Can ChatGPT provide reliable guidance to families of children with Duchenne Muscular Dystrophy? A dual perspective of parents and experts Authors: Gülşah Çelik, Hanife Hale Hekim, Meral Bilgilisoy Filiz, Naciye Füsun Toraman, Dilek Yapar Published: 2026 ABSTRACT BackgroundArtificial intelligence (AI)-driven chatbots, such as ChatGPT, have the potential to improve healthcare communication by providing timely and accessible medical information. Duchenne Muscular Dystrophy (DMD) is a progressive neuromuscular disorder that significantly impacts patients and their families. Parents of DMD patients seek reliable information on disease progression, complications, and care management. However, the effectiveness of AI-generated medical guidance remains unclear.MethodsThis study evaluated ChatGPT’s capacity to provide guidance to DMD patients’ families by analyzing expert and parental assessments. Ten frequently asked questions were compiled from parental inquiries and submitted to ChatGPT-4o. Responses were evaluated by physiatrists, pediatric neurologists, and parents using structured rubrics. Experts rated responses based on Accuracy, Currency, Comprehensiveness, Usefulness, and Understandability, while parents assessed them in terms of Satisfaction, Empathy, Comprehensiveness, and Understandability.ResultsExperts rated Accuracy highest (4.3 ± 0.6; 91.8% of ratings ≥ 4), followed by Usefulness (4.2 ± 0.6; 88.6%), Currency (4.2 ± 0.7; 87.1%), Comprehensiveness (4.2 ± 0.7; 85.1%), and Understandability (4.2 ± 0.7; 83.4%). Parents rated Understandability highest (4.5 ± 0.8; 86.0%), followed by Comprehensiveness (4.4 ± 0.8; 84.0%) and Satisfaction (4.4 ± 0.8; 83.8%). Empathy received the lowest parental score (4.2 ± 0.9; 75.1%). Although both groups generally evaluated the responses favorably (median scores ≥ 4), experts tended to prioritize technical accuracy, whereas parents placed greater emphasis on clarity and practical usefulness.ConclusionChatGPT shows promise as an informational tool for parents of DMD, particularly in providing accurate and applicable medical guidance. However, limitations in empathy suggest a need for refinement. While ChatGPT should complement rather than replace human expertise, future improvements could expand its role in patient education and public health. Further research is needed to optimize AI-driven healthcare communication for reliability and ethical considerations.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12887-026-07126-5. Introduction Advancements in digital technology have significantly improved public access to medical information [1]. Today, rapidly evolving artificial intelligence (AI) applications play a crucial role across a wide spectrum, from biomedical research to healthcare services and medical education [2]. Natural language processing (NLP) serves as a fundamental component in AI applications, enabling computers to comprehend, interpret, and generate human language. One of the most prominent tools developed in this field is large language models (LLMs) [3]. Among these, ChatGPT (OpenAI Incorporated, Mission District, San Francisco, USA) stands out as a state-of-the-art language model [4]. With its advanced language processing capabilities, ChatGPT has the potential to enhance clinical decision support systems in disease diagnosis and management [5]. By facilitating the presentation of medical information and improving the comprehension of complex health data, it can increase access to healthcare services, particularly for patients in rural and underserved areas, ultimately enhancing the quality of care. This, in turn, may help patients make more informed decisions regarding their health [6, 7]. Additionally, ChatGPT can contribute to the optimization of treatment protocols, thereby improving therapeutic efficacy, increasing patient satisfaction, and reducing healthcare costs [8, 9]. In this context, examining the capacity of ChatGPT to provide medical guidance is a crucial step in exploring its potential to enhance the overall quality and accessibility of public health services. Duchenne Muscular Dystrophy (DMD) is a progressive X-linked neuromuscular disorder characterized by muscle weakness and loss of function [10]. Despite advancements in treatment approaches, individuals with DMD gradually lose their independence and require external assistance for daily living activities [11]. Given the extensive biological, socio-emotional, psychosocial and economic impacts of the disease, family members providing care for these patients face numerous challenges throughout the process. The absence of a definitive cure, the psychological burden associated with a limited life expectancy, the increasing financial costs related to long-term medical care and supportive equipment, and the physical and emotional strain of caregiving constitute significant sources of stress for patient caregivers [12]. Parents caring for individuals with DMD often experience concerns regarding disease progression, potential complications, and appropriate caregiving practices, frequently seeking guidance from healthcare professionals [13]. The complexity of the disease and the long-term nature of care further necessitate access to information on various aspects, including disease trajectory, possible complications, what to expect in the future, and their own responsibilities in the caregiving process [14, 15]. This situation exacerbates anxiety among caregivers due to information gaps, thereby increasing their psychosocial burden. Consequently, there is a growing need for innovative solutions that facilitate access to accurate and up-to-date medical information, providing families with the guidance they require. In this regard, ChatGPT emerges as a promising AI tool capable of addressing this need by delivering timely information and support to patients and their families [16]. In a study published by Peng et al., ChatGPT/GPT-4 was described as an indispensable tool for Physical Medicine and Rehabilitation (PMR) specialists in the modern era [17]. Similarly, Mittal et al. explored the potential of ChatGPT in developing personalized rehabilitation plans for elderly patients, emphasizing that rather than replacing physicians’ expertise, ChatGPT could serve as a supportive tool for clinical decision-making [18]. Additionally, Silver et al. highlighted that when carefully integrated within ethical guidelines and in collaboration with medical professionals, AI can be a valuable resource for providing support and guidance in the field of rehabilitation medicine [19]. AI-driven systems such as ChatGPT have the potential to provide guidance to families of DMD patients regarding disease progression, care management, and current treatment approaches. However, the effectiveness and reliability of such systems should not only be evaluated from a technological perspective but also through the assessments of expert physicians and parents. While the existing literature has explored the role of ChatGPT/GPT-4 in various medical disciplines, including radiology, gynecology, infectious diseases, and neurosurgery, there is a noticeable gap in studies within the field of PMR [17]. Specifically, to the best of our knowledge, no study has yet investigated the guidance capacity of AI-based systems in the context of DMD, a disease where parents have a high demand for information and seek answers to numerous questions. This study aims to evaluate ChatGPT’s potential to provide guidance and information to families of children with DMD from the perspectives of both parents and expert physicians. Materials and methods Ethical approval Ethical approval was obtained from the University of Health Sciences Antalya Training and Research Hospital Clinical Research Ethics Committee (Approval Date: January 30, 2025; Approval Number: 2/16). All participating experts and parents were informed about the study’s objectives, and written informed consent was obtained prior to their participation. Question preparation and ChatGPT responses The ten most frequently asked questions about DMD were identified by PMR specialists working at the Neuromuscular Diseases Center of the University of Health Sciences Antalya Training and Research Hospital. These experts routinely manage children diagnosed with DMD and provide long-term follow-up care, enabling them to capture commonly expressed parental concerns. The questions were structured to reflect the original phrasing used by parents during clinical consultations. In 31 January 2025, these questions were submitted to ChatGPT-4o. Responses were recorded without any word limit, and each question was asked only once. The full text of the questions and ChatGPT responses is provided in Supplementary Appendix 1. Evaluator groups Two groups of evaluators participated in this study: parents of children with DMD (n = 45) and specialist physicians in PMR and Pediatric Neurology (n = 51). Eligible physicians were required to be practicing in Turkey and to have at least four years of professional experience in the care of patients with DMD. Pediatric neurologists were included for their primary role in diagnosis, monitoring, and treatment initiation, while PMR specialists were involved for their expertise in rehabilitation and functional management. Parents eligible for participation were primary caregivers of children with a confirmed diagnosis of DMD who agreed to take part in the study. The required number of evaluators in rubric assessments depends on the study’s scope and originality. Although no strict rule exists, at least five are generally recommended for content validity, while complex studies may include up to 40 depending on evaluator expertise [20, 21]. In this study, 51 specialists from various institutions were engaged using snowball sampling, and online evaluation forms were distributed to them via email. Additionally, 45 parents participated in the study, and their evaluations were conducted face-to-face. All invited participants who met the eligibility criteria completed the evaluation forms in full, resulting in a 100% response rate with no excluded or incomplete responses. Evaluation of ChatGPT responses The evaluation was conducted using the rubric method. The response to each question was scored between 1 and 5 by the evaluator according to the categories we determined. Although rubric evaluations are commonly used in educational fields, especially for grading written assignments or projects, they’re also applicable in research and various other domains, including health [22–25]. The selection of evaluation domains for this study was guided by a comprehensive review of existing literature on the assessment of large language models (LLMs) in healthcare. Two key studies played a pivotal role in shaping our evaluation framework [25, 26]. Drawing from these studies, we defined two sets of evaluation domains. Experts and parents evaluated ChatGPT responses using distinct domain sets reflecting their different roles and perspectives. Experts assessed medically oriented domains, whereas parents evaluated user-centered domains. Accordingly, analyses and comparisons were conducted within each evaluator group only, and no direct comparisons were made between expert and parent ratings. For parental assessments, the primary domains included Satisfaction, Empathy, Comprehensiveness, and Understandability, ensuring a user-centered evaluation of ChatGPT’s responses to questions about DMD. Parents evaluated ChatGPT’s answers to questions using a rubric covering satisfaction, empathy, comprehensiveness and understandability, each rated on a 1–5 scale. In addition, expert evaluations were conducted using a structured scoring rubric to assess ChatGPT’s responses in terms of accuracy, currency, comprehensiveness, usefulness, and understandability. These domains ensured a systematic and multidimensional assessment of ChatGPT’s performance from both expert and user perspectives. Although this study did not aim to develop a scale or formally evaluate psychometric properties, additional exploratory analyses were conducted to examine the internal consistency and structural coherence of the rubric-based evaluation domains. These analyses are provided in the Supplementary Materials (Supplementary Appendix 2–4). Likert Response of ≥ 4” reflects the percentage of evaluators rating responses as highly satisfactory for each criterion. The ‘Overall Evaluation Percentage’ represents