or more races | 17 (3.40) | 19 (3.87) |
| Other | 10 (2.00) | 10 (1.71) |
| Education | ||
| No high school | 24 (4.80) | 31 (6.11) |
| High school graduate | 159 (31.80) | 159 (31.81) |
| Some college | 98 (19.60) | 101 (20.24) |
| Associate degree (2 years) | 46 (9.20) | 46 (9.19) |
| Bachelor’s degree (4 years) | 107 (21.40) | 103 (20.65) |
| Postgraduate/advanced degree | 66 (13.20) | 60 (12.00) |
| Insurance* | ||
| Current or former employer or union of yours or spouse’s | 191 (38.20) | 186 (37.12) |
| Insurance company | 45 (9.00) | 44 (8.82) |
| Medicare | 119 (23.80) | 121 (24.28) |
| Medicaid | 100 (20.00) | 101 (20.26) |
| Tricare | 12 (2.40) | 13 (2.50) |
| Veterans Affairs | 11 (2.20) | 10 (2.06) |
| Indian Health Service | 4 (0.80) | 7 (1.43) |
| Other | 22 (4.40) | 22 (4.42) |
| None | 62 (12.40) | 66 (13.18) |
| Marital status | ||
| Married | 221 (44.20) | 217 (43.36) |
| Separated | 9 (1.80) | 8 (1.58) |
| Divorced | 64 (12.80) | 67 (13.45) |
| Widowed | 28 (5.60) | 26 (5.19) |
| Never married | 156 (31.20) | 161 (32.30) |
| Domestic/civil partnership | 22 (4.40) | 21 (4.12) |
| Employment status | ||
| Full time | 189 (37.80) | 177 (35.34) |
| Part time | 50 (10.00) | 55 (11.11) |
| Unemployed | 90 (18.00) | 93 (18.54) |
| Retired | 115 (23.00) | 112 (22.38) |
| Permanently disabled | 31 (6.20) | 34 (6.83) |
| Student | 15 (3.00) | 18 (3.64) |
| Other | 10 (2.00) | 11 (2.16) |
| Family income | ||
| Less than US$10 000 | 33 (6.60) | 35 (6.95) |
| US$10 000–US$49 999 | 160 (32.00) | 169 (33.80) |
| US$49 999–US$99 999 | 147 (29.40) | 144 (28.74) |
| US$99 999–US$199 999 | 103 (20.60) | 92 (18.45) |
| US$199 999–US$249 999 | 9 (1.80) | 8 (1.63) |
| US$249 999 or more | 7 (1.40) | 8 (1.62) |
| Prefer not to say | 41 (8.20) | 44 (8.81) |
| Children under 18 in household | ||
| Yes | 113 (22.60) | 104 (20.81) |
| No | 387 (77.40) | 396 (79.18) |
| Home ownership | ||
| Own | 347 (69.40) | 348 (69.71) |
| Rent | 147 (29.40) | 143 (28.53) |
| Other | 6 (1.20) | 9 (1.76) |
| Voter registration status | ||
| Yes | 412 (82.40) | 410 (82.09) |
| No | 79 (15.80) | 81 (16.12) |
| Don’t know | 9 (1.80) | 9 (1.79) |
| 3 point political party | ||
| Democrat | 138 (27.60) | 138 (27.54) |
| Republican | 137 (27.40) | 143 (28.55) |
| Independent | 167 (33.40) | 167 (33.34) |
| Other | 29 (5.80) | 27 (5.49) |
| Not sure | 29 (5.80) | 25 (5.08) |
| 2024 election vote | ||
| Kamala Harris | 140 (28.00) | 136 (27.18) |
| Donald Trump | 148 (29.60) | 143 (28.62) |
| Other | 10 (2.00) | 5 (1.07) |
| Did not vote | 202 (40.40) | 216 (43.13) |
Values are presented as unweighted counts and percentages and survey-weighted counts and percentages calculated using sampling weights to reflect population-representative estimates.
