Content and structural needs assessment for an artificial intelligence education mobile app in healthcare: a mixed methods study Authors: Seyyedeh Fatemeh Mousavi Baigi, Reyhane Norouzi Aval, Masoumeh Sarbaz, Seyyed Mohammad Tabatabaei, Khalil Kimiafar Published: 2025 ABSTRACT ObjectiveThis study aimed to identify and prioritize the core content and structural requirements for developing a high-quality mobile app designed to teach AI concepts and skills in a healthcare context.MethodsA mixed-methods design was employed. First, two systematic reviews were conducted: [1] a review of scholarly articles to extract educational frameworks for AI in healthcare, and [2] a review of 47 AI education apps from three app stores (Google Play, App Store, Café Bazaar), assessed using the Mobile App Rating Scale (MARS). As no healthcare-specific AI education apps were found during the search, general-purpose AI learning apps were included, which constitutes a limitation in terms of domain specificity. Based on these insights, a preliminary content framework was developed and validated by 12 experts in medical informatics and health information management. Subsequently, a structural needs assessment was carried out with 97 healthcare students using custom-designed questionnaires. Open-ended responses were analyzed using Braun and Clarke’s thematic analysis method.ResultsThe systematic review of 37 articles revealed 10 key domains essential for AI education in healthcare, including foundational knowledge, data science, practical clinical applications, ethics, and communication. The app review showed a mean MARS quality score of 2.92 out of 5, highlighting significant deficiencies in content coherence, interactivity, and privacy implementation. Expert validation confirmed all proposed domains, and thematic analysis of expert feedback led to the inclusion of an additional domain: Practical Tools and Platforms. Healthcare students strongly favored features such as interactive learning, offline functionality, and personalized learning paths (mean scores > 4.76/5), with no significant differences across gender or field of study.ConclusionThis study presents a validated, evidence-based framework for developing a healthcare-focused AI education app. The finalized structure includes 11 content domains and 20 prioritized structural features aimed at promoting practical, ethical, and engaging learning experiences. The findings underscore the urgent need for structured, user-centered digital tools to prepare healthcare students and professionals for the responsible integration of AI into clinical practice.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12909-025-08328-8. Introduction In recent years, Artificial Intelligence (AI), particularly machine learning and deep learning algorithms, has experienced significant growth, leading to widespread applications across educational and professional domains [1, 2]. As AI continues to transform healthcare and other sectors, the demand for comprehensive, specialized training to prepare the future workforce for the effective integration of AI into clinical practice has intensified. AI education must extend beyond theoretical knowledge to include the development of practical competencies and the ability to utilize AI-based tools effectively in healthcare environments [3]. While AI education in healthcare has been widely studied from a global perspective, integrating local perspectives is crucial for designing effective interventions. For example, recent research by Younesian et al. (2025) on Iranian physicians’ views regarding AI’s role in healthcare highlights regional concerns about trust and AI’s potential to augment or replace human expertise. These insights demonstrate the need for AI education tools that are not only globally relevant but also culturally and professionally tailored to meet the specific needs of healthcare professionals [4]. Despite its growing importance, recent studies have highlighted insufficient levels of AI knowledge, attitudes, and practical skills among healthcare professionals and students, although many hold a positive outlook on the future of AI in healthcare [5]. A systematic review found that more than half of healthcare students reported low levels of AI knowledge and skills and limited hands-on experience, underscoring the urgent need for practical, application-focused AI training in health education [5]. Another study reported low AI literacy among healthcare professionals and students, with significant improvements following targeted educational interventions [4]. These findings highlight the critical need to incorporate effective, accessible, and practice-oriented AI education into healthcare curricula. To address these gaps, educational mobile applications (apps) have emerged as effective tools that offer interactive, personalized, and flexible access to AI training, enhancing the reach and impact of educational programs [6, 7]. These apps employ multimedia formats, including text, images, videos, and podcasts, to create engaging learning experiences and provide learners with convenient access to high-quality educational resources anytime and anywhere [8, 9]. Despite the proliferation of AI educational apps, their quality and content consistency remain variable, necessitating systematic evaluation using standardized tools. The Mobile App Rating Scale (MARS) provides a comprehensive framework for evaluating app quality across dimensions such as engagement, functionality, aesthetics, content quality, and subjective user experience [10]. Developing effective AI education tools requires a clear understanding of the content and structural needs of the target users. To inform the design of our mobile app for AI education in healthcare, we first conducted a systematic review of the literature to extract and synthesize existing frameworks and key domains for AI education, identifying essential competencies, instructional strategies, and ethical considerations pertinent to AI literacy in healthcare. We then performed a systematic review of existing AI educational mobile apps across three major app stores, systematically evaluating them using the MARS framework to assess current gaps and best practices in app-based AI education. Based on insights from the literature and app evaluations, we developed a preliminary content and structural framework for an AI education app in healthcare. To validate this framework, we designed a structured questionnaire to gather expert feedback on content priorities and to assess structural usability with healthcare students. Content validity was confirmed through expert review, while structural usability and acceptability were assessed by healthcare students to ensure the design aligns with user needs and practical app in learning environments. Given the accelerating integration of AI in healthcare, this study aims to bridge the gap between theoretical knowledge and practical competency by identifying and prioritizing content and structural elements critical for developing an effective AI education mobile app for healthcare students and professionals. By aligning educational content with evidence-based frameworks and user-centered needs assessments, this study contributes to the development of a high-quality, accessible, and practical AI educational intervention, supporting the advancement of AI competencies within healthcare education and clinical practice. Methods Systematic review The initial objective of this study was to identify and review existing studies and mobile apps specifically designed for AI education within the healthcare domain. However, our