Perceptions of doctors disclosing AI use in their online biographies: evidence from an experiment

For better or worse, artificial intelligence (AI) seems to be everywhere, including the health-care space. From the Centers for Disease Control and Prevention (2026) and World Health Organization (2025) now prioritizing AI in the global public health sphere, to individual general practitioners using AI to help with their diagnostic capabilities (Blease et al., 2024), it appears AI is likely here to stay; and it makes sense. In a space that is already often hard to navigate for lay populations (e.g. health literacy), AI can potentially open pathways to assist in uncomplicating the complicated, while maybe even allowing physicians to spend more time with patients via less time completing burdensome paperwork (Čartolovni et al., 2023; Pavuluri et al., 2024). However, what remains to be seen is whether the marketing and mentioning of providers using AI in their own practices provides a resulting competitive edge when it comes to patients’ perceptions of those providers. While its use might yield potential benefits in highly complicated surgical settings (Ali Mohamad et al., 2023) or for improved resource optimization (e.g. scheduling operating rooms more efficiently) for health-care systems (Hennrich et al., 2024), providers’ own disclosure of its use might be viewed by patients as a bridge too far.

A recent, national survey of US adults revealed low levels of trust in their health-care systems to use AI responsibly and in ways that would not harm them (Nong and Platt, 2025). One reason for this could be that health care is primarily a human-centered field, where having empathic providers can lead to improved patient outcomes (Hojat et al., 2023). Conversely, despite AI’s likely integration into health-care systems, a provider indicating they actively use AI might seem to a potential patient to lack this human or empathic touch patients could be seeking in their future care (Esmaeilzadeh et al., 2021). Providers’ use of AI also has the potential to lead to a dehumanization of patient care (Akingbola et al., 2024). As Akingbola et al. (2024) note, “although AI has the potential for the transformation of healthcare, the effect on the erosion of [the] doctor-patient interaction poses a challenge which must be addressed through research” (p. 2). Therefore, by using physicians’ online biographies, a key location patients seek information about future providers (Perrault and Hildenbrand, 2018), this study looks to determine what impact a physician’s disclosure towards using AI in their practice might have on patient perceptions and intentions to visit the provider.

Despite already being a key player in biomedical research (da Silva, 2024) and medical devices (Muralidharan et al., 2024), AI has taken a bit longer to be fully integrated into individual clinical practices, likely because of regulatory and reimbursement hurdles (Davenport and Glaser, 2022). However, a recent survey of practicing physicians conducted by the American Medical Association revealed that about three-quarters see AI as having at least some advantages in caring for patients, with about a similar number indicating they currently use AI in their practices for things such as assisting in diagnoses and generating discharge and care plans (AMA, 2026). Some physicians even feel that increased use of AI to perform more administrative tasks has the potential to free up more time to converse and meet with patients (Čartolovni et al., 2023).

Yet, it is essential to acknowledge that there are some concerning elements that the use of AI in clinical settings could lead to. For example, its automated use in insurance decisions could lead to denials of care (Ross and Herman, 2023), as well as subjecting patients to data privacy concerns (AMA, 2024). Additionally, AI has the potential to lead to a rise in new medical paternalism where the ambiguity in how AI tools reach decisions “returns patients to a paternalistic model of the physician patient relationship, in which physicians make medical decisions for their patients without disclosing information which is relevant for decision making” (Xu and Shuttleworth, 2024, p. 55). In other words, the use of AI can potentially take away patients’ ability to be fully informed about their care (Xu and Shuttleworth, 2024). Additionally, Longoni et al. (2019) note that people’s belief that AI might not be fully attuned to the unique circumstances of humans (i.e. uniqueness neglect) may make consumers reticent to adopt AI in their own healthcare.