Selections are not mutually exclusive; participants can select more than one option.
Comfort with healthcare providers using AI
Overall, 25.4% of weighted respondents in the general population reported being comfortable or very comfortable with their healthcare provider using AI, 43.0% reported being uncomfortable or very uncomfortable and 31.5% were neutral (figure 1, online supplemental file 1). Comfort did not significantly differ across the other sociodemographic factors examined after FDR correction (SMDs 0.10–0.26; all p>0.05) (online supplemental file 1). Rates were similar across age groups (18–59: 26.2%; ≥60 general cohort: 25.0%; older adult oversample: 23.1%) (online supplemental file 1).
Figure 1. Public attitudes towards artificial intelligence (AI) in healthcare by demographic group.
Regional analysis
Exploratory geographic analysis demonstrated modest variation in comfort with healthcare providers using AI across US census divisions (figure 2). The proportion reporting being uncomfortable or very uncomfortable ranged from 48% in the Middle Atlantic division to 83% in New England, with a range of intermediate values observed throughout the other divisions.
Figure 2. Exploratory analysis of discomfort with healthcare providers using artificial intelligence (AI) across US Census Divisions.
Attitudes towards clinician use of AI second opinions
Regarding clinicians’ use of AI for generating second opinions, 28.9% favoured their use, 27.0% opposed, and 44.0% were neutral. Favourability increased markedly with higher educational attainment (46.3% among college graduates vs 14.6% among those without a high school diploma; p<0.001). No statistically significant differences were observed by political affiliation, health status or insurance status.
Prior use of AI for healthcare
Overall, 34.0% of weighted respondents reported prior use of AI for healthcare purposes. Use increased with education (46.5% among college graduates vs 22.9% among high school graduates; p=0.013). Differences by the other sociodemographic factors were not statistically significant. Prior AI use was descriptively more common among adults aged 18–59 (34.2%) compared with adults aged ≥60 in the general cohort (29.6%) and the older adult oversample (27.3%) (online supplemental file 1).
Relative trust in AI versus clinicians
When asked to compare trust in AI versus clinicians for health information, 44.5% trusted AI much less than clinicians, 40.3% selected the midpoint and 15.2% trusted AI more. Trust in AI versus clinicians did not significantly differ by the examined sociodemographic factors after FDR correction.
Preference for AI chatbot versus healthcare provider
A large majority (88.4%) agreed their healthcare provider would better help them understand their care plan than an AI chatbot (online supplemental file 1). This preference did not significantly differ across demographic groups. The proportion preferring providers ranged from 85.4% among Republicans to 91.2% among Democrats (SMD=0.05; p=0.60). By health status, 92.1% of respondents in poor or fair health preferred providers compared with 87.3% in good to excellent health (SMD=0.28; p=0.20).
Perceived effect of AI on fairness
The largest proportion of respondents (43.7%) reported that AI would have no effect on healthcare fairness. In contrast, 35.7% believed AI would make care fairer, and 20.6% believed it would make care less fair. Educational differences were borderline but did not remain significant after FDR correction (SMD=0.26; p=0.058). Perceptions did not significantly differ by political affiliation (p=0.324), health status (p=0.123), or insurance status (p=0.29). Younger adults (18–59) were more likely to perceive AI as improving fairness (38.2%) than adults aged 60 or older in the general cohort (20.0%) or the older adult oversample (32.3%).
Attitudes towards AI in healthcare
Prior AI experience was consistently associated with more favourable attitudes across applied use cases of AI in clinical care (figure 3). Among prior users, 36.6% trusted AI ‘a lot’ or ‘a great deal’ to interpret laboratory or imaging results compared with 10.4% of non-users, and 55.5% rated AI chatbots as very or extremely useful compared with 17.9% of non-users. Prior users were also more likely to favour automated referrals (24.4% vs 11.9%) and less likely to strongly oppose AI involvement in emergency care (11.0% vs 32.7%).