systematic searches revealed no studies or mobile apps exclusively focused on AI education in healthcare. Consequently, we expanded our scope to include general AI educational mobile apps regardless of domain while concurrently reviewing studies proposing content frameworks and curricular guidelines for AI education within healthcare settings. The systematic review was conducted on January 31, 2025. Two researchers with expertise in health information technology and prior training in mobile app evaluation using the MARS systematically searched for AI educational mobile apps across three official app stores: Google Play, Cafe Bazaar, and the Apple App Store. The search was conducted using the keywords: “Artificial Intelligence”, “AI”, “Machine Learning”, and “Deep Learning” in both English and Persian. In parallel, a systematic review was conducted across four major databases: Scopus, PubMed, Embase, and Web of Science to identify studies that proposed or described content frameworks, curricular models, or competency guidelines for AI education in healthcare. Three concept blocks of keywords were searched across the title, abstract, and keyword fields of articles, combining the following terms: (“artificial intelligence education” OR “AI education” OR “artificial intelligence literacy” OR “AI literacy”) AND (“educational framework” OR “curriculum” OR “content framework” OR “curriculum development” OR “training framework” OR “educational programs”) AND (“healthcare” OR “health care” OR “health” OR “medical education” OR “health professions education” OR “clinical education”). Equivalent search structures were adapted for each database using platform-specific syntax, such as MeSH terms in PubMed and Emtree terms in Embase, when applicable. The full, database-specific search strategies are available in Supplementary Materials 1. Eligibility criteria For the app store review, mobile apps were included if their primary objective was to teach concepts and applications related to AI. Mobile apps that focused on topics unrelated to AI or machine learning were excluded. Additionally, apps in languages other than Persian and English, those with significant technical issues, or those requiring unreasonably high download costs were excluded. However, mobile apps offering free initial download and use with partial paid content remained eligible for evaluation in this study. For the systematic review, studies were included if they proposed or described content frameworks, curricular guidelines, or structured educational models specifically for AI education within the healthcare domain. Studies addressing general AI education frameworks outside healthcare were excluded. Additionally, letters to the editor and studies that did not present a structured content framework or educational program were excluded. Selection of studies and mobile apps The titles and download pages of the identified mobile apps were independently screened by two reviewers to determine eligibility. Eligible mobile apps were entered into a centralized database, and duplicates were removed. The remaining mobile apps were then downloaded and reviewed in detail. Disagreements between reviewers were resolved through discussion, and when necessary, a third reviewer (supervisor) was consulted for final decisions. For the systematic review, independent screening of titles and abstracts was performed according to the predefined eligibility criteria. Articles not meeting the inclusion criteria were excluded. The full texts of potentially relevant studies were retrieved and assessed for eligibility independently by two researchers. Discussions were conducted to resolve disagreements, and when a consensus was not reached, the third author made the final decision. Data extraction and quality assessment of mobile apps Two health information technology specialists independently evaluated the identified mobile apps using the MARS, whose Persian version has demonstrated high reliability (coefficient = 0.94) [11]. Prior to the assessment, the reviewers watched the official MARS training video and received the necessary instructions to ensure consistency in evaluations [12]. Each app was assessed using MARS’s 23 items across five sections (Engagement, Functionality, Aesthetics, Information Quality, Subjective Quality), and six perceived impact items evaluating the app’s influence on knowledge, attitudes, intention to change, and behavioral change likelihood. Each item was scored on a 5-point scale (1 = inadequate, 5 = excellent), with the overall quality score calculated as the mean across the five primary sections. For structured data extraction, a standardized checklist was applied to the mobile apps, capturing: app name, version, year (last version), platform (Café Bazaar, Google Play, App Store), paid content (free or in-app purchases), downloads, developer, affiliation (academic/commercial), theoretical background/strategies, core educational topics, technical and privacy aspects, privacy statements, app functionality description, user ratings, and selected comments. Data analysis included calculating the mean scores assigned by both reviewers for each app. Descriptive statistics (means, standard deviations) were used to summarize app quality overall and by MARS sections, facilitating a clear comparison of app quality and content alignment across evaluated mobile apps. Data extraction and quality assessment of included studies To ensure methodological rigor, all included studies in the systematic review were appraised using a structured quality assessment protocol based on four predefined criteria: [1] clarity of the research objective [2], appropriateness of the study design [3], consistency between data analysis and study design, and [4] sufficiency and relevance of supporting evidence for the conclusions. Only studies that satisfactorily met al.l four criteria were included in the final review. Two independent reviewers conducted the assessments. Disagreements were resolved through discussion, with a third reviewer consulted when necessary. This approach ensured the inclusion of studies with sound methodological foundations, thereby enhancing the validity and reliability of the overall synthesis. A structured checklist was also used to extract key data from the included studies, including: study reference, country of origin, year of publication, study method, target population, and core educational topics related to AI education in healthcare. Sample size The research population included healthcare students at Mashhad University of Medical Sciences, across all available fields. Given the large total student population (N = 913), the required sample size was calculated using the Cochran formula for limited populations, considering α = 0.05, p = q = 0.5 (to maximize the sample size), and a tolerable error rate of d = 0.1. Accordingly, the calculated sample size for healthcare students was 87 individuals to ensure sufficient statistical power and representativeness in the content and structural needs assessment of the proposed AI education mobile app.