the average percentage of evaluators providing ratings of 4 or higher across all responses and criteria. An overall index score for each domain was then computed by aggregating and averaging all ratings within that category, providing a single composite measure per domain [25]. For example, if an evaluator rated the ‘comprehensiveness’ of 10 ChatGPT responses as x₁, x₂, …, x₁₀ on a 5-point Likert scale. The Accuracy Index Score for an expert was computed as: Accuracy Overall Index Score = (x₁ + x₂ + … + x₁₀) / 10. This calculation was repeated for all 51 experts, and the descriptive statistics of these index scores were reported in the tables. All evaluators completed all ratings in full, and no domain-level data were missing. Therefore, no imputation or special handling of missing data was required when computing overall index scores. Statistical analysis Statistical analyses were performed with IBM SPSS 26.0 (SPSS Inc., Chicago, USA). The distribution of continuous variables was assessed through both visual methods (such as histograms and probability plots) and statistical tests (Kolmogorov-Smirnov and Shapiro-Wilk tests).Descriptive statistics were presented as number, percentage, mean±standard deviation and median (minimum; maximum). Pairwise comparisons between evaluation domains were performed using the Wilcoxon Signed-Ranks Test to compare paired scores provided by the same evaluator across identical scenarios. Analyses were conducted separately for experts and parents, as each group rated different evaluation domains based on their respective roles and perspectives. Because each comparison involved repeated ratings from the same evaluator, all analyses were treated as dependent (within-subject) comparisons. A p-value of < 0.05 was considered statistically significant. Results Characteristics of parents performing the evaluation A total of 45 parents of children with DMD were included (mean age 39.4 ± 6.1 years, range 29–52). The majority were mothers (71.1%, n = 32), while 28.9% (n = 13) were fathers. Participants’ highest level of education completed was as follows: 17.8% (n = 8) had completed primary or middle school, 31.1% (n = 14) high school, and 51.1% (n = 23) university. The mean duration since diagnosis was 6.4 ± 5.1 years (range 1–20). Regarding ChatGPT familiarity, 60.0% (n = 27) reported no prior knowledge, 24.4% (n = 11) some knowledge, 13.3% (n = 6) moderate knowledge, and 2.2% (n = 1) good knowledge. After completion of the evaluation process, participants were additionally asked about their curiosity regarding ChatGPT and whether they believed its use in healthcare would increase in the future. Accordingly, 95.6% (n = 43) of the parents reported increased curiosity about ChatGPT, while 97.8% (n = 44) believed that its use in healthcare would increase in the future. Characteristics of specialist physicians performing the evaluation A total of 51 specialists participated (35.3% male, 64.7% female; mean age 40.8 ± 11.2 years, range 30–68). Academic titles included research assistants (21.6%, n = 11), assistant/associate professors (15.6%, n = 8), professors (17.6%, n = 9), and specialist physicians (45.1%, n = 23). All evaluators in the physician group were licensed medical doctors (MD), including resident physicians (research assistants in their final year), specialists, and academic faculty members in PMR or Pediatric Neurology. Most were PMR specialists (80.4%, n = 41), while 19.6% (n = 10) were pediatric neurologists. Knowledge of LLMs was substantial in 15.7% (n = 8), moderate in 31.4% (n = 16), limited in 43.1% (n = 22), and absent in 9.8% (n = 5). The majority (70.6%, n = 36) reported using ChatGPT-3 or 4. Curiosity about ChatGPT increased in 96.1% (n = 49) of participants, and 98.0% (n = 50) anticipated broader use of ChatGPT in healthcare. Evaluation of ChatGPT responses For each question, experts rated ChatGPT’s responses as 4 or higher (≥ 4) in all five categories more than 80%. Considering the overall index scores calculated by summing and averaging the scores from all questions, the highest mean scores were observed for accuracy (4.3 ± 0.6). Accuracy had the highest overall index percentage of responses scoring ≥ 4 (91.8%), while the percentages for the other four categories ranged between 87.1% and 83.4% (Table 1.) (Fig. 1.). The median overall index scores for the evaluation criteria ranged from 4.0 (2.7–5.0) for currency to 4.3 (2.9–5.0) for understandability. Pairwise comparisons revealed statistically significant differences between accuracy and currency (p = 0.001) and between accuracy and comprehensiveness (p = 0.025), indicating that accuracy was rated significantly higher than currency and comprehensiveness by experts (Table 2.). Table 1Evaluation of ChatGPT’s answers to the questions by the experts according to the scoring rubric in terms of accuracy, currency, comprehensiveness, usefulness, and understandabilityExperts, N = 51Experts with Likert Response of ≥ 4,n (%)Mean ± SDScoreMedian (Min-max) Score Question 1 Accuracy48 (94.1)4.4 ± 0.64.0 (3.0–5.0) Currency38 (74.5)3.9 ± 0.84.0 (2.0–5.0) Comprehensiveness36 (70.6)3.9 ± 0.84.0 (2.0–5.0) Usefulness41 (80.4)4.0 ± 0.74.0 (2.0–5.0) Understandability42 (82.4)4.2 ± 0.74.0 (3.0–5.0) Question 2 Accuracy47 (92.1)4.3 ± 0.64.0 (3.0–5.0) Currency47 (92.1)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness40 (78.4)4.1 ± 0.74.0 (3.0–5.0) Usefulness45 (88.2)4.1 ± 0.64.0 (3.0–5.0) Understandability38 (74.5)4.1 ± 0.84.0 (3.0–5.0) Question 3 Accuracy43 (84.3)4.2 ± 0.74.0 (3.0–5.0) Currency41 (80.4)4.1 ± 0.74.0 (3.0–5.0) Comprehensiveness40 (78.4)4.1 ± 0.74.0 (3.0–5.0) Usefulness46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Understandability43 (84.3)4.3 ± 0.74.0 (3.0–5.0) Question 4 Accuracy47 (92.1)4.3 ± 0.64.0 (3.0–5.0) Currency45 (88.2)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness46 (90.2)4.2 ± 0.64.0 (3.0–5.0) Usefulness47 (92.2)4.2 ± 0.64.0 (3.0–5.0) Understandability45 (88.2)4.3 ± 0.74.0 (3.0–5.0) Question 5 Accuracy46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Currency45 (88.2)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness43 (84.3)4.2 ± 0.74.0 (3.0–5.0) Usefulness43 (84.3)4.2 ± 0.74.0 (3.0–5.0) Understandability39 (76.5)4.1 ± 0.84.0 (2.0–5.0) Question 6 Accuracy47 (92.2)4.3 ± 0.64.0 (3.0–5.0) Currency46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Comprehensiveness46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Usefulness46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Understandability44 (86.3)4.4 ± 0.75.0 (3.0–5.0) Question 7 Accuracy46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Currency44 (86.3)4.2 ± 0.64.0 (3.0–5.0) Comprehensiveness42 (82.4)4.2 ± 0.74.0 (3.0–5.0) Usefulness44 (86.3)4.2 ± 0.74.0 (3.0–5.0) Understandability44 (86.3)4.3 ± 0.74.0 (3.0–5.0) Question 8 Accuracy47 (92.2)4.3 ± 0.64.0 (3.0–5.0) Currency44 (86.3)4.2 ± 0.74.0 (3.0–5.0) Comprehensiveness48 (94.1)4.3 ± 0.64.0 (3.0–5.0) Usefulness45 (88.2)4.2 ± 0.74.0 (3.0–5.0) Understandability43 (84.3)4.2 ± 0.74.0 (2.0–5.0) Question 9 Accuracy50 (98.0)4.4 ± 0.54.0 (3.0–5.0) Currency49 (96.1)4.3 ± 0.54.0 (3.0–5.0) Comprehensiveness48 (94.1)4.3 ± 0.64.0 (3.0–5.0) Usefulness48 (94.1)4.4 ± 0.64.0 (3.0–5.0) Understandability46 (90.2)4.4 ± 0.75.0 (3.0–5.0) Question 10 Accuracy47 (92.2)4.4 ± 0.64.0 (3.0–5.0) Currency45 (88.2)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness45 (88.2)4.4 ± 0.74.0 (3.0–5.0) Usefulness47 (92.2)4.3 ± 0.64.0 (3.0–5.0) Understandability41 (80.4)4.3 ± 0.84.0 (3.0–5.0) All QuestionsOverall Index Percentage of Responses Scoring ≥ 4, %*Overall Index Score**Overall Index Score** Accuracy91.84.3 ± 0.64.2 (3.1-5.0) Currency87.14.2 ± 0.74.0 (2.7-5.0) Comprehensiveness85.14.2 ± 0.74.1 (3.0–5.0) Usefulness88.64.2 ± 0.64.2 (3.1-5.0) Understandability83.44.2 ± 0.74.3 (2.9-5.0)Bold values indicate statistical significance (p < 0.05)*: The percentage of responses scoring ≥ 4 across all ten questions are provided for each of the five evaluation areas**: The overall index scores for all questions are derived by aggregating and averaging the scores from all five areas across each question Table 1Evaluation of ChatGPT’s answers to the questions by the experts according to the scoring rubric in terms of accuracy, currency, comprehensiveness, usefulness, and understandabilityExperts, N = 51Experts with Likert Response of ≥ 4,n (%)Mean ± SDScoreMedian (Min-max) Score Question 1 Accuracy48 (94.1)4.4 ± 0.64.0 (3.0–5.0) Currency38 (74.5)3.9 ± 0.84.0 (2.0–5.0) Comprehensiveness36 (70.6)3.9 ± 0.84.0 (2.0–5.0) Usefulness41 (80.4)4.0 ± 0.74.0 (2.0–5.0) Understandability42 (82.4)4.2 ± 0.74.0 (3.0–5.0) Question 2 Accuracy47 (92.1)4.3 ± 0.64.0 (3.0–5.0) Currency47 (92.1)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness40 (78.4)4.1 ± 0.74.0 (3.0–5.0) Usefulness45 (88.2)4.1 ± 0.64.0 (3.0–5.0) Understandability38 (74.5)4.1 ± 0.84.0 (3.0–5.0) Question 3 Accuracy43 (84.3)4.2 ± 0.74.0 (3.0–5.0) Currency41 (80.4)4.1 ± 0.74.0 (3.0–5.0) Comprehensiveness40 (78.4)4.1 ± 0.74.0 (3.0–5.0) Usefulness46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Understandability43 (84.3)4.3 ± 0.74.0 (3.0–5.0) Question 4 Accuracy47 (92.1)4.3 ± 0.64.0 (3.0–5.0) Currency45 (88.2)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness46 (90.2)4.2 ± 0.64.0 (3.0–5.0) Usefulness47 (92.2)4.2 ± 0.64.0 (3.0–5.0) Understandability45 (88.2)4.3 ± 0.74.0 (3.0–5.0) Question 5 Accuracy46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Currency45 (88.2)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness43 (84.3)4.2 ± 0.74.0 (3.0–5.0) Usefulness43 (84.3)4.2 ± 0.74.0 (3.0–5.0) Understandability39 (76.5)4.1 ± 0.84.0 (2.0–5.0) Question 6 Accuracy47 (92.2)4.3 ± 0.64.0 (3.0–5.0) Currency46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Comprehensiveness46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Usefulness46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Understandability44 (86.3)4.4 ± 0.75.0 (3.0–5.0) Question 7 Accuracy46 (90.2)4.3 ± 0.64.0 (3.0–5.0) Currency44 (86.3)4.2 ± 0.64.0 (3.0–5.0) Comprehensiveness42 (82.4)4.2 ± 0.74.0 (3.0–5.0) Usefulness44 (86.3)4.2 ± 0.74.0 (3.0–5.0) Understandability44 (86.3)4.3 ± 0.74.0 (3.0–5.0) Question 8 Accuracy47 (92.2)4.3 ± 0.64.0 (3.0–5.0) Currency44 (86.3)4.2 ± 0.74.0 (3.0–5.0) Comprehensiveness48 (94.1)4.3 ± 0.64.0 (3.0–5.0) Usefulness45 (88.2)4.2 ± 0.74.0 (3.0–5.0) Understandability43 (84.3)4.2 ± 0.74.0 (2.0–5.0) Question 9 Accuracy50 (98.0)4.4 ± 0.54.0 (3.0–5.0) Currency49 (96.1)4.3 ± 0.54.0 (3.0–5.0) Comprehensiveness48 (94.1)4.3 ± 0.64.0 (3.0–5.0) Usefulness48 (94.1)4.4 ± 0.64.0 (3.0–5.0) Understandability46 (90.2)4.4 ± 0.75.0 (3.0–5.0) Question 10 Accuracy47 (92.2)4.4 ± 0.64.0 (3.0–5.0) Currency45 (88.2)4.2 ± 0.74.0 (2.0–5.0) Comprehensiveness45 (88.2)4.4 ± 0.74.0 (3.0–5.0) Usefulness47 (92.2)4.3 ± 0.64.0 (3.0–5.0) Understandability41 (80.4)4.3 ± 0.84.0 (3.0–5.0) All QuestionsOverall Index Percentage of Responses Scoring ≥ 4, %*Overall Index Score**Overall Index Score** Accuracy91.84.3 ± 0.64.2 (3.1-5.0) Currency87.14.2 ± 0.74.0 (2.7-5.0) Comprehensiveness85.14.2 ± 0.74.1 (3.0–5.0) Usefulness88.64.2 ± 0.64.2 (3.1-5.0) Understandability83.44.2 ± 0.74.3 (2.9-5.0)Bold values indicate statistical significance (p < 0.05)*: The percentage of responses scoring ≥ 4 across all ten questions are provided for each of the five evaluation areas**: The overall index scores for all questions are derived by aggregating and averaging the scores from all five areas across each question Evaluation of ChatGPT’s answers to the questions by the experts according to the scoring rubric in terms of accuracy, currency, comprehensiveness, usefulness, and understandability Bold values indicate statistical significance (p < 0.05) *: The percentage of responses scoring ≥ 4 across all ten questions are provided for each of the five evaluation areas **: The overall index scores for all questions are derived by aggregating and averaging the scores from all five areas across each question Fig. 1“Distribution of overall index percentage of all questions’ responses scoring ≥ 4 across the five evaluation areas (based on experts’ assessment) “Distribution of overall index percentage of all questions’ responses scoring ≥ 4 across the five evaluation areas (based on experts’ assessment) Table 2Pairwise comparisons of the overall index scores of evaluation criteria by experts and parentsAll Questions Overall Index Score*P values**Median (Min-max) Score Evaluation