Similarly, perceptions of artificial intelligence use in the context of doctor–patient consultations may be influenced by algorithm aversion. Algorithm aversion is the tendency of people to prefer human recommendations to those from algorithms (Sunstein and Gaffe, 2025). Mahmud et al. (2022) identify in their systematic review several influencing factors such as algorithms’ black box nature, their accuracy, and their supporting (or replacing) role in human decision-making. For example, in medicine, people may prefer human decision making supported by algorithms rather than algorithms replacing humans entirely (Bigman and Gray, 2018). One location where patients could get a realistic preview of the kind of care they may receive from potential providers, and how those providers might engage with AI tools, is via their online biographies provided by health-care systems.

ZocDoc recently found that more than half of patients they surveyed went to an initial doctor’s appointment only to find the doctor was not the best fit for themselves; and about 1-in-4 providers indicated they also met new patients who did not seem to be a proper fit for their clinical expertise (Zocdoc, 2024). In short, having initial medical visits where there is not an ideal patient-physician fit or connection wastes significant time and resources for all parties involved. Therefore, it comes as no surprise that online biographies are becoming an increasingly important marketing tool to help patients make the best care decisions for themselves (Reichwald, 2025). Perrault and Hildenbrand (2018) found that providers’ online biographies are the top location patients seek information about future health-care providers, as they can provide a realistic glimpse into the kind of care the clinician might provide through philosophies of care and even short video introductions (Perrault, 2021; Perrault, 2016).

It therefore makes sense that providers who use AI in their medical decision-making might indicate in their biographies that they use AI as part of their care priorities/interests. Signaling in a biography the ways the doctor uses to make their decisions can help a prospective patient determine if the doctor’s care style matches the patient’s own care preferences. For example, a doctor who indicates they use AI might signal to a patient that they have a more paternalistic approach to medicine (Xu and Shuttleworth, 2024), whereas someone who says they like to consult with their physician colleagues might signal they take a more deliberative approach to care (Emanuel and Emanuel, 1992). Thus, this research was interested in answering the following two research questions:

A 2 (collaboration strategy: with AI vs with physician colleagues) × 2 (sex of provider: male vs female), between-subjects experiment was conducted to answer our research questions. The experiment was approved by the Purdue University Institutional Review Board, and data collection occurred via the online participation pool within that university’s school of communication between April–August 2025.

Upon consenting, participants read a vignette where they were asked to imagine themselves having moved to a new city after graduating and not feeling well after having a meal at a nearby restaurant. After still feeling unwell for the next few days (e.g. having pain that makes sleeping/working difficult), they decided it was time to seek a medical appointment, and that they went online to find a provider in their insurance network to visit. See  Appendix 1 for the vignette. They were then randomly shown one of the four biographies and asked to complete questions related to the biography they just viewed. To ensure participants viewed the biography, they could not advance until at least 10 s had elapsed on the page showing the biography.

The creation of the doctor’s biography was informed by prior content analyses of provider biographies (Perrault et al., 2021; Perrault and Smreker, 2013) and included their educational background, residency information, certifications, professional and personal interests and their philosophy of care. The information remained constant in all conditions, except for changing the doctor’s first name (Jane vs James) and indicating within the biography their interests in consulting with either AI or their colleagues within their practices. See  Appendix 2 for the exact text of the biographies. All biographies were the exact same word length.

After viewing the biography, participants were asked single item questions about their likelihood of making an appointment with the doctor (1 = extremely unlikely, 7 = extremely likely), as well as then answering a dichotomous yes/no question regarding whether they would make an appointment.

Perceived expertise was assessed by averaging six, seven-point, semantic differential items adapted from McCroskey and Teven (1999) and Ohanian (1990), asking participants to indicate whether the doctor they just read about was: not an expert/expert, inexperienced/experienced, incompetent/competent, unqualified/qualified, unskilled/skilled, stupid/smart.

The average of four, seven-point semantic differential word pairs adapted from Richmond et al. (1998), asked participants to rate the kind of medical care they would receive from the doctor they read about as: impersonal/personal, uncaring/caring, unconcerned/concerned, and unsatisfactory/satisfactory.