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:n=\frac{Nqp{z}^{2}}{{Nd}^{2}+pq{z}^{2}}$$\end{document} To account for an anticipated 10% attrition rate among student participants, the final number of students invited to participate in the study was adjusted as:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:n\:\left(\text{a}\text{d}\text{j}\text{u}\text{s}\text{t}\text{e}\text{d}\right)=\frac{1}{1-\text{a}\text{t}\text{t}\text{r}\text{i}\text{t}\text{i}\text{o}\text{n}\:\text{r}\text{a}\text{t}\text{e}}$$\end{document} Therefore, 97 students were invited to participate in the survey. As a motivational incentive, participants were informed that they would receive free access to the final version of the designed AI educational mobile app upon completion, encouraging participation and reducing dropout likelihood. For the expert content validation phase, the entire accessible faculty population in the fields of Health Information Management and Medical Informatics (n = 12 faculty members) was invited to participate using purposive total population sampling due to the limited number of domain experts and to ensure maximum content coverage in refining the educational app framework. Given the small size, no sample size calculation was applied in this phase, and all eligible experts were included to ensure the validity and comprehensiveness of the qualitative thematic analysis. Data collection instruments The instruments used in this study consisted of two researcher-made questionnaires, which were developed based on a review of 47 mobile apps for AI education in health and 37 framework articles. Mobile App Content Validation Questionnaire. This questionnaire was designed to validate and assess the educational content needs of the mobile app for AI education in health from the perspective of medical informatics specialists and health information management specialists. The tool included: A demographic section (specialty, gender, age, years of experience).Ten content sections aligned with the framework extracted from the systematic review. A demographic section (specialty, gender, age, years of experience). A demographic section (specialty, gender, age, years of experience). Ten content sections aligned with the framework extracted from the systematic review. Ten content sections aligned with the framework extracted from the systematic review. Respondents were asked to rate the importance and necessity of each section based on their experience using a 1–5 scale, and at the end, to provide additional suggestions for improvement. Mobile App Structural Needs Assessment Questionnaire. This questionnaire was designed to assess the structural needs of the mobile app for AI education in health from the perspective of healthcare students and included: A demographic section (age, gender, field of study).Items regarding digital learning tools used by students.Attitudes toward learning via the mobile app for AI education (10 items rated on a 5-point Likert scale).Assessment of the structural feature needs of the mobile app for AI education (20 features rated on a 1–5 scale). A demographic section (age, gender, field of study). A demographic section (age, gender, field of study). Items regarding digital learning tools used by students. Items regarding digital learning tools used by students. Attitudes toward learning via the mobile app for AI education (10 items rated on a 5-point Likert scale). Attitudes toward learning via the mobile app for AI education (10 items rated on a 5-point Likert scale). Assessment of the structural feature needs of the mobile app for AI education (20 features rated on a 1–5 scale). Assessment of the structural feature needs of the mobile app for AI education (20 features rated on a 1–5 scale). Validity and reliability The face and content validity of the questionnaires were evaluated by five experts (two health information management specialists, two medical informatics specialists, and one biostatistics specialist), and necessary modifications were applied. Subsequently, content validity (CVI and CVR) was calculated using the Delphi method with the participation of six experts (two PhD-level AI specialists, two medical informatics specialists, and two health information management specialists), yielding a CVI value of 0.98. The reliability of the instruments was assessed using Cronbach’s alpha, resulting in a value of 0.89 for the content validation questionnaire and 0.76 for the structural needs assessment questionnaire. The questionnaires were collected in-person, with the researcher present to clarify any ambiguities during completion. The content and structural needs assessment questionnaires are available in Supplementary Materials 2 and 3, respectively. Statistical analysis The obtained quantitative data were aggregated and summarized using descriptive statistics to provide a clear basis for inference and interpretation. The chi-square test was applied to examine the independence between qualitative variables. All quantitative analyses were conducted using SPSS Version 16. For qualitative analysis, open-ended questions were analyzed using thematic content analysis to interpret the meaning of responses provided by professors and students. The analysis followed the Braun and Clarke method of thematic analysis [13, 14]. Responses were initially coded using pre-established codes identified in prior studies, and in instances where responses did not align with existing codes, new codes were generated by the research team. To ensure reliability of the coding, a peer review approach was employed [15, 16]. Two peer reviewers with expertise in medical informatics were invited to review and confirm the generated codes. In cases where disagreements occurred between the two reviewers, a third reviewer was engaged to resolve discrepancies and finalize the codes. The intercoder agreement rate for both questions exceeded 80%. Following the completion of the coding process, similar codes were grouped into overarching themes for structured interpretation [13]. Results Systematic review A total of 247 records were retrieved through database searches. After removing 67 duplicates, 180 unique records were screened based on titles and abstracts. Of these, 140 were excluded for not meeting the inclusion criteria. Forty full-text articles were sought for retrieval, all of which were successfully obtained and assessed for eligibility. Three studies were excluded due to a lack of specific focus on AI education within the healthcare domain, resulting in the inclusion of 37 studies, representing 47 individual reports. In parallel, 47 mobile apps were identified from three official app stores (Google Play, Café Bazaar, and Apple App Store). All identified apps were retrieved, assessed in full, and included in the evaluation, as they met the predefined eligibility criteria. The complete selection process is illustrated in the PRISMA flow diagram (Fig. 1).Fig. 1PRISMA flow diagram of the study selection PRISMA flow diagram of the study selection Descriptive characteristics of the included mobile apps A total of 47 mobile apps designed for AI education were included for evaluation in this study. All identified apps met the eligibility criteria and were retained for detailed assessment. Among these, 23 apps were available on the App Store/iOS, 12 apps on Google Play, and 12 apps on Café Bazaar. All apps were developed and published by commercial developers or companies, with none supported directly by universities or governmental institutions. In terms of accessibility, 33 apps were offered free of charge, while 14 apps required in-app purchases or paid versions for full access to content. The number of downloads varied widely, ranging from over 10 to more than 100,000 downloads, indicating differences in user reach and popularity across platforms. All the evaluated apps were designed to deliver theoretical and practical education in AI; however, none were specifically developed to teach AI within the healthcare domain. Regarding privacy and data protection, fewer than 15% of the apps included clear privacy statements or options for creating secure user accounts. Among the reviewed