Figure 3. Applied attitudes towards artificial intelligence (AI) in healthcare by age and prior AI use.
Overall, trust in AI for laboratory or imaging interpretation was moderate across age groups, with 51.7% of adults aged 18–59 reporting at least moderate trust, compared with 47.4% of adults aged ≥60 in the general cohort and 52.8% in the older adult oversample (online supplemental file 1). Most respondents preferred speaking with their primary healthcare provider before being automatically referred for additional testing based on AI findings (81.9% among adults aged 18–59 and 88.8% among adults aged ≥60 in the general cohort). Perceived usefulness of AI chatbots was mixed, with 31.6% of adults aged 18–59 rating them as very or extremely useful and 39.7% as not at all or only slightly useful, with similar patterns across age groups. Opposition to AI-assisted emergency triage was common, reported by 46.0% of adults aged 18–59, 50.0% of adults aged ≥60 in the general cohort, and 54.0% in the older adult oversample.
Discussion
In this national survey, fewer than one in four US adults reported comfort with their healthcare provider using AI, and a large majority placed greater trust in clinicians than in AI for health information. A substantial proportion were neutral across multiple key outcomes, including comfort with provider AI use (31.5%), AI second opinions (44.0%), and relative trust in AI versus clinicians (40.3%). Together, these findings suggest that technical progress in healthcare AI has outpaced the social licence necessary for widespread acceptance.27 Understanding the sources of this hesitation is critical for safe and equitable implementation.28,29
One of the most consistent findings was the strong association between prior AI use for healthcare and more favourable attitudes across applied clinical use cases. Prior users were substantially more likely to trust AI for laboratory and imaging interpretation, rate AI chatbots as highly useful, support automated referral pathways, and oppose AI involvement in emergency triage at lower rates. This aligns with adoption science frameworks and evidence suggesting that direct experience is a primary determinant of technology acceptance and may shift public perceptions in a more favourable direction.30,31
Educational attainment emerged as the most consistent demographic predictor, consistent with prior domestic and international research.32,33 Higher education was associated with greater support for clinician use of AI for second opinions and higher rates of prior AI use, likely reflecting greater digital literacy.34 These findings raise equity concerns: if comfort with healthcare AI is concentrated among more highly educated individuals, clinical deployment may preferentially serve already-advantaged populations.35,36 Equitable integration will require active outreach and co-design processes engaging communities with lower historical exposure to digital health technologies.37
Unlike many contemporary issues that show strong political polarisation,38,39 attitudes towards healthcare AI were similar across political party affiliations, with no significant differences observed after correction for multiple comparisons. This study also did not identify significant differences by insurance status or general health. The absence of differences by insurance status is noteworthy; contrary to prior research suggesting uninsured adults are more likely to turn to AI for health information,40 uninsured individuals in our sample did not show greater acceptance of provider-deployed AI. The intentional oversample of adults 60 and older revealed meaningful heterogeneity: while substantial proportions trusted AI for imaging and laboratory interpretation, older adults showed a stronger preference for clinician oversight before AI-initiated referrals (88.8% vs 81.9%), highlighting a broader tension between patient autonomy and system-level standardisation.41,42
Public opposition to AI in emergency triage was consistent across age groups, suggesting that patients prefer AI in a supportive rather than autonomous decision-making role, a distinction well-documented in prior research.43,44 Views on fairness were largely neutral, though more respondents anticipated benefit than harm. Those with less education were more likely to view AI as reducing fairness, consistent with prior evidence of greater medical mistrust among lower-income and less-educated populations.45
Exploratory geographic analysis suggested modest regional variation in comfort with healthcare providers using AI, though differences were descriptive given small regional sample sizes. These patterns suggest AI acceptability may be shaped by baseline disparities in healthcare access, workforce density, and local infrastructure, where attitudes reflect distinct care constraints. Adequately powered studies are needed to determine whether regional variations reflect true differences in public perception or sampling variability.