The average of six, seven-point semantic differential items adapted from McCroskey and Teven (1999) and Ohanian (1990), asked participants to rate the doctor on the following word pairs: untrustworthy/trustworthy; phony/genuine; insincere/sincere; unreliable/reliable, dishonest/honest; and undependable/dependable.

Participants’ perception about the approachability of the provider was assessed as the average of five Likert-scaled items (1 = strongly disagree, 7 = strongly agree) from Hackett and Jacobson (1995). The items asked whether participants thought the doctor would make them comfortable to raise issues they had, would take a real interest in them, would understand them as a person, would not be stressful to be with, and would understand all the health problems they had.

The degree of likability participants felt towards the provider they read about was measured as the average of four Likert-scaled items (1 = strongly disagree, 7 = strongly agree) adapted from Jayanti and Whipple (2008). Participants assessed whether the doctor they read about was friendly, pleasant, likeable, and seemed like a nice person.

The average of three semantic differential items adapted from Richmond et al. (1998) asked people to rate how displeased/pleased, dissatisfied/satisfied, and uncomfortable/comfortable they would be if they had a visit with the doctor they just read about.

Participants (n = 466) ranged in age from 18 to 42 years (M = 19.75 and SD = 1.9). Most identified as Caucasian (54.1%), followed by Asian (26.8%) and Hispanic (7.9%). About a third (35.6%) were in their first year in school, followed by second-year (28.8%), third-year (19.3%) and fourth-year (14.6%; Table 1).

RQ1 was interested in determining how a doctor’s disclosure of using AI in their care strategy might impact prospective patients’ perceptions of that provider. A two-way MANOVA was conducted with physician sex (James/Jane) and consultation strategy (AI/colleagues) as the between-subjects variables, and the six perceptions of perceived expertise, anticipated quality of medical care, trustworthiness, approachability, liking, and anticipated satisfaction as the dependent variables. The overall effect of physician sex F (6, 457) = 1.71 and p = 0.321, on the dependent variables was nonsignificant, as was the interaction between sex and consultation strategy (p = 0.507). However, a significant overall main effect was found for consultation strategy on the dependent variables, F (6, 457) = 4.70, p < 0.001, partial η² = 0.058 and Wilks’ Λ = 0.942. Therefore, we proceeded to investigate this main effect on the various dependent perceptions.

Participants perceived the doctor who indicated a consultative preference for AI as less expert (p = 0.01), trustworthy (p < 0.001) and approachable (p < 0.001). Participants also perceived they would receive a lower quality of care (p < 0.001) and reported less patient satisfaction (p < 0.001) with the doctor who mentioned AI than with the doctor who indicated a consultative preference with their human colleagues. There was no difference between conditions on liking (p = 0.11; Table 2).

RQ2 was interested in determining whether a preference for AI in the biographies might have an impact on people’s willingness to want to visit that provider for future medical care. To answer this RQ, a two-way ANOVA was conducted, with intention to make an appointment as the dependent variable. Similar to RQ1, the main effect for both provider sex (p = 0.358) and the interaction (p = 0.371) were non-significant. There was a significant main effect for the impact of consultation strategy on people’s intention to make an appointment, F (1, 462) = 34.56, p < 0.001 and partial η² = 0.070. Those who viewed the biography of the doctor who mentioned consulting with AI had a significantly reduced intention to want to visit that provider (M = 4.82 and SD = 1.65), compared to those who viewed the biography of the provider who mentioned consulting with their colleagues (M = 5.62 and SD = 1.27).

Subsequently, a Chi-Square analysis was conducted crossing the variables of consultation strategy (AI vs colleagues) and decision to visit the provider (yes/no), revealing a significant finding: χ² (1) = 20.65, p < 0.001 and Φ = 0.211. A greater percentage of people who viewed the biography of the AI-consulting provider indicated not wanting to make a future appointment with that provider (32.2%; 74 of 230), than those who viewed the biography of the provider who indicated a preference in consulting with their colleagues (14.4%; 34 of 236). In other words, more than twice as many people indicated not wanting to make a future appointment with the AI-consulting physician compared with the provider who did not mention AI.