apps, Artificial Intelligence by Softecks, Learn AI & ML with Python by Coding & Programming, and Artificial intelligence__AL by Woow apps had the highest recorded downloads, each exceeding 100,000 installations, indicating a higher level of user engagement and interest. Table S1 presents a detailed summary of the included mobile apps, highlighting their names, platforms, release years, download counts, payment status, and developer information. Table S2 outlines the privacy statements, technical specifications, and theoretical frameworks adopted by these apps. Table S3 summarizes the user ratings, selected user comments, primary educational topics, and functional descriptions, providing insights into the user experience and the educational content offered by the evaluated mobile apps. App quality assessment results The mean quality scores of the evaluated AI learning mobile apps ranged from 1.80 to 4.10, with an overall mean of 2.92 (SD: 0.58) on a 1 (inadequate) to 5 (excellent) scale. The highest domain-specific score was observed in functionality (mean: 3.12, SD: 0.63), followed by information quality (mean: 3.05, SD: 0.59). Lower scores were found in aesthetics (mean: 2.94, SD: 0.61) and subjective quality (mean: 2.76, SD: 0.65), indicating these as areas where the evaluated apps may require improvement to enhance user engagement and satisfaction. Among the evaluated apps, Learn AI with Python and Mila, AI Learning achieved the highest overall quality scores (mean: 4.10, SD: 0.37), demonstrating strong performance across engagement, functionality, aesthetics, and content quality. In contrast, AI & ChatGPT Training (FA) and Learn AI & ML with Python (Google Play) received the lowest overall scores (mean: 1.80, SD: 0.50), reflecting limitations in design, content depth, and user-perceived value. The perceived impact domain demonstrated a moderate average score (mean: 2.79, SD: 0.57), reflecting the apps’ influence on users’ knowledge acquisition, attitudes toward AI, and behavioral intentions regarding AI-related practices. Despite relatively acceptable functionality and information scores, the lower subjective quality ratings highlight that many apps may not fully meet user expectations in terms of motivation and sustained engagement. Artificial Intelligence by Softecks, Learn AI & ML with Python by Coding & Programming, and Artificial intelligence__AL by Woow Apps had the highest recorded download numbers (over 100,000 installs), aligning with higher user interest and potential utility despite variability in MARS ratings across these apps. Supplementary Table S4 presents detailed MARS scores across engagement, functionality, aesthetics, information quality, subjective quality, and perceived impact for each evaluated app, providing a comprehensive overview of the strengths and areas for improvement within these AI education tools. Overview of included studies A total of 37 studies were identified through a systematic review of major databases [17–53]. The methodologies employed across these studies included 16 systematic and narrative reviews [17, 19, 21, 26, 28, 29, 32, 39, 43, 45, 48, 51–53] and three Delphi studies [25, 30, 47]. Other methodologies encompassed workshop development and evaluation, cross-sectional surveys, the design and assessment of educational interventions, perspective analyses, and framework-based mixed-methods approaches [18, 20, 22–24, 27, 31, 33–38, 40, 41, 44, 46, 49, 50]. Among these, seven studies were conducted in low- and middle-income countries (LMICs), including the Philippines, Turkey, Colombia, Pakistan, India, UAE/India, and Oman [46–51, 53], while 28 studies were conducted in high-income countries (HICs) such as Canada, the United States, Germany, Sweden, the Netherlands, the United Kingdom, South Korea, Hong Kong, Japan, Finland, and Singapore [17–45]. A comprehensive overview of the included studies, detailing the first author, year, country, core educational topics, study methodology, and target population, is presented in Table S5. Development of an adaptive core content framework Based on the systematic review of 37 included studies [17–53] and the comprehensive evaluation of AI educational mobile apps, a unified adaptive framework of core content topics for AI education in healthcare was developed. This framework integrates content derived from the highest-quality studies and systematically mapped mobile app functionalities, ensuring alignment with real-world competencies needed for medical learners and healthcare professionals. The framework encompasses ten major domains, including foundational AI concepts, data science literacy, practical healthcare applications, ethical and legal considerations, communication and humanistic skills, applied technical skills, curriculum integration strategies, advanced AI topics, critical evaluation of AI tools, and research enhancement using AI. Each domain is further divided into targeted subcategories informed by both systematic review and mobile app content analysis to ensure a pragmatic, learner-centered structure for implementation in medical curricula. A detailed mapping of these domains, corresponding subcategories, associated mobile apps, and aligned references is presented in Table S6. Expert content needs assessment for the AI in health education mobile app A total of 12 health information management and medical informatics experts were invited to participate in the quantitative content validation of the AI in Health Education mobile app. Ten experts completed the questionnaire, yielding a response rate of 83.3%. The results are presented in two sections: demographic characteristics and content section prioritization. Demographic characteristics Participants had a mean age of 42.6 years (SD = 9.5) with an average work experience of 11.1 years (SD = 6.8). The gender distribution was 60% male and 40% female, and the specialty distribution included 50% from medical informatics and 50% from health information management (Table 1).Table 1Demographic characteristics of participantsVariableCategoryFrequency (n)Percentage (%)GenderMale660Female440SpecialtyMedical Informatics550Health Information Management550Mean Age (years)42.6 (SD 9.5)-Mean Work Experience (years)11.1 (SD 6.8)- Table 1Demographic characteristics of participantsVariableCategoryFrequency (n)Percentage (%)GenderMale660Female440SpecialtyMedical Informatics550Health Information Management550Mean Age (years)42.6 (SD 9.5)-Mean Work Experience (years)11.1 (SD 6.8)- Demographic characteristics of participants Content section prioritization Experts rated 10 structured content sections extracted from the prior systematic review using a 1–5 scale for importance and necessity in the AI in Health Education curriculum. All sections received high ratings, with means ranging from 4.3 to 5.0, indicating a shared consensus among experts on the relevance of the proposed educational framework. The highest-rated sections were: Basic Concepts of AI and Machine Learning in Health (M = 5.0, SD = 0.0),Data Literacy and Data Science (M = 4.9, SD = 0.3),Practical Applications of AI in Health (M = 4.9, SD = 0.3). Basic Concepts of AI and Machine Learning in Health (M = 5.0, SD = 0.0), Basic Concepts of AI and Machine Learning in Health (M = 5.0, SD = 0.0), Data Literacy and Data Science (M = 4.9, SD = 0.3), Data Literacy and Data Science (M = 4.9, SD = 0.3), Practical Applications of AI in Health (M = 4.9, SD = 0.3). Practical Applications of AI in Health (M = 4.9, SD = 0.3). The lowest-rated section was Advanced Topics in AI, while still highly rated, at (M = 4.3, SD = 0.5). Table S7 presents the means and standard