by experts Accuracy4.2 (3.1-5.0) Accuracy vs. Currency = 0.001 Accuracy vs. Comprehensiveness = 0.025 Accuracy vs. Usefulness = 0.179Accuracy vs. Understandability = 0.457Currency vs. Comprehensiveness = 0.752Currency vs. Usefulness = 0.105Currency vs. Understandability = 0.404Comprehensiveness vs. Usefulness = 0.111Comprehensiveness vs. Understandability = 0.281Usefulness vs. Understandability = 0.916 Currency4.0 (2.7-5.0) Comprehensiveness4.1 (3.0–5.0) Usefulness4.2 (3.1-5.0) Understandability4.3 (2.9-5.0) Evaluation by parents Satisfaction4.5 (3.0–5.0) Satisfaction vs. Empathy = 0.002 Satisfaction vs. Comprehensiveness = 0.935Satisfaction vs. Understandability = 0.074 Empathy vs. Comprehensiveness = 0.001 Empathy vs. Understandability = 0.001 Comprehensiveness vs. Understandability = 0.090 Empathy4.3 (2.6-5.0) Comprehensiveness4.5 (2.8-5.0) Understandability4.8 (2.6-5.0)Bold values indicate statistical significance (p < 0.05)*: The overall index scores for all scenarios is derived by aggregating and averaging the scores from all areas across each scenario**: Wilcoxon Signed Ranks Test Table 2Pairwise comparisons of the overall index scores of evaluation criteria by experts and parentsAll Questions Overall Index Score*P values**Median (Min-max) Score Evaluation by experts Accuracy4.2 (3.1-5.0) Accuracy vs. Currency = 0.001 Accuracy vs. Comprehensiveness = 0.025 Accuracy vs. Usefulness = 0.179Accuracy vs. Understandability = 0.457Currency vs. Comprehensiveness = 0.752Currency vs. Usefulness = 0.105Currency vs. Understandability = 0.404Comprehensiveness vs. Usefulness = 0.111Comprehensiveness vs. Understandability = 0.281Usefulness vs. Understandability = 0.916 Currency4.0 (2.7-5.0) Comprehensiveness4.1 (3.0–5.0) Usefulness4.2 (3.1-5.0) Understandability4.3 (2.9-5.0) Evaluation by parents Satisfaction4.5 (3.0–5.0) Satisfaction vs. Empathy = 0.002 Satisfaction vs. Comprehensiveness = 0.935Satisfaction vs. Understandability = 0.074 Empathy vs. Comprehensiveness = 0.001 Empathy vs. Understandability = 0.001 Comprehensiveness vs. Understandability = 0.090 Empathy4.3 (2.6-5.0) Comprehensiveness4.5 (2.8-5.0) Understandability4.8 (2.6-5.0)Bold values indicate statistical significance (p < 0.05)*: The overall index scores for all scenarios is derived by aggregating and averaging the scores from all areas across each scenario**: Wilcoxon Signed Ranks Test Pairwise comparisons of the overall index scores of evaluation criteria by experts and parents Accuracy vs. Currency = 0.001 Accuracy vs. Comprehensiveness = 0.025 Accuracy vs. Usefulness = 0.179 Accuracy vs. Understandability = 0.457 Currency vs. Comprehensiveness = 0.752 Currency vs. Usefulness = 0.105 Currency vs. Understandability = 0.404 Comprehensiveness vs. Usefulness = 0.111 Comprehensiveness vs. Understandability = 0.281 Usefulness vs. Understandability = 0.916 Satisfaction vs. Empathy = 0.002 Satisfaction vs. Comprehensiveness = 0.935 Satisfaction vs. Understandability = 0.074 Empathy vs. Comprehensiveness = 0.001 Empathy vs. Understandability = 0.001 Comprehensiveness vs. Understandability = 0.090 Bold values indicate statistical significance (p < 0.05) *: The overall index scores for all scenarios is derived by aggregating and averaging the scores from all areas across each scenario **: Wilcoxon Signed Ranks Test Except for empathy, parents rated ChatGPT’s responses as 4 or higher (≥ 4) more than 80% for each question in the other three categories. Considering the overall index scores calculated by summing and averaging the scores from all questions, the highest mean score was observed for understandability (4.5 ± 0.8). Understandability also had the highest overall index percentage of responses scoring ≥ 4 (86.0%), while the percentages for the other three categories ranged between 84.0% and 75.1% (Table 3.) (Fig. 2). The median overall index scores for the evaluation criteria ranged from 4.3 (2.6–5.0) for empathy to 4.8 (2.6–5.0) for understandability. Pairwise comparisons revealed statistically significant differences between satisfaction and empathy (p = 0.002), as well as between empathy and both comprehensiveness (p = 0.001) and understandability (p = 0.001), indicating that empathy was rated significantly lower than satisfaction, comprehensiveness, and understandability by parents (Table 2.). Table 3Evaluation of ChatGPT’s answers by parents according to the scoring rubric in terms of satisfaction, empathy, comprehensiveness and understandabilityParents, N = 45Parents with Likert Response of ≥ 4,n (%)Mean ± SDScoreMedian (Min-max) ScoreQuestion 1 Satisfaction33 (73.3)4.1 ± 0.94.0 (2.0–5.0) Empathy33 (73.3)3.6 ± 1.14.0 (1.0–5.0) Comprehensiveness34 (75.5)4.1 ± 0.94.0 (2.0–5.0) Understandability37 (82.2)4.2 ± 0.94.0 (1.0–5.0)Question 2 Satisfaction37 (82.2)4.3 ± 0.84.0 (2.0–5.0) Empathy33 (71.1)4.0 ± 1.04.0 (2.0–5.0) Comprehensiveness37 (82.2)4.2 ± 0.94.0 (2.0–5.0) Understandability36 (80.0)4.4 ± 0.95.0 (2.0–5.0)Question 3 Satisfaction34 (75.5)4.2 ± 0.94.0 (2.0–5.0) Empathy30 (66.7)4.0 ± 1.04.0 (2.0–5.0) Comprehensiveness35 (77.8)4.2 ± 0.94.0 (2.0–5.0) Understandability38 (84.4)4.4 ± 0.75.0 (3.0–5.0)Question 4 Satisfaction40 (88.9)4.4 ± 0.75.0 (3.0–5.0) Empathy34 (75.5)4.1 ± 1.04.0 (2.0–5.0) Comprehensiveness38 (84.4)4.5 ± 0.85.0 (3.0–5.0) Understandability38 (84.4)4.5 ± 0.85.0 (3.0–5.0)Question 5 Satisfaction37 (82.2)4.3 ± 0.84.0 (3.0–5.0) Empathy36 (80.0)4.2 ± 0.94.0 (2.0–5.0) Comprehensiveness36 (80.0)4.3 ± 0.85.0 (3.0–5.0) Understandability37 (82.2)4.4 ± 0.85.0 (2.0–5.0)Question 6 Satisfaction41 (91.1)4.5 ± 0.75.0 (3.0–5.0) Empathy35 (77.8)4.3 ± 0.85.0 (3.0–5.0) Comprehensiveness40 (88.9)4.5 ± 0.75.0 (3.0–5.0) Understandability41 (91.1)4.6 ± 0.75.0 (3.0–5.0)Question 7 Satisfaction41 (91.1)4.5 ± 0.75.0 (3.0–5.0) Empathy36 (80.0)4.3 ± 0.85.0 (3.0–5.0) Comprehensiveness40 (88.9)4.5 ± 0.75.0 (3.0–5.0) Understandability41 (91.1)4.6 ± 0.75.0 (2.0–5.0)Question 8 Satisfaction37 (82.2)4.4 ± 0.85.0 (3.0–5.0) Empathy33 (71.1)4.2 ± 1.05.0 (2.0–5.0) Comprehensiveness37 (82.2)4.5 ± 0.85.0 (3.0–5.0) Understandability38 (84.4)4.5 ± 0.95.0 (2.0–5.0)Question 9 Satisfaction39 (86.6)4.5 ± 0.85.0 (2.0–5.0) Empathy40 (88.9)4.5 ± 0.75.0 (3.0–5.0) Comprehensiveness41 (91.1)4.5 ± 0.95.0 (1.0–5.0) Understandability42 (93.3)4.6 ± 0.75.0 (2.0–5.0)Question 10 Satisfaction38 (84.4)4.5 ± 0.85.0 (3.0–5.0) Empathy39 (86.6)4.3 ± 0.95.0 (2.0–5.0) Comprehensiveness38 (84.4)4.5 ± 0.85.0 (2.0–5.0) Understandability39 (86.6)4.6 ± 0.85.0 (2.0–5.0) All QuestionsOverall Index Percentage of Responses Scoring ≥ 4, %*Overall Index Score**Overall Index Score** Satisfaction83.84.4 ± 0.84.5 (3.0–5.0) Empathy75.14.2 ± 0.94.3 (2.6-5.0) Comprehensiveness84.04.4 ± 0.84.5 (2.8-5.0) Understandability86.04.5 ± 0.84.8 (2.6-5.0)*: The percentage of responses scoring ≥ 4 across all ten questions are provided for each of the four evaluation areas**: The overall index scores for all questions are derived by aggregating and averaging the scores from all four areas across each question Table 3Evaluation of ChatGPT’s answers by parents according to the scoring rubric in terms of satisfaction, empathy, comprehensiveness and understandabilityParents, N = 45Parents with Likert Response of ≥ 4,n (%)Mean ± SDScoreMedian (Min-max) ScoreQuestion 1 Satisfaction33 (73.3)4.1 ± 0.94.0 (2.0–5.0) Empathy33 (73.3)3.6 ± 1.14.0 (1.0–5.0) Comprehensiveness34 (75.5)4.1 ± 0.94.0 (2.0–5.0) Understandability37 (82.2)4.2 ± 0.94.0 (1.0–5.0)Question 2 Satisfaction37 (82.2)4.3 ± 0.84.0 (2.0–5.0) Empathy33 (71.1)4.0 ± 1.04.0 (2.0–5.0) Comprehensiveness37 (82.2)4.2 ± 0.94.0 (2.0–5.0) Understandability36 (80.0)4.4 ± 0.95.0 (2.0–5.0)Question 3 Satisfaction34 (75.5)4.2 ± 0.94.0 (2.0–5.0) Empathy30 (66.7)4.0 ± 1.04.0 (2.0–5.0) Comprehensiveness35 (77.8)4.2 ± 0.94.0 (2.0–5.0) Understandability38 (84.4)4.4 ± 0.75.0 (3.0–5.0)Question 4 Satisfaction40 (88.9)4.4 ± 0.75.0 (3.0–5.0) Empathy34 (75.5)4.1 ± 1.04.0 (2.0–5.0) Comprehensiveness38 (84.4)4.5 ± 0.85.0 (3.0–5.0) Understandability38 (84.4)4.5 ± 0.85.0 (3.0–5.0)Question 5 Satisfaction37 (82.2)4.3 ± 0.84.0 (3.0–5.0) Empathy36 (80.0)4.2 ± 0.94.0 (2.0–5.0) Comprehensiveness36 (80.0)4.3 ± 0.85.0 (3.0–5.0) Understandability37 (82.2)4.4 ± 0.85.0 (2.0–5.0)Question 6 Satisfaction41 (91.1)4.5 ± 0.75.0 (3.0–5.0) Empathy35 (77.8)4.3 ± 0.85.0 (3.0–5.0) Comprehensiveness40 (88.9)4.5 ± 0.75.0 (3.0–5.0) Understandability41 (91.1)4.6 ± 0.75.0 (3.0–5.0)Question 7 Satisfaction41 (91.1)4.5 ± 0.75.0 (3.0–5.0) Empathy36 (80.0)4.3 ± 0.85.0 (3.0–5.0) Comprehensiveness40 (88.9)4.5 ± 0.75.0 (3.0–5.0) Understandability41 (91.1)4.6 ± 0.75.0 (2.0–5.0)Question 8 Satisfaction37 (82.2)4.4 ± 0.85.0 (3.0–5.0) Empathy33 (71.1)4.2 ± 1.05.0 (2.0–5.0) Comprehensiveness37 (82.2)4.5 ± 0.85.0 (3.0–5.0) Understandability38 (84.4)4.5 ± 0.95.0 (2.0–5.0)Question 9 Satisfaction39 (86.6)4.5 ± 0.85.0 (2.0–5.0) Empathy40 (88.9)4.5 ± 0.75.0 (3.0–5.0) Comprehensiveness41 (91.1)4.5 ± 0.95.0 (1.0–5.0) Understandability42 (93.3)4.6 ± 0.75.0 (2.0–5.0)Question 10 Satisfaction38 (84.4)4.5 ± 0.85.0 (3.0–5.0) Empathy39 (86.6)4.3 ± 0.95.0 (2.0–5.0) Comprehensiveness38 (84.4)4.5 ± 0.85.0 (2.0–5.0) Understandability39 (86.6)4.6 ± 0.85.0 (2.0–5.0) All QuestionsOverall Index Percentage of Responses Scoring ≥ 4, %*Overall Index Score**Overall Index Score** Satisfaction83.84.4 ± 0.84.5 (3.0–5.0) Empathy75.14.2 ± 0.94.3 (2.6-5.0) Comprehensiveness84.04.4 ± 0.84.5 (2.8-5.0) Understandability86.04.5 ± 0.84.8 (2.6-5.0)*: The percentage of responses scoring ≥ 4 across all ten questions are provided for each of the four evaluation areas**: The overall index scores for all questions are derived by aggregating and averaging the scores from all four areas across each question Evaluation of ChatGPT’s answers by parents according to the scoring rubric in terms of satisfaction, empathy, comprehensiveness and understandability *: The percentage of responses scoring ≥ 4 across all ten questions are provided for each of the four evaluation areas **: The overall index scores for all questions are derived by aggregating and averaging the scores from all four areas across each question Fig. 2“Distribution of overall index percentage of all questions’ responses scoring ≥ 4 across the four evaluation areas (based on parents’ assessment) “Distribution of overall index percentage of all questions’ responses scoring ≥ 4 across the four evaluation areas (based on parents’ assessment) Discussion This study evaluated ChatGPT’s responses to frequently asked questions about DMD from both expert and parental perspectives, revealing key insights into its strengths and limitations across multiple evaluation criteria. Notably, this is the first study to assess ChatGPT’s potential for providing information and guidance specifically to families of children with DMD, as evaluated by both parents and expert physicians. Access to accurate, comprehensible, and up-to-date health information is particularly crucial for families affected by rare diseases like DMD, where misinformation or lack of clarity can significantly impact decision-making and care strategies [27]. The findings provide an important contribution to understanding the role of AI-driven chatbots in healthcare communication and highlight their implications for improving patient education and clinical support in the context of complex and chronic diseases. In this study, experts and parents evaluated ChatGPT’s responses using different sets of criteria. Experts assessed responses based on Accuracy, Currency, Comprehensiveness, Usefulness, and Understandability, while parents evaluated them in terms of Satisfaction, Empathy, Comprehensiveness, and Understandability. Expert ratings indicated that ChatGPT’s responses were considered satisfactory in the range of 83–92% across these domains, whereas parental ratings showed satisfaction levels between 75 and 86%. Among the expert evaluations, accuracy received the highest score. This outcome can be attributed to ChatGPT’s ability to generate responses based on extensive medical literature, scientific research, and clinical guidelines [28]. Its access to well-documented and comprehensive sources, particularly regarding genetic and neuromuscular disorders like DMD, may contribute to the relatively high accuracy observed in expert evaluations. In the expert evaluations, the understandability parameter received the lowest score, whereas it was rated the highest among the parameters assessed by parents. Experts prioritized medical accuracy, the use of appropriate