Doctors faced an AI-use penalty after disclosing AI use to patients

By Dr. Priyom Bose, Ph.D.Reviewed by Lauren HardakerOct 5 2026

US adults were less willing to choose, recommend, or follow the advice of family doctors who disclosed using AI, with negative perceptions particularly pronounced among younger consumers.

Study: Influence of physician and consumer demographics on AI-use penalties in primary care: a vignette-based study. Image credit: MUNGKHOOD STUDIO/Shutterstock.com

A recent npj Digital Medicine study examined how physician and consumer demographics influence the "AI-use penalty" in primary care settings.

AI integration and consumer perceptions in healthcare

Artificial intelligence (AI) is becoming increasingly common in healthcare, but its use may also affect how patients view their doctors. When patients have limited information about a physician’s abilities or intentions, knowing that the doctor uses AI could influence their first impressions.

Previous research suggests that this disclosure can work against physicians. Doctors who say they use AI may be viewed less favorably than otherwise similar doctors who say they do not: an effect researchers describe as the “AI-use penalty.” Studies have also found that simply believing AI is involved can reduce trust in medical information, while physicians who disclose using AI may be seen as less competent by both their peers and the public.

It is less clear whether this penalty differs depending on whether the doctor or patient is using AI. Patients already judge physicians differently based on characteristics such as race, gender, and age, while their own backgrounds can also influence how they respond to AI. Previous research, for example, has found differences in attitudes toward healthcare AI between men and women and between Black and White individuals. The researchers therefore set out to determine whether physician and patient demographics could strengthen or weaken the AI-use penalty.

Assessing consumer judgments on physician AI use

Data were collected in November 2025 using Prolific’s “representative sample” feature to approximate the demographic composition of the US adult population based on the 2021 US census. After applying manipulation-check exclusions, the final analytic sample comprised 1,030 participants (females: 520, males: 487, non-binary: 18, and not disclosed: 5), including 123 Black, 72 Asian, 96 Hispanic, 11 Native American, 674 White, 47 other, and 7 preferred not to disclose. More than half of participants reported holding a university degree.

The study employed a 2 × 2 × 2 × 3 factorial design, with physician gender, age, and AI use manipulated between subjects and physician race manipulated within subjects. A “never uses AI” group served as the control. AI use, physician gender, and age were manipulated between subjects to minimize demand characteristics and socially desirable responding, while the order of physician profiles was randomized to minimize carryover and contrast effects.

Participants evaluated three randomized advertisements for family doctors (representing White, Black, and Asian physicians) and were assigned to one of eight groups reflecting combinations of gender, age, and AI use. Measures included attention checks, perceived warmth and competence, and whether participants would choose, recommend, or follow the doctor’s advice. Participants also reported their use of AI tools, perceptions of AI integration in healthcare, and demographic characteristics.

AI-use penalty was stronger among younger consumers

Analysis demonstrated a consistent aversion to physicians who disclosed AI use, with these physicians rated as less warm and less competent. Participants also showed reduced willingness to make or repeat appointments, recommend these physicians, or adhere to treatment plans. Participants were also more likely to seek a second opinion and were willing to travel less far, pay less, and wait less long for an appointment with physicians who disclosed AI use. These outcomes were robust across physician age, gender, and race, indicating that the AI-use penalty operated largely independently of these factors.

Exploratory analyses revealed that the negative effects of disclosing AI use were especially pronounced among participants younger than 45. For perceived warmth, the AI-use penalty also varied by the combination of physician and participant race, and was strongest for White physicians evaluated by White and Asian participants and for Asian physicians evaluated by Black participants.

These results align with cue-based models, which suggest that without direct information about a physician’s ability or intentions, consumers rely on observable cues. Disclosure of AI use may be interpreted as outsourcing judgment, leading to doubts about both intent and expertise. The heightened negative response among younger participants could reflect greater familiarity with AI and increased awareness of its limitations. As these are exploratory findings, further research is needed to clarify these patterns.

The interaction between physician and participant race indicates that AI-use disclosure may intensify pre-existing differences in trust and expectations among diverse consumer–physician pairings, at least in perceptions of physician warmth, although this effect was relatively modest.

Social implications and communication strategies for AI adoption

The current study findings highlight that AI adoption in healthcare involves not only technical and organizational considerations but also substantial social implications. Communicating AI use effectively, by emphasizing physician responsibility, shared decision-making, and patient benefits, may help mitigate negative perceptions, particularly among younger consumers. Tailored communication strategies and sustained attention to consumer concerns remain essential as AI becomes further integrated into clinical practice.