deviations of expert ratings for the proposed content sections to be included in the AI in Health Education content framework. Inferential analysis Fisher’s exact test was performed to evaluate potential associations between gender, specialty, and content prioritization scores across the ten sections. The analysis revealed no statistically significant differences across gender (p = 0.89) or specialty (p = 0.76), confirming that experts shared a unified perception regarding the necessity and importance of the proposed content regardless of their background. Qualitative thematic analysis of expert recommendations The open-ended qualitative responses of 10 health informatics and health information management specialists were analyzed using Braun and Clarke’s thematic analysis to refine the initial AI in Health Education framework. Pre-established codes were derived from the systematic review of 47 AI education apps and 37 content framework papers. Experts were invited to suggest additional modules or refinements to improve the educational app framework for teaching AI in health. Ten primary themes emerged from the analysis, representing the most frequently emphasized and practically relevant content areas recommended by experts. As illustrated in Fig. 2, Health Data Management & Infrastructure was the most prominent theme, followed by Practical Tools and Platform Training, and Transparency and Explainability. This distribution highlights the experts’ strong preference for curricula that are hands-on, technically grounded, and ethically conscious—especially in the context of real-world AI integration in healthcare.Fig. 2Frequency distribution of expert-suggested themes for AI in health education app content Frequency distribution of expert-suggested themes for AI in health education app content Theme 1: Health data management & infrastructure Experts emphasized the need to train learners on practical health data management, including Big Data handling, lifecycle data governance, and the use of FHIR, HL7, SQL HealthVault, and OMOP standards. These additions were proposed to ensure learners gain practical skills to manage health datasets within AI workflows. “Teaching FHIR and HL7 standards is essential for interoperability in health AI systems.” (Expert #2). “Teaching FHIR and HL7 standards is essential for interoperability in health AI systems.” (Expert #2). Theme 2: Practical tools and platform training Participants recommended incorporating practical training in open-source platforms, including TensorFlow, PyTorch, and prompt engineering for LLMs and medical NLP applications, aligning with current AI development practices. “Hands-on training with TensorFlow and PyTorch should be part of the curriculum to facilitate real-world skill building.” (Expert #3). “Hands-on training with TensorFlow and PyTorch should be part of the curriculum to facilitate real-world skill building.” (Expert #3). Theme 3: Transparency and explainability Several experts highlighted the need to cover Explainable AI (XAI), algorithm transparency, and interpretable clinical models to address critical ethical and practical considerations in health AI adoption. “A module on Explainable AI in medicine is necessary.” (Expert #2). “A module on Explainable AI in medicine is necessary.” (Expert #2). Theme 4: Tool evaluation and quality Recommendations included adding simulation-based learning, sandbox environments, and structured evaluation frameworks to allow learners to evaluate AI models using practical case studies (e.g., Epic Sepsis Model, AI Radiology). “Simulation-based learning with case studies can enhance practical understanding.” (Expert #7). “Simulation-based learning with case studies can enhance practical understanding.” (Expert #7). Theme 5: Data governance and legal aspects Participants emphasized incorporating GDPR, ISO/IEC 42,001 frameworks, and local data protection laws into training, ensuring learners are aware of legal and regulatory requirements. “Local and international privacy standards like GDPR should be covered.” (Expert #6). “Local and international privacy standards like GDPR should be covered.” (Expert #6). Theme 6: Practical training and simulation Establishing sandbox environments for safe AI experimentation with real health data and managing data cleaning, abstraction, and governance projects were suggested to enhance experiential learning. “Sandbox environments enable safe, practical AI learning.” (Expert #6). “Sandbox environments enable safe, practical AI learning.” (Expert #6). Theme 7: Ethics and standards training. The inclusion of consistent ethical guidelines, digital ethics, and considerations of algorithmic bias and fairness was suggested for responsible AI use in health contexts. “Ethical implications must be integrated across all modules.” (Expert #9). “Ethical implications must be integrated across all modules.” (Expert #9). Theme 8: Edge AI and IoMT applications The adoption of Lightweight Models and Edge AI for healthcare, as well as IoMT (Internet of Medical Things) integration, was proposed to enhance real-world relevance. “Edge AI applications should be discussed for practical scenarios in remote health monitoring.” (Expert #5). “Edge AI applications should be discussed for practical scenarios in remote health monitoring.” (Expert #5). Theme 9: Fundamentals based on references Experts emphasized using authoritative references such as Russell & Norvig’s AI book and coverage of intelligent agents and their environments to strengthen theoretical grounding. “Foundations should use references like Russell’s AI book to ensure standardization.” (Expert #8). “Foundations should use references like Russell’s AI book to ensure standardization.” (Expert #8). Theme 10: Change management and AI culture Experts recommended adding training on organizational resistance management and building a culture of AI adoption within health systems to improve implementation outcomes. “Managing organizational resistance is critical for AI adoption.” (Expert #4). “Managing organizational resistance is critical for AI adoption.” (Expert #4). Figure 2. Frequency distribution of expert-suggested themes for AI in health education app content. Final content framework for AI in health education app Based on thematic analysis of expert feedback using Braun & Clarke methodology, the initial 10-category framework was refined to an 11-category structure by adding “Practical Tools and Platform Training” as a separate category, ensuring focused module development on hands-on, platform-specific skill-building in AI for health (Table 2).Table 2Final content framework for AI in health education appMain CategorySubcategories1. Foundations & Basic Concepts of AI & ML1.1 Definitions, History, Types of AI & ML1.2 Supervised, Unsupervised, Reinforcement Learning1.3 Deep Learning (ANN, CNN, RNN, Transformers)1.4 Generative Models and LLMs1.5 Intelligent Agents & Environments (Russell Reference)2. Data Science & Health Data Management2.1 Data Preprocessing and Cleaning2.2 Feature Engineering and Data Analysis2.3 Programming with Python/R2.4 Data Visualization2.5 Big Data Handling & Infrastructure (FHIR, HL7, SQL HealthVault, OMOP)2.6 Data Lifecycle Governance3. Practical Tools and Platforms3.1 Hands-on TensorFlow, PyTorch Training3.2 Prompt Engineering for LLMs and NLP3.3 Open-source Platforms for AI Development3.4 Lightweight Models & Edge AI for IoMT4. Practical Applications of AI in Health4.1 Clinical Decision Support & Diagnostics4.2 Personalized Medicine & Patient Data Analysis4.3 Simulation and Scenario-based Learning4.4 Use of ChatGPT and Clinical Reasoning