With more than half of US adults using consumer-facing large language models daily,46 many patient-facing AI tools operate outside clinical oversight, and current regulatory frameworks do not ensure they are rigorously monitored.47 Trust therefore may depend on whether these systems are validated, limited in scope, and designed with safety guardrails, including escalation to human clinicians when risk is detected.48 Trust may be further strengthened when patients are informed of AI use and provided clear, understandable explanations of its role in their care to support informed consent, helping to address concerns about safety, bias and data use.49,50
Limitations
This study has several limitations. First, its cross-sectional design limits causal inference about the relationships observed between demographic characteristics and attitudes towards healthcare AI. Second, although post-stratification weights were applied to match national demographic benchmarks, the opt-in YouGov panel inherently overrepresents digitally literate and internet-connected adults, entirely excluding non-internet users. Because our findings demonstrate that prior technology experience and digital exposure are strongly associated with more favourable perceptions of healthcare AI, this sampling method may have introduced a systematic bias towards more positive attitudes. Consequently, the low rates of comfort and strong preference for human clinicians observed here may represent conservative estimates; public scepticism may be even more pronounced in the broader, less digitally connected general population. Third, response bias is possible, as those willing to complete a lengthy survey about AI may hold stronger views than the general public. Fourth, sample sizes within several census divisions were small, particularly in New England and East South Central. Consequently, regional variations were not powered for subgroup comparisons and should be interpreted as exploratory hypotheses regarding local healthcare context. Fifth, some survey items used terminology that may have required relatively high health or digital literacy, and the instrument was not formally cognitively pre-tested with members of the public prior to fielding. Finally, survey weights aligned the sample on key demographic characteristics but may not correct for unmeasured differences in AI-relevant attitudes, as self-selection into an online panel likely favours individuals with greater technology engagement than the broader population.
Future research should build on these findings through: (1) dedicated analyses of the older adult oversample to examine heterogeneity in AI-specific concerns, (2) larger, geographically powered studies linking regional attitudes to contextual factors such as healthcare access, provider shortages and broadband availability, (3) co-designed trust-building interventions with digitally underserved and lower-literacy communities and (4) qualitative analysis of the survey’s open-ended responses, which is currently underway in a dedicated follow-up study
Conclusions
Despite rapid advances in healthcare AI, public acceptance remains limited. In this national survey, fewer than one-quarter of respondents reported comfort with clinicians using AI in healthcare, and most continued to place trust in physicians over AI systems. These findings suggest that successful integration of healthcare AI will depend not only on technological capability, but on transparency, public exposure with human oversight and sustained efforts to build patient trust.
Supplementary material
Footnotes
Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Provenance and peer review: Part of a Topic Collection; Not commissioned; externally peer reviewed.
Patient consent for publication: Consent obtained directly from patient(s).
Ethics approval: This study protocol was reviewed by the Stanford University Institutional Review Board (IRB-83837) and determined to be exempt under federal human subjects research regulations (45 CFR 46 / 21 CFR 56). All participants provided electronic informed consent prior to completing the survey.
Data availability free text: De-identified survey data supporting the findings of this study are available from the corresponding author on reasonable request. The survey instrument is available as online supplemental file 2. Data are not publicly deposited due to the terms of the data use agreement with YouGov; however, the authors will consider data sharing requests on a case-by-case basis in accordance with applicable institutional and ethical requirements.
Map disclaimer: The depiction of boundaries on this map does not imply the expression of any opinion whatsoever on the part of BMJ (or any member of its group) concerning the legal status of any country, territory, jurisdiction or area or of its authorities. This map is provided without any warranty of any kind, either express or implied.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Data availability statement
Data are available on reasonable request.
