J Particip Med. 2026 Jul 24. 18
e93720
Background: Generative artificial intelligence (GenAI) tools are widely accessible to the public, who are engaging with them for a wide range of health care applications. Existing research has focused predominantly on clinician-facing adoption. Far less is known about how patients and family members use GenAI tools, particularly in rare disease contexts, where diagnostic delay, limited specialist access, and unmet informational needs are common.
Objective: This study aimed to examine the experiences and opinions of adult patients with rare diseases and parents or guardians of children with rare diseases regarding the use of GenAI tools.
Methods: Between November 2025 and January 2026, we conducted an exploratory mixed methods web-based survey using convenience sampling through rare disease community organizations in the United States. The survey included closed-ended items assessing prior GenAI use, purposes of use, perceived influence on medical decisions and diagnoses, trust, concerns, communication with clinicians, and experiences of harm, alongside open-text questions capturing qualitative reflections. Descriptive statistics were used to summarize quantitative data. Inductive qualitative analysis was applied to the open-text responses.
Results: A total of 115 respondents completed the survey. A majority of respondents were parents or guardians of a child with a rare disease (n=74, 64.3%), and the remaining respondents were patients with a rare disease (n=41, 35.7%). Slightly more than half of respondents (n=63, 54.8%) reported prior use of GenAI tools in the context of rare disease. Common purposes included exploring new treatments or clinical trials (n=53, 46.1%), interpreting medical tests or clinical notes (n=37, 32.2%), locating specialists or care centers (n=29, 25.2%), and suggesting possible diagnoses (n=28, 24.3%). Nearly one-third of respondents (n=37, 32%) reported some degree of influence of GenAI on their medical decisions. Nearly 10% (n=12) reported contributions of GenAI to a formal diagnosis. Concern about GenAI accuracy was widespread; 71 of 115 (61.8%) respondents reported moderate to extreme concern. Most respondents (n=90, 78.3%) had not discussed AI-generated information with a clinician. Few respondents (n=7, 6.1%) reported experiencing harm. Qualitative analysis identified 3 themes: (1) GenAI as a practical tool for augmenting patient and caregiver expertise and advocacy, (2) conditional trust and bounded use of GenAI with an emphasis on verification and human oversight, and (3) perceived risks, harms, and structural concerns, including inaccuracies, genetic misinterpretation, and privacy and commercialization issues.
Conclusions: In this exploratory study, patients and families affected by rare diseases were actively experimenting with GenAI tools to support information seeking, preparation, and advocacy while simultaneously expressing substantial caution and concern about the reliability, safety, and appropriate boundaries of use. Our findings contrast sharply with clinician concerns that patients lack the capacity to use GenAI tools judiciously. Notwithstanding, the sample was skewed toward highly educated participants. Future research should prioritize more representative samples to better capture the range of patient and caregiver experiences with GenAI in rare disease care.
Keywords: education; general practice; generative AI; large language models; online survey questionnaire; primary care; qualitative research; training