Assistants4.5 Sandbox Labs for Safe AI Practice5. Ethics, Law, and Governance in AI5.1 Privacy and Data Security (GDPR, ISO/IEC 42001)5.2 Algorithm Transparency & Accountability5.3 Algorithmic Bias and Fairness5.4 Legal and Societal Implications5.5 Digital Ethics Integration Across Modules6. Communication & Humanistic Skills in AI6.1 Explaining AI to Patients6.2 Collaborative AI Use in Health Teams6.3 Digital Literacy and Empathy6.4 Change Management and Trust-Building in AI Culture7. Practical & Applied AI Skills7.1 Designing & Implementing AI Models7.2 Prompt Engineering & Using AI Tools7.3 Real-World Projects with Health Data8. Enhancing Medical Education Quality with AI8.1 Designing AI Curriculum for Medical Education8.2 Faculty Empowerment in AI Teaching8.3 Personalized Learning via AI9. Advanced Topics9.1 Natural Language Processing (NLP)9.2 Computer Vision & Image Processing9.3 Expert Systems & AGI9.4 Transfer Learning & Explainable Models10. Evaluation & Critical Analysis of AI Tools10.1 Performance Analysis & Limitations10.2 Evaluation of Practical Utility & Usability10.3 Research Articles & Critical Appraisal Skills11. Enhancing Research Quality with AI11.1 AI for Systematic Literature Reviews11.2 AI for Quantitative & Qualitative Data Analysis11.3 AI for Scientific Writing and Editing Table 2Final content framework for AI in health education appMain CategorySubcategories1. Foundations & Basic Concepts of AI & ML1.1 Definitions, History, Types of AI & ML1.2 Supervised, Unsupervised, Reinforcement Learning1.3 Deep Learning (ANN, CNN, RNN, Transformers)1.4 Generative Models and LLMs1.5 Intelligent Agents & Environments (Russell Reference)2. Data Science & Health Data Management2.1 Data Preprocessing and Cleaning2.2 Feature Engineering and Data Analysis2.3 Programming with Python/R2.4 Data Visualization2.5 Big Data Handling & Infrastructure (FHIR, HL7, SQL HealthVault, OMOP)2.6 Data Lifecycle Governance3. Practical Tools and Platforms3.1 Hands-on TensorFlow, PyTorch Training3.2 Prompt Engineering for LLMs and NLP3.3 Open-source Platforms for AI Development3.4 Lightweight Models & Edge AI for IoMT4. Practical Applications of AI in Health4.1 Clinical Decision Support & Diagnostics4.2 Personalized Medicine & Patient Data Analysis4.3 Simulation and Scenario-based Learning4.4 Use of ChatGPT and Clinical Reasoning Assistants4.5 Sandbox Labs for Safe AI Practice5. Ethics, Law, and Governance in AI5.1 Privacy and Data Security (GDPR, ISO/IEC 42001)5.2 Algorithm Transparency & Accountability5.3 Algorithmic Bias and Fairness5.4 Legal and Societal Implications5.5 Digital Ethics Integration Across Modules6. Communication & Humanistic Skills in AI6.1 Explaining AI to Patients6.2 Collaborative AI Use in Health Teams6.3 Digital Literacy and Empathy6.4 Change Management and Trust-Building in AI Culture7. Practical & Applied AI Skills7.1 Designing & Implementing AI Models7.2 Prompt Engineering & Using AI Tools7.3 Real-World Projects with Health Data8. Enhancing Medical Education Quality with AI8.1 Designing AI Curriculum for Medical Education8.2 Faculty Empowerment in AI Teaching8.3 Personalized Learning via AI9. Advanced Topics9.1 Natural Language Processing (NLP)9.2 Computer Vision & Image Processing9.3 Expert Systems & AGI9.4 Transfer Learning & Explainable Models10. Evaluation & Critical Analysis of AI Tools10.1 Performance Analysis & Limitations10.2 Evaluation of Practical Utility & Usability10.3 Research Articles & Critical Appraisal Skills11. Enhancing Research Quality with AI11.1 AI for Systematic Literature Reviews11.2 AI for Quantitative & Qualitative Data Analysis11.3 AI for Scientific Writing and Editing Final content framework for AI in health education app 1.1 Definitions, History, Types of AI & ML 1.2 Supervised, Unsupervised, Reinforcement Learning 1.3 Deep Learning (ANN, CNN, RNN, Transformers) 1.4 Generative Models and LLMs 1.5 Intelligent Agents & Environments (Russell Reference) 2.1 Data Preprocessing and Cleaning 2.2 Feature Engineering and Data Analysis 2.3 Programming with Python/R 2.4 Data Visualization 2.5 Big Data Handling & Infrastructure (FHIR, HL7, SQL HealthVault, OMOP) 2.6 Data Lifecycle Governance 3.1 Hands-on TensorFlow, PyTorch Training 3.2 Prompt Engineering for LLMs and NLP 3.3 Open-source Platforms for AI Development 3.4 Lightweight Models & Edge AI for IoMT 4.1 Clinical Decision Support & Diagnostics 4.2 Personalized Medicine & Patient Data Analysis 4.3 Simulation and Scenario-based Learning 4.4 Use of ChatGPT and Clinical Reasoning Assistants 4.5 Sandbox Labs for Safe AI Practice 5.1 Privacy and Data Security (GDPR, ISO/IEC 42001) 5.2 Algorithm Transparency & Accountability 5.3 Algorithmic Bias and Fairness 5.4 Legal and Societal Implications 5.5 Digital Ethics Integration Across Modules 6.1 Explaining AI to Patients 6.2 Collaborative AI Use in Health Teams 6.3 Digital Literacy and Empathy 6.4 Change Management and Trust-Building in AI Culture 7.1 Designing & Implementing AI Models 7.2 Prompt Engineering & Using AI Tools 7.3 Real-World Projects with Health Data 8.1 Designing AI Curriculum for Medical Education 8.2 Faculty Empowerment in AI Teaching 8.3 Personalized Learning via AI 9.1 Natural Language Processing (NLP) 9.2 Computer Vision & Image Processing 9.3 Expert Systems & AGI 9.4 Transfer Learning & Explainable Models 10.1 Performance Analysis & Limitations 10.2 Evaluation of Practical Utility & Usability 10.3 Research Articles & Critical Appraisal Skills 11.1 AI for Systematic Literature Reviews 11.2 AI for Quantitative & Qualitative Data Analysis 11.3 AI for Scientific Writing and Editing Student structural needs assessment for the AI in health education mobile app A total of 97 healthcare students participated in the needs assessment phase of the AI in Health Education mobile app with a 100% response rate. The results are presented in three sections: demographic characteristics, attitudes toward learning via mobile apps, and prioritization of app structural features. Demographic characteristics The majority of participants were male (64.7%), with a mean age of 22.4 years (SD = 1.8). All participants reported using mobile phones and personal laptops as their primary tools for digital learning (Table 3).Table 3Demographic characteristics of participantsVariableCategoryFrequency (n)Percentage (%)GenderMale6364.9Female3435.1Field of StudyHealth Information Technology3435.1Physiotherapy1616.5Speech Therapy1616.5Optometry1111.3Laboratory Sci.1111.3Radiotherapy99.3Digital Learning ToolsMobile Phone97100Laptop/PC97100 Table 3Demographic characteristics of participantsVariableCategoryFrequency (n)Percentage (%)GenderMale6364.9Female3435.1Field of StudyHealth Information Technology3435.1Physiotherapy1616.5Speech Therapy1616.5Optometry1111.3Laboratory Sci.1111.3Radiotherapy99.3Digital Learning ToolsMobile Phone97100Laptop/PC97100 Demographic characteristics of participants Attitudes toward learning via the mobile app Students demonstrated highly positive attitudes toward using the AI education app, with mean scores across all 10 items exceeding 4.7 on a 5-point Likert scale. The highest agreement was observed for the item “Learning with the app can make complex AI concepts easier” (M = 4.94, SD = 0.24) (Table 4).Table 4Students’ attitudes toward learning via the mobile appItemsMean (SD)Willingness to use the app for AI learning4.91 (0.29)Preference for interactive app-based learning4.91 (0.29)Belief in easing complex AI concept learning4.94 (0.24)Expectation for instructors to use the app4.91 (0.29)Belief in enhanced focus through app-based learning4.91 (0.29)Belief in long-term retention of concepts4.85 (0.36)App-based