References
- 1.Bajwa J, Munir U, Nori A, et al. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J. 2021;8:e188–94. doi: 10.7861/fhj.2021-0095. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Fahim YA, Hasani IW, Kabba S, et al. Artificial intelligence in healthcare and medicine: clinical applications, therapeutic advances, and future perspectives. Eur J Med Res. 2025;30:848. doi: 10.1186/s40001-025-03196-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Faiyazuddin Md, Rahman SJQ, Anand G, et al. The Impact of Artificial Intelligence on Healthcare: A Comprehensive Review of Advancements in Diagnostics, Treatment, and Operational Efficiency. Health Science Reports . 2025;8:e70312. doi: 10.1002/hsr2.70312. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Howell MD, Corrado GS, DeSalvo KB. Three Epochs of Artificial Intelligence in Health Care. JAMA. 2024;331:242–4. doi: 10.1001/jama.2023.25057. [DOI] [PubMed] [Google Scholar]
- 5.Angus DC, Khera R, Lieu T, et al. AI, Health, and Health Care Today and Tomorrow: The JAMA Summit Report on Artificial Intelligence. JAMA . 2025;334:1650–64. doi: 10.1001/jama.2025.18490. [DOI] [PubMed] [Google Scholar]
- 6.Alelyani T. A validated framework for responsible AI in healthcare autonomous systems. Sci Rep. 2025;15:44432. doi: 10.1038/s41598-025-25266-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Osnat B. Patient perspectives on artificial intelligence in healthcare: A global scoping review of benefits, ethical concerns, and implementation strategies. Int J Med Inform. 2025;203:106007. doi: 10.1016/j.ijmedinf.2025.106007. [DOI] [PubMed] [Google Scholar]
- 8.Brauner P, Glawe F, Liehner GL, et al. Mapping public perception of artificial intelligence: Expectations, risk–benefit tradeoffs, and value as determinants for societal acceptance. Technol Forecast Soc Change. 2025;220:124304. doi: 10.1016/j.techfore.2025.124304. [DOI] [Google Scholar]
- 9.Takita H, Kabata D, Walston SL, et al. A systematic review and meta-analysis of diagnostic performance comparison between generative AI and physicians. npj Digit Med. 2025;8:175. doi: 10.1038/s41746-025-01543-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Han R, Acosta JN, Shakeri Z, et al. Randomised controlled trials evaluating artificial intelligence in clinical practice: a scoping review. Lancet Digit Health. 2024;6:e367–73. doi: 10.1016/S2589-7500(24)00047-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Vo V, Chen G, Aquino YSJ, et al. Multi-stakeholder preferences for the use of artificial intelligence in healthcare: A systematic review and thematic analysis. Soc Sci Med. 2023;338:116357. doi: 10.1016/j.socscimed.2023.116357. [DOI] [PubMed] [Google Scholar]
- 12.Antes AL, Burrous S, Sisk BA, et al. Exploring perceptions of healthcare technologies enabled by artificial intelligence: an online, scenario-based survey. BMC Med Inform Decis Mak. 2021;21:221. doi: 10.1186/s12911-021-01586-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Molnár L, Horváth K. Examining demographic and psychographic factors in AI perception. Social Sciences & Humanities Open. 2026;13:102368. doi: 10.1016/j.ssaho.2025.102368. [DOI] [Google Scholar]
- 14.Health Equity in the Era of Large Language Models. Am J Manag Care. 2025;31:112–7. doi: 10.37765/ajmc.2025.89695. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Budhu JA, Anderson N, Branson C, et al. Health Equity Considerations in the Age of Artificial Intelligence. Neurology. 2025;105:e214356. doi: 10.1212/WNL.0000000000214356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Beets B, Newman TP, Howell EL, et al. Surveying Public Perceptions of Artificial Intelligence in Health Care in the United States: Systematic Review. J Med Internet Res. 2023;25:e40337. doi: 10.2196/40337. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Nong P, Platt J. Patients’ Trust in Health Systems to Use Artificial Intelligence. JAMA Netw Open. 2025;8:e2460628. doi: 10.1001/jamanetworkopen.2024.60628. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Busch F, Hoffmann L, Xu L, et al. Multinational Attitudes Toward AI in Health Care and Diagnostics Among Hospital Patients. JAMA Netw Open . 2025;8:e2514452. doi: 10.1001/jamanetworkopen.2025.14452. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.YouGov Panel mehodolgy. https://yougov.co.uk/about/panel-methodology n.d. Available.