Despite many physicians indicating they presently use AI tools in their practices for diagnostic purposes (AMA, 2026; Blease et al., 2024), the findings from this study indicate that prospective patients do not find these physicians as expert, trustworthy, or approachable as human-consulting physicians. Additionally, the participants in this study indicated they believed they would get a lower quality of care and less patient satisfaction from the AI-consulting provider. Most importantly for health-care marketers, this AI-disclosure within the biographies significantly reduced the number of patients who were willing to make an appointment with that provider.

In addition to these significant quantitative findings, an open-ended question was also asked of participants to explain why they decided against wanting to visit the physician whose biography they just saw. While we did not do a formal qualitative analysis of these responses, as that falls outside of the scope of this study’s research questions, the answers people provided offer a little insight into their ultimate decision-making processes. Almost all of the participants who saw the AI-consulting doctor’s biography and indicated not wanting to make an appointment, wrote an open-ended response (71-of-74). Of those 71 responses, 58 of them (or 81.7%) mentioned the doctor’s disclosure of AI-use in the biography as a reason for not wanting to visit the doctor (e.g. “he uses AI” [Participant 54], “I don’t care for the emphasis on AI” [Participant 27], “She mentions the use of AI and that is kind of a turn off as a patient” [Participant 207]). One participant indicated, “Although there very well may be a future where consulting AI is a normalized part of health care, I am strongly dissuaded by the collaboration with AI at the moment. While AI can be useful, I would not want to risk the chance of the AI hallucinating results regarding my medical tests… I would strongly prefer a primary health provider who does not regularly consult with AI” (participant 418). Another participant indicated, “The use of AI in the doctor’s description is concerning. When I see a doctor, I want them to use their expertise to analyze my chart, not rely on a tool that can’t see past the information presented to treat me. This is something that I could do for myself if I needed to” (Participant 78). These open-ended responses echo responses of prior qualitative research regarding concerns toward AI-use in medical care; primarily, that if doctors start indicating they use AI, patients might simply stop going to providers and diagnose themselves – potentially incorrectly – as well as a loss of the human/empathic element needed to provide compassionate care (Čartolovni et al., 2023).

One limitation of this study was the dichotomous/binary nature of the consultative strategy the doctor disclosed using in their practice (i.e. consulting with AI or with colleagues). It is entirely possible that a doctor uses AI as just one of a plethora of tools in their medical toolbox. Therefore, future studies of this type might want to develop biographies that showcase a more combined consultative strategy, where consulting with AI – in addition to consulting with colleagues or the latest peer-reviewed research – is presented to participants.

Another limitation of this study is that the AI use in the biographies was mentioned solely by the physician in the course of doing their own individual jobs. However, in practice, AI is currently being embedded within entire health-care systems, sometimes in unknown or stealth fashions (Gillner, 2024). Future research should investigate what impact the promotion of AI use health-care systemwide might have on prospective patients deciding to select providers within those networks. For example, large networks such as Kaiser Permanente (Yang, 2025) and the Cleveland Clinic (2025) tout their use of AI within their systems – systems that individual physicians are situated within. Would patients have the same reluctance to visit a doctor who works within one of these systems if AI use within the system was presented in biographies, possibly as boilerplate language at the bottom of the biography? Additionally, the mention of AI in the biography was presented in a broad, generalized fashion, rather than identifying a providers’ use of certain, specific AI tools in their practices. Future research might want to investigate which specific AI tools (e.g. ones that lean more assistive versus autonomous) prospective patients would be more comfortable having their providers use.

Furthermore, this study focused on experimentally determining whether disclosing AI as a consultation strategy in providers' biographies impacts prospective patients’ perceptions of expertise, quality of medical care, trustworthiness, approachability, liking and anticipated satisfaction. However, as the Technology Acceptance Model highlights (Davis and Granić, 2024), there are dozens of other individual-level characteristics, antecedents, or moderators that might impact someone’s willingness to choose a doctor who signals using AI in their practices. Future studies might also want to investigate how variables such as perceived loss of empathy versus competence signaling might influence perceptions.