learning perceived as more engaging4.88 (0.32)Learning is better than using the app for entertainment4.85 (0.36)Belief in the app’s support for collaborative learning4.85 (0.36)Quick feedback increases motivation4.91 (0.29) Table 4Students’ attitudes toward learning via the mobile appItemsMean (SD)Willingness to use the app for AI learning4.91 (0.29)Preference for interactive app-based learning4.91 (0.29)Belief in easing complex AI concept learning4.94 (0.24)Expectation for instructors to use the app4.91 (0.29)Belief in enhanced focus through app-based learning4.91 (0.29)Belief in long-term retention of concepts4.85 (0.36)App-based learning perceived as more engaging4.88 (0.32)Learning is better than using the app for entertainment4.85 (0.36)Belief in the app’s support for collaborative learning4.85 (0.36)Quick feedback increases motivation4.91 (0.29) Students’ attitudes toward learning via the mobile app Prioritization of app structural features Students rated all 20 proposed structural features as highly necessary, with mean scores ranging from 4.76 to 4.97. The top-rated features included interactive quizzes, offline functionality, and personalized learning paths (Table 5).Table 5Prioritization of structural features of the appFeaturesMean (SD)Interactive quizzes and tests4.97 (0.17)Use of flashcards for review4.91 (0.29)Progress tracking with graphical display4.94 (0.24)Gamification elements4.85 (0.36)Step-by-step structured learning4.91 (0.29)Offline usability4.97 (0.17)Notifications and reminders4.91 (0.29)Course completion certificate4.85 (0.36)In-app AI coding practice4.88 (0.32)In-app Q&A or chatbot support4.94 (0.24)Note-taking and bookmarking4.94 (0.24)Multimedia content (videos, infographics)4.91 (0.29)User-friendly interface for all levels4.94 (0.24)Continuous content updates4.97 (0.17)Prompt engineering support4.85 (0.36)Personalized learning paths4.97 (0.17)Peer interaction within the app4.88 (0.32)Privacy and data security statements4.94 (0.24)User registration requirement4.76 (0.43) Table 5Prioritization of structural features of the appFeaturesMean (SD)Interactive quizzes and tests4.97 (0.17)Use of flashcards for review4.91 (0.29)Progress tracking with graphical display4.94 (0.24)Gamification elements4.85 (0.36)Step-by-step structured learning4.91 (0.29)Offline usability4.97 (0.17)Notifications and reminders4.91 (0.29)Course completion certificate4.85 (0.36)In-app AI coding practice4.88 (0.32)In-app Q&A or chatbot support4.94 (0.24)Note-taking and bookmarking4.94 (0.24)Multimedia content (videos, infographics)4.91 (0.29)User-friendly interface for all levels4.94 (0.24)Continuous content updates4.97 (0.17)Prompt engineering support4.85 (0.36)Personalized learning paths4.97 (0.17)Peer interaction within the app4.88 (0.32)Privacy and data security statements4.94 (0.24)User registration requirement4.76 (0.43) Prioritization of structural features of the app Inferential statistics Chi-square tests were conducted to examine the association between gender, field of study, attitudes toward learning with the app, and prioritization of app structural features. The results indicated no statistically significant differences across gender and field of study regarding students’ attitudes and their prioritization of structural features (p = 1.00). This consistency suggests universal high interest and demand for mobile-based AI education across gender and disciplinary backgrounds among students. Discussion This mixed-methods study sought to explore and define the essential content and structural features of an educational mobile app for AI literacy in healthcare. By synthesizing findings from a systematic review, evaluating existing educational mobile apps using the MARS tool, and integrating insights from both experts and end-users, we developed and validated a comprehensive framework intended to guide the future development of a learner-centered AI education mobile app. Our findings indicate significant variability in the quality and focus of existing AI-related educational mobile apps, with an average MARS score of 2.92 out of 5. Functionality was rated highest, while engagement and aesthetics lagged behind, and privacy features were poorly implemented or entirely absent. These findings align with recent studies that similarly point to the fragmented development and lack of standardization in mHealth education mobile apps, particularly those targeting complex domains like AI [7, 11, 54]. These shortcomings emphasize the need for structured, evidence-based design in mobile app development to ensure educational effectiveness, especially in fields requiring conceptual and practical understanding such as AI. The content framework derived from the systematic review encompassed ten core domains essential for AI education in healthcare: [1] Foundations and Basic Concepts of AI & Machine Learning [2], Data Science & Data Literacy [3], Practical Applications of AI in Health [4], Ethics, Law, and Governance in AI [5], Communication and Humanistic Skills in AI [6], Practical and Applied Skills [7], Enhancing the Quality of Medical Education with AI [8], Advanced Topics [9], Evaluation and Critical Analysis of AI Tools, and [10] Enhancing Research Quality with AI. These domains echo the competencies proposed in international AI literacy frameworks [17–19]. Notably, the emphasis on both theoretical and humanistic components — such as algorithmic transparency and patient-centered communication — reflects growing recognition of the dual technical-ethical nature of AI integration in healthcare. While the ethical considerations around AI in healthcare are crucial, it is equally important to address issues of fairness, bias, and equity. AI systems, by their nature, can perpetuate or even exacerbate biases if not properly designed and trained, particularly when applied to sensitive healthcare environments [55]. Our framework acknowledges these concerns by emphasizing the need for fairness and equity in AI education, ensuring that future healthcare professionals are equipped to understand and address potential biases in AI tools. By integrating these ethical principles into the design of our educational app, we aim to foster a critical awareness of AI’s potential biases, ensuring that students not only learn how to utilize AI but also understand how to mitigate its risks in clinical practice. The qualitative findings, drawn from expert commentary via open-ended questions, led to the addition of a new thematic domain: “Practical Tools and Platforms.” This domain underscores the need to expose learners to real-world applications (e.g., FHIR, EHR systems, and basic platform literacy), bridging the gap between abstract AI concepts and clinical reality. This result resonates with recent commentaries advocating for hands-on exposure to clinical informatics tools as part of AI literacy [21, 22]. In parallel, the structural needs assessment conducted among healthcare students (n = 97) revealed a preference for accessible, personalized, and interactive learning experiences. Features such as offline functionality, multimedia content, and modular structure were highly rated — supporting the idea that digital learning platforms must cater to flexibility and learner autonomy [6, 9, 56, 57]. These findings underscore the importance of incorporating broader pedagogical frameworks, such as constructivism and self-regulated learning, in the design of AI education apps. According to studies by Patrick et al. [58] and Parsons et al. [59], learner-centered approaches that promote active engagement, autonomy, and personalized learning paths are critical for fostering