- 20.Graham A, Pickett JT, Cullen FT. Advantages of Matched Over Unmatched Opt-in Samples for Studying Criminal Justice Attitudes: A Research Note. Crime & Delinquency . 2021;67:1962–81. doi: 10.1177/0011128720977439. [DOI] [Google Scholar]
- 21.Kalmoe NP, Mason L. How to Measure Public Support for Political Violence. Public Opin Q. 2026;90:118–51. doi: 10.1093/poq/nfaf069. [DOI] [Google Scholar]
- 22.Bureau UC . CensusGov; [29-Mar-2026]. American community survey (acs)https://www.census.gov/programs-surveys/acs.html Available. Accessed. [Google Scholar]
- 23.Cohen J. Statistical Power Analysis for the Behavioral Sciences. 0. Routledge; 2013. edn. [Google Scholar]
- 24.Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society Series B. 1995;57:289–300. doi: 10.1111/j.2517-6161.1995.tb02031.x. [DOI] [Google Scholar]
- 25.U.S. Census Bureau Census regions and divisions of the united states. https://www.census.gov n.d. Available.
- 26.SAS Institute Inc . SAS/STAT® 9.4 User’s Guide. Cary, NC: 2013. [Google Scholar]
- 27.Chew HSJ, Achananuparp P. Perceptions and Needs of Artificial Intelligence in Health Care to Increase Adoption: Scoping Review. J Med Internet Res . 2022;24:e32939. doi: 10.2196/32939. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Abdelwanis M, Simsekler MCE, Gabor AF, et al. Artificial intelligence adoption challenges from healthcare providers’ perspectives: A comprehensive review and future directions. Saf Sci. 2026;193:107028. doi: 10.1016/j.ssci.2025.107028. [DOI] [Google Scholar]
- 29.Reddy S. Generative AI in healthcare: an implementation science informed translational path on application, integration and governance. Implement Sci. 2024;19:27. doi: 10.1186/s13012-024-01357-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Lee AT, Ramasamy RK, Subbarao A. Understanding Psychosocial Barriers to Healthcare Technology Adoption: A Review of TAM Technology Acceptance Model and Unified Theory of Acceptance and Use of Technology and UTAUT Frameworks. Health Care (Don Mills) 2025;13:250. doi: 10.3390/healthcare13030250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Isaksen AA, Schaarup JR, Bjerg L, et al. Changes in public perception of artificial intelligence in healthcare after exposure to ChatGPT. npj Digit Med. 2025;8:795. doi: 10.1038/s41746-025-02169-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Yakar D, Ongena YP, Kwee TC, et al. Do People Favor Artificial Intelligence Over Physicians? A Survey Among the General Population and Their View on Artificial Intelligence in Medicine. Value Health. 2022;25:374–81. doi: 10.1016/j.jval.2021.09.004. [DOI] [PubMed] [Google Scholar]
- 33.Aras S, Drakos C, Manimangalam V, et al. Influencing public acceptance of artificial intelligence (AI) in healthcare delivery. Front Digit Health . 2025;7:1664345. doi: 10.3389/fdgth.2025.1664345. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Markus A, Carolus A, Wienrich C. Objective measurement of AI literacy: Development and validation of the AI competency objective scale (AICOS) Computers and Education: Artificial Intelligence. 2025;9:100485. doi: 10.1016/j.caeai.2025.100485. [DOI] [Google Scholar]
