Patients’ Perceptions of AI Use in Health Care

How to cite

Binelli, Chiara; Sartori, Laura; Macis, Mario. 2026. "Patients’ Perceptions of AI Use in Health Care." AEA RCT Registry. AEARCTR-0017724. https://doi.org/10.1257/rct.17724-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Chiara Binelli* , Laura Sartori , Mario Macis

* Primary investigator

Registry ID
AEARCTR-0017724
DOI
https://doi.org/10.1257/rct.17724-1.0
First registered
Last updated
Status
In development
Trial dates
2026-02-04 – 2026-02-17
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17724

Abstract

Artificial intelligence (AI) tools are moving rapidly from back-office uses to health care to shape diagnoses, treatment choices, and follow-up recommendations. Yet adoption depends not only on whether AI can improve accuracy, but also on whether patients view the use of AI in health care as trustworthy, competent, and appropriately accountable, and whether those perceptions translate into concrete behavioral responses. This study uses randomized clinical vignettes to estimate how patients perceive physicians’ use of AI across common medical applications. The goal is to quantify the impact of AI use and of how it is framed on patients’ willingness to follow recommendations, their propensity to request second opinions, their trust in clinicians and institutions, and the perceived quality of health care. The goal is to identify which design and communication choices can preserve trust while enabling safe, effective integration of AI into medical decision-making.

Experimental design

We have designed an online survey that will include two randomized clinical vignettes and one A/B experiment. Vignette 1 presents a routine chronic care scenario (diabetes' medication adjustment), and Vignette 2 presents a high-stakes specialist decision (oncology treatment plan). In both vignettes, respondents are randomly assigned to either a control condition with no mention of AI or a treatment condition in which AI is described as an additional verification step to the doctor's diagnosis.
The survey also includes an A/B experiment in which respondents are shown a scenario where the doctor and AI disagree on the recommended action. In one arm, the doctor advises the patient to follow the doctor's recommendation; in the other arm, the doctor seeks the opinion of a human colleague who agrees with the AI recommendation.

Design details

Randomization method
Randomization done by survey company
Randomization unit
Individual
Clusters
There are no clusters.
Observations
2,000
Arms
Vignette A and B: 500 individuals control, 500 individuals treatment A/B experiment: 1,000 individuals in scenario A, 1,000 individuals in scenario B
Primary outcomes
Primary outcomes for Vignette 1 and Vignette 2: 1. Probability to follow the recommendation of the doctor 2. Probability to request a second opinion from a different doctor Primary outcomes for A/B Experiment: 1. Probability to follow the recommendation of the doctor versus probability to follow recommendation of AI 3. Probability to request a second opinion from a different doctor

Cited by

No citation data available for this registration yet.

Keywords