Patient Trust in Artificial Intelligence in Medicine

Olena Vyrva, MD · Updated: 30 October 2026

Why is patient trust becoming a key condition for adopting AI?

Trust in medical artificial intelligence (AI) is becoming one of the factors that shape how society views new technologies for diagnosis and treatment. As algorithms move into clinical practice, an important question remains: are patients ready to entrust a diagnostic decision to AI, and what role should the physician keep?

A study by Michael Constantin Kirchberger, published in JMIR Formative Research in 2026, examined how adults in the United States view the use of AI in medicine, how far they trust different diagnostic scenarios, and what they expect from the training of future physicians [1]. Particular attention was paid to concerns about clinicians becoming over-reliant on algorithms and to support for mandatory AI training.

The main finding is that public trust in medical AI is conditional: respondents mostly favoured decisions based on the physician's own expertise and considered it necessary to train future doctors to critically assess what AI can and cannot do.

Key points

62.8% of 524 US adults trusted a diagnosis based on the physician's expertise alone; only 7.8% preferred a model in which AI makes the first diagnosis.
93.9% rated formal training of physicians on AI limitations and errors as essential or very important.
81.1% were very or extremely concerned that future doctors may rely on AI without critical review.
85.1% supported mandatory teaching on AI use, ethics and limitations in medical school.
The survey measured preferences, not diagnostic accuracy or patient outcomes.

How was the study conducted, and who was surveyed?

A total of 524 adults living in the United States took part in a web-based survey.

The questionnaire had 12 questions: 7 core questions on perceptions of AI in medicine and 5 demographic questions. The researchers assessed trust in different diagnostic scenarios, the importance of training physicians on AI limitations, concern about over-reliance on algorithms, support for mandatory training, and views on the priorities of medical education.

Which diagnostic model do respondents trust most?

A preference for physician expertise was the clearest result of the survey. When asked which approach to a complex diagnostic situation they would trust most, 62.8% chose a physician who decides based solely on their own professional expertise.

An option in which the physician first makes the diagnosis and then uses AI as an additional check was chosen by 26.7%. Only 7.8% preferred a model in which AI forms the initial diagnosis and the physician acts as the reviewer.

These results point to a preference for the physician leading the diagnostic process. However, they cannot be read as evidence that diagnosis without AI is clinically superior: the study assessed public preferences in a hypothetical scenario, not diagnostic accuracy or treatment outcomes across different models.

Physicians reviewing a chest X-ray analysed by an AI assistant on a monitor

Why does training physicians to work with AI matter so much?

Training on AI limitations was closely tied to respondents' trust in physicians' decisions. Overall, 93.9% rated formal training of physicians on the limitations and potential errors of AI as "essential" or "very important".

At the same time, 81.1% reported strong concern that future physicians might rely too heavily on AI output without critically checking it.

Support for mandatory training reflects these concerns: 85.1% backed including training on AI use, its ethical aspects and its limitations in the compulsory medical school curriculum.

The results show that the public links trust in AI not only to the technology itself but also to the competence of the specialist who is responsible for interpreting its results.

Should medical education shift its focus towards human-centred skills?

Human-centred skills include empathy, communication with patients and ethical judgement. In the study, 70.2% said medical education should give more weight to such skills, rather than to clinical procedures alone, provided that AI can reliably take over more technical tasks.

This reflects respondents' view of the physician's future role: technological competence should be combined with communication skills, professional judgement and preserved human contact. The question was framed around a hypothetical development of AI, so the answer cannot be taken as evidence that technical clinical skills should be reduced in real curricula.

Perceptions of physicians' digital skills were generally positive: 79.6% rated them as good or very good.

How could the results influence the training of future physicians?

Mandatory AI training is the most direct educational implication of the survey. Strong support for it suggests that respondents expect physicians not merely to use digital tools, but to understand the limitations of algorithms, recognise potential errors and remain responsible for the clinical decision.

For medical education programmes, this means treating AI literacy as a combination of technical knowledge, critical appraisal of outputs and ethical competence.

Specialists working with an AI assistant displayed on a large screen

Trust in AI depends on preserving the physician's professional role

The key practical takeaway is that the public does not reject AI in medicine outright, but prefers models in which the physician keeps the leading role. Strong support for mandatory training, concern about over-reliance on algorithms and the priority given to human-centred skills all point to an expectation that AI will be used responsibly and with critical judgement.

For medical education, the results support systematic training of physicians to work with these technologies. However, a cross-sectional survey cannot establish cause and effect, and it does not show that the proposed educational changes would by themselves increase trust or improve clinical outcomes. The next step is to evaluate educational approaches in more representative samples and to study how physician training affects decision-making, communication and patient trust.

For clinicians

Respondents accepted AI mainly as a second check on a physician's own diagnosis, not as the first reader. Being open with patients about when and how AI is used, and showing that its output has been critically reviewed, is likely to matter as much as the tool itself. The data are self-reported preferences from a single online US sample.

Q&A

Which diagnostic approach do people consider most trustworthy?

In the survey, 62.8% preferred a diagnosis based on the physician's expertise, 26.7% an initial physician assessment followed by an AI check, and 7.8% an initial AI diagnosis followed by physician review.

What educational changes did participants support?

85.1% supported mandatory teaching on AI use, ethics and limitations in medical school. In addition, 70.2% favoured prioritising human-centred skills, provided AI can reliably handle more technical tasks.

How concerned are people about doctors relying too much on AI?

Very: 81.1% said they were very or extremely concerned that future physicians might accept AI results without critical review.

Does the study show that diagnosis without AI is more accurate?

No. The survey measured public preferences in a hypothetical scenario. It did not compare the accuracy of different diagnostic models or their effect on patient outcomes.

References

Kirchberger MC. Public Perceptions of AI in Medicine and Implications for Future Medical Education: Cross-Sectional Survey. JMIR Form Res. 2026;10:e89123. doi: 10.2196/89123. PMID: 41984594.