- 35.Marko JGO, Neagu CD, Anand PB. Examining inclusivity: the use of AI and diverse populations in health and social care: a systematic review. BMC Med Inform Decis Mak. 2025;25:57. doi: 10.1186/s12911-025-02884-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Chinta SV, Wang Z, Palikhe A, et al. AI-driven healthcare: A review on ensuring fairness and mitigating bias. PLOS Digit Health . 2025;4:e0000864. doi: 10.1371/journal.pdig.0000864. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Johnson KB, Horn IB, Horvitz E. Pursuing Equity With Artificial Intelligence in Health Care. JAMA Health Forum . 2025;6:e245031. doi: 10.1001/jamahealthforum.2024.5031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Steiglechner P, Smaldino PE, Merico A. How opinion variation among in-groups can skew perceptions of ideological polarization. PNAS Nexus. 2025;4:gaf184. doi: 10.1093/pnasnexus/pgaf184. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Castelo N, Ward AF. Conservatism predicts aversion to consequential Artificial Intelligence. PLoS ONE. 2021;16:e0261467. doi: 10.1371/journal.pone.0261467. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Montero A, Montalvo J, III, Kearney A, et al. KFF tracking poll on health information and trust: use of ai for health information and advice. 2026. https://www.kff.org/public-opinion/kff-tracking-poll-on-health-information-and-trust-use-of-ai-for-health-information-and-advice/ Available.
- 41.Martinho D, Sobreiro P, Domingues A, et al. Ethical Responsibility in Medical AI: A Semi-Systematic Thematic Review and Multilevel Governance Model. Health Care (Don Mills) 2026;14:287. doi: 10.3390/healthcare14030287. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Kim D, Vegt N, Visch V, et al. How much decision power should (a)i have?: investigating patients’ preferences towards ai autonomy in healthcare decision making. CHI ’24; Honolulu HI USA. May 11, 2024. pp. 1–17. Available. [DOI] [Google Scholar]
- 43.Gundlack J, Thiel C, Negash S, et al. Patients’ Perceptions of Artificial Intelligence Acceptance, Challenges, and Use in Medical Care: Qualitative Study. J Med Internet Res. 2025;27:e70487. doi: 10.2196/70487. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Zhu X, Stroud AM, Minteer SA, et al. Key Information Influencing Patient Decision-Making About AI in Health Care: Survey Experiment Study. J Med Internet Res. 2026;28:e75615. doi: 10.2196/75615. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Shukla M, Schilt-Solberg M, Gibson-Scipio W. Medical Mistrust: A Concept Analysis. Nurs Rep. 2025;15:103. doi: 10.3390/nursrep15030103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Consumer-facing AI can improve public health — but more evidence is needed. Nat Health. 2026;1:159. doi: 10.1038/s44360-025-00052-3. [DOI] [Google Scholar]
- 47.Abulibdeh R, Celi LA, Sejdić E. The illusion of safety: A report to the FDA on AI healthcare product approvals. PLOS Digit Health . 2025;4:e0000866. doi: 10.1371/journal.pdig.0000866. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Kaur D, Uslu S, Rittichier KJ, et al. Trustworthy Artificial Intelligence: A Review. ACM Comput Surv. 2023;55:1–38. doi: 10.1145/3491209. [DOI] [Google Scholar]
- 49.Hurley ME, Lang BH, Kostick-Quenet KM, et al. Patient Consent and The Right to Notice and Explanation of AI Systems Used in Health Care. Am J Bioeth. 2025;25:102–14. doi: 10.1080/15265161.2024.2399828. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Chau M, Rahman MG, Debnath T. From black box to clarity: Strategies for effective AI informed consent in healthcare. Artif Intell Med. 2025;167:103169. doi: 10.1016/j.artmed.2025.103169. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data are available on reasonable request.