The student sample used in the study is also a limitation. Even though Hanel and Vione (2016) found that student samples were as heterogeneous as the general public on multiple personality variables, caution should still be taken in generalizing study findings beyond the sample studied. Future research should seek to replicate this study’s findings using a more diverse sample, possibly of varying education, age, and socio-economic status.

Finally, this study dealt with someone dealing with an acute health condition. While prior research has found fairly minimal differences between those with acute and chronic conditions and their perceptions towards AI use in their medical care (Esmaeilzadeh et al., 2021), future research might want to test if differences exist when even more serious ailments are considered. For example, for someone battling more complicated, rare, or serious diseases, might a doctor’s AI use be perceived differently?

Even though research finds many physicians are currently using AI within their own practices, the current study indicates that disclosing their use of it within their personal biographies has potentially negative ramifications on how prospective patients view those providers, and their willingness to select them for their personal care. Despite many health-care systems already deploying AI in their networks, this study finds that having AI mentioned in individual provider biographies has the potential to alienate one-third of a system’s prospective patient base. Given that AI is likely here to stay in healthcare for the foreseeable future, additional research needs to continue to be conducted to determine how best to communicate clinicians’ use of AI to patients while they are seeking care to ensure this powerful tool does not solely serve as a negative heuristic for wholesale rejection of care.

Evan K. Perrault (PhD, Michigan State University, 2014) is an Associate Professor of Health Communication in the Brian Lamb School of Communication at Purdue University.

Gladys M. Momade (MS, Illinois State University, 2023) is currently a Doctoral Candidate in the Brian Lamb School of Communication at Purdue University.

Eylül Yel (PhD, Purdue University, 2026) is currently an Assistant Professor of Communication Studies at Ferris State University, but was a doctoral candidate at Purdue University during the time of this research.

This study was approved as exempt by the Purdue University Institutional Review Board (protocol number: 2025–533).

Please carefully and completely read the following paragraphs.

Imagine that you are the patient in the following scenario.

After graduating from school, you moved to a new city. One evening, you decide to eat dinner at a nearby restaurant. After eating, you start feeling a tight, burning sensation in your stomach. You tried to go to bed, thinking the pain was caused by the spicy food you just ate and that it would go away by morning. However, it becomes even worse the next morning.

You recall that you have experienced this before, and the last time it happened, you went to see a doctor who had to run many tests, including bloodwork and scans, to no avail. You decide to pick up some over-the-counter medicines from the nearest pharmacy, but they only help for a little bit and the pain returns even worse than before. The pain continues for days and starts to distract your daily life. Concentrating on work or sleeping is hard because of the pain, and you are worried it might be something more serious that needs medical attention.

To make an appointment, you go to a health-care system’s website that is within your insurance network to look up a provider you might want to visit.

On the next page you will see the biography of that healthcare provider. Please read the biography with your utmost attention, and then answer the series of questions following the biography.

[Jane/James] G. O’Connor, MD

Physician

Education, Residency and Certifications:

• University of Michigan – Biology, BA (2007)

• Indiana University School of Medicine – Family Medicine, MD (2011)

• Duke University – Residency (2014)

• ABFM (American Board of Family Medicine) certified

Professional Interests and Specializations:

I specialize in primary care. I am particularly interested in preventive medicine and how collaboration with [AI tools/other clinicians] can improve medicine.

What is your philosophy of care?

I believe in a patient-centered approach to medicine. I strive to tailor treatments to meet patients’ needs, give them options, and help them choose the best options for themselves and their unique lifestyles. For example, I am a strong believer in seeking multiple views when developing treatment plans. I regularly consult with [the latest artificial intelligence tools/my team of physician colleagues] to interpret medical records and test results to enhance diagnostic accuracy.

What do you like to do in your free time?

I love spending time with my family and cooking meals together. I also enjoy playing word games and playing soccer with my friends and family on weekends.