terminology, and scientific clarity in their assessments, which may have led them to rate complex or overly technical responses lower in terms of understandability [29]. In contrast, parents, who generally seek more fundamental and practically applicable information for everyday use, may have found ChatGPT’s responses more aligned with their expectations, leading to higher scores for understandability [30, 31]. This discrepancy between experts and parents likely stems from differences in knowledge levels, expectations, and evaluation criteria among the two groups. In the parental evaluation, the empathy parameter received the lowest score, which may be attributed to several factors. First, ChatGPT’s responses are generated using an objective and standardized language, lacking personalized or emotionally supportive expressions, which may have led parents to perceive them as insufficiently empathetic. AI-based systems do not incorporate nonverbal cues present in human interactions, such as tone, facial expressions, or gestures, making responses appear more mechanical and lacking emotional depth [32]. Additionally, parents of children with DMD often have high expectations for psychosocial support, and AI-generated responses may fall short of meeting these needs [33]. Furthermore, parents are accustomed to receiving medical information in conjunction with the guidance of a healthcare professional, since interactions with AI systems inherently differ from human communication and lack personalized emotional exchange; responses may appear less personal and emotionally engaging. Collectively, these factors may have contributed to empathy receiving a lower rating compared to other parameters. Recent studies conducted predominantly in common medical conditions have demonstrated that ChatGPT is generally well-received across various medical fields, with high evaluation scores reflecting its perceived reliability and effectiveness [34, 35]. Sarraju et al. reported that ChatGPT provided suitable responses to 84% of cardiovascular disease prevention queries, highlighting its significance in clinical decision support [36]. Similarly, Ayers et al. emphasized its empathetic approach and high-quality responses, suggesting that AI-powered conversational agents could enhance patient engagement [37]. Deiana et al. further reinforced ChatGPT’s potential as a reliable informational tool by demonstrating that it achieved 85.4% clarity and accuracy in answering vaccination-related queries [38]. Additionally, Yapar et al. found that ChatGPT exhibited significant potential in assisting orthopedic patients with home care by providing accurate and applicable medical recommendations. Across various scenarios, ChatGPT demonstrated 79.8% accuracy, 75.2% applicability, 70.6% comprehensiveness, and 75.6% communication clarity, further supporting its role as a valuable adjunct in healthcare guidance [25]. Against this background, and recognizing that evidence derived from common, data-rich conditions may not be directly generalizable to rare diseases, recent research has increasingly focused on evaluating the performance of LLMs specifically within rare disease contexts [39–44]. Recent evidence suggests that LLMs, including ChatGPT, can provide meaningful support in the context of rare diseases. A comparative benchmarking study has shown that ChatGPT can achieve diagnostic accuracy similar to clinical expert panels in genetically confirmed rare skeletal disorders, supporting its potential role as a complementary decision-support tool rather than a replacement for expert judgment [39]. Similarly, an evaluation of generative AI in rare disease–focused genetic counseling has found that ChatGPT generally offers accurate, professional, and non-harmful responses across areas such as diagnosis, prognosis, and patient counseling, while emphasizing the importance of proper clinical oversight [40]. Moreover, disease-specific assessments in rare oncologic and nephrologic populations indicate that ChatGPT can deliver empathetic and informative responses to patient-centered questions, although performance varies depending on question type and clinical complexity [41, 42, 44]. Emerging dialogue-based systems that incorporate phenotypic reasoning and Bayesian frameworks further demonstrate the potential of AI-assisted approaches in rare disease differential diagnosis, especially when used as an adjunct within structured clinical workflows [44]. A recent large-scale benchmarking study based on national rare disease catalogs has demonstrated that ChatGPT-4o can achieve high diagnostic accuracy for rare diseases, outperforming several open-source LLMs [43]. Overall, these findings suggest that although rare diseases present unique challenges for LLM-based systems, careful, supervised use of ChatGPT can still offer clinically relevant informational support in rare disease management. Rare diseases such as Duchenne muscular dystrophy are less represented in the data used to train large language models, which may limit the depth and consistency of disease-specific information generated by ChatGPT. Therefore, our findings should be interpreted as reflecting informational support rather than diagnostic decision-making. In rare disease settings, ChatGPT may assist with early understanding and communication, while diagnosis and long-term management should remain clinician-led. The median evaluation scores exceeding 4 across all responses in our study indicate that both experts and parents generally perceived ChatGPT’s responses positively. Families of DMD patients require accurate and accessible information regarding disease progression, potential complications, and appropriate care procedures [13]. At the same time, the increasing volume of patient inquiries has also driven a growing demand for digital health solutions among healthcare professionals. AI-based chatbots like ChatGPT have the potential to enhance the healthcare experience for both patients and experts by providing rapid responses and facilitating access to reliable medical information. While our study provides significant findings regarding ChatGPT’s capacity to deliver information and guidance to families of DMD patients, the results should be interpreted with caution. AI-based systems can play a supportive role in providing information and guidance to patients and their families; however, they should not replace human expertise. Particularly in chronic and progressive diseases such as DMD, AI-driven guidance applications raise ethical and safety concerns that must be carefully considered. Existing studies indicate that AI systems may generate biased responses, provide incomplete or inaccurate information, and fail to fully address patients’ individual needs [45–47]. This highlights the potential ethical implications and patient safety risks associated with excessive reliance on AI-driven systems [48]. Therefore, AI-based tools like ChatGPT should be positioned as complementary and supportive resources for healthcare professionals in the care of DMD patients, facilitating access to information rather than replacing clinical decision-making processes. This study has several limitations that should be acknowledged. First, the sample size was relatively small and limited to a single center, which may restrict the generalizability of the findings. Second, only one version of the AI model (ChatGPT-4o) was evaluated, and no comparisons were made with alternative models or updates that may yield different results. In addition, outputs generated by LLMs may vary depending on interaction conditions and system settings, which may limit the reproducibility of AI-generated responses. Third, the evaluation relied on subjective ratings from experts and parents, which might have introduced response bias. Fourth, the rubric-based evaluation approach used in this study is commonly applied in the literature, no formal scale was employed. We did not aim to develop a scale, and therefore no total scores were calculated, as doing so would have implied scale-based measurement. Nevertheless, the use of previously validated instruments could have strengthened the robustness of the evaluation and should be considered in future studies. In addition, domain selection was guided by existing literature, no formal theoretical validation, expert review or piloting, or assessment of content or face validity was performed, which may limit how comprehensively these domains reflect the intended constructs. Fifth, the questions evaluated in this study were developed by clinicians based on caregiver inquiries most frequently encountered in clinical practice; as such, they may not fully capture the range of questions caregivers might independently ask outside the context of direct healthcare provider interactions. The structured nature of the questions used in this study, along with the absence of direct patient and caregiver involvement, may have influenced the selection and framing of questions, potentially reflecting clinician perspectives more than the full range of real-world caregiver concerns. Moreover, the use of standardized and neutrally phrased questions may not fully reflect real-world interactions, where users may pose emotionally driven, leading, or context-dependent queries. In such scenarios, the adaptive and “people-pleasing” nature of large language models may influence the content and tone of responses. Additionally, selecting questions based on clinician experience may have introduced selection bias, as it may not fully reflect the diversity of concerns expressed in real-world settings such as online parent communities or support groups. Furthermore, the study did not include subspecialists who play a central role in the multidisciplinary management of DMD, such as cardiologists, pulmonologists, orthopedic surgeons or experts in neuromuscular genetics. The clinician sample was predominantly composed of physical medicine and rehabilitation specialists, representing only one component of the multidisciplinary care team, which may have introduced perspective bias and influenced the scope and depth of the evaluation. In addition, certain highly specialized aspects of DMD management, such as exon-skipping therapies, involve complex genetic considerations, including the presence of amenable out-of-frame deletions rather than simply mutations within specific exons. Such nuances may not be fully captured in generalized AI-generated responses, and the absence of subspecialists may have further affected the assessment of accuracy in these technically nuanced domains. Overall, the lack of multidisciplinary input may also have influenced the selection and framing of questions, potentially affecting the scope and relevance of the data. Finally, the study focused exclusively on written responses addressing informational needs at the time of diagnosis and early post-diagnostic stages, without evaluating interactive, longitudinal, or multimodal applications of AI systems. Additionally, cultural and linguistic factors may have influenced the perception of empathy and understandability, limiting the applicability of the findings across different populations. In conclusion, this study provides important insights into ChatGPT’s capacity to deliver informational support and guidance to families of patients with DMD. Our results suggest that ChatGPT may contribute to improving access to disease-related information and supporting families’ understanding of the disease at the time of diagnosis and patient–caregiver communication. To our knowledge, this is the first study to specifically evaluate ChatGPT’s performance in addressing the informational needs of families affected by DMD, representing an initial step toward understanding the role of LLMs in rare muscle disease contexts. While rare diseases are less represented in the data used to train LLMs, our results indicate that, when used appropriately, such tools may still offer meaningful support, particularly in settings with limited access to specialized healthcare services. The integration of AI-based informational tools into healthcare systems, under appropriate clinical oversight, has the potential to enhance access to reliable medical information, promote more informed health decisions, and contribute to broader public health efforts without replacing clinician-led diagnosis or long-term management. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.