deeper understanding and sustained motivation. These theories advocate for creating learning environments that allow students to take control of their learning process, set personal goals, and adjust their learning strategies, which align with the features highly rated by our participants, such as offline functionality, modular structures, and multimedia content. By embedding such theoretical principles, the app could enhance learners’ ability to manage their learning, making it more adaptive and effective in a rapidly evolving technological landscape. Similarly, addressing concerns related to trust and competence in AI use, as highlighted by Firuzpour et al. [60] in their study on physician confidence in Iran, can enhance the app’s ability to effectively engage healthcare professionals. By integrating the identified gaps in physician trust and AI competence, our app framework aims to build confidence through practical, targeted educational experiences. No statistically significant differences were observed between students based on gender or academic background, suggesting that the proposed structure holds cross-demographic applicability. However, the uniformly high ratings observed (mean >4.7 across most features) may partially reflect a ceiling effect or social desirability bias. While responses likely represent genuine interest in accessible AI education, it is important to note that participants were informed they would receive free access to the finalized app, which may have influenced the positivity of responses. Future studies should explore alternative data triangulation methods, such as behavioral usage metrics or follow-up engagement data, to mitigate this potential bias and validate user preferences more robustly. An important implication of our findings is the integrated design of AI education that addresses both competence and confidence. While many learners express interest in AI, previous studies have shown a lack of confidence due to limited exposure and unclear curricular structures (20, 50). Our validated framework helps mitigate this issue by aligning content with learners’ expectations and pedagogical standards, and is positioned to support both formal curricula and self-directed learning pathways. While our framework addresses ethical and legal dimensions through global standards such as GDPR and ISO/IEC 42,001, its practical application in low-resource settings presents unique challenges. Kaushik et al. [61] identified major implementation barriers in LMICs, including limited health data literacy, inconsistent data systems, and regulatory ambiguity—factors that complicate the translation of international frameworks into actionable local practice. Similarly, Yu and Yu [55] introduced the concept of the “AI Deployment Paradox,” which suggests that poorly contextualized AI initiatives may exacerbate rather than reduce health disparities. Their thematic analysis emphasizes three key issues: data poverty and contextual misalignment; trade-offs between cost-effectiveness and health equity; and the risk of technological colonization. These insights underscore the importance of embedding locally sensitive governance models and participatory ethics into AI education, ensuring that learners are not only aware of international standards, but also capable of critically examining their relevance and adaptability in diverse healthcare environments. Also, this highlights the need for ongoing, longitudinal studies to better understand how AI education tools perform in real-world settings. Long-term evaluations will be essential to assess the sustained impact on learning outcomes, motivation, and academic achievement. While our current framework lays the foundation for an effective AI education tool, future research will need to focus on evaluating the app’s effectiveness over time, particularly in diverse healthcare settings. This study presents several notable strengths, including the use of multiple data sources (systematic review, mobile app analysis, expert and user input), the mobile app of the validated MARS tool for evaluating existing AI education apps, and a strong user-centered design approach. Although developing a mobile app prototype was beyond the scope of the current research, the validated content and structural framework offers a solid foundation for future development and implementation. In addition, it is important to highlight that, given the fast-evolving nature of AI technologies and the educational needs of healthcare students, adopting an agile approach to the development of educational mobile apps is crucial. This approach allows for rapid iteration and adaptation to meet the immediate needs of learners, ensuring that students are not deprived of essential educational tools due to delays in app design and implementation. This responsiveness to user feedback and evolving technological trends is key to ensuring that the app remains relevant and effective as AI technologies continue to advance. Nonetheless, several limitations should be acknowledged. While this study utilized open-ended questions and thematic analysis for qualitative insights, we acknowledge that more in-depth qualitative methods, such as student interviews or focus groups, are essential to gain a richer understanding of learners’ experiences. Future phases of this research will incorporate these qualitative approaches to provide a more comprehensive perspective on user engagement and the effectiveness of the app in addressing learners’ needs. Both the student and expert participants were recruited from a single institution, which may introduce institutional bias and limit the generalizability of the findings. Although we included individuals from a variety of academic levels and disciplines to improve representativeness, further multi-institutional validation is needed to strengthen external validity. Additionally, the app review included both Persian and English-language applications to capture a broader range of available resources. While this bilingual inclusion expanded the scope of analysis, it may have introduced regional or cultural biases, potentially limiting the broader applicability of the results. The study’s focus on healthcare students and professionals may also restrict generalizability to learners in non-health domains, such as engineering, business, or the social sciences. Finally, due to methodological heterogeneity among the included studies in the systematic review, a standardized quality appraisal tool (e.g., AMSTAR 2 or MMAT) was not applied. Instead, we used a predefined set of four criteria to ensure consistent and rigorous assessment across diverse study designs. Conclusion This mixed-methods study developed and validated a multidimensional framework for the design of an AI literacy mobile app tailored to healthcare education. Integrating evidence from a systematic literature review, app quality analysis using MARS, and stakeholder input, the resulting framework addresses both content and structural dimensions across eleven core domains, from foundational AI concepts to applied and ethical considerations. The findings reveal major gaps in current AI educational apps—particularly in interactivity, content coherence, and privacy—underscoring the need for structured, user-centered design. The inclusion of a “Practical Tools and Platforms” domain, based on expert recommendations, enhances the framework’s applicability to real-world training. Student feedback further supports the demand for interactive, flexible mobile learning environments. This framework offers a strong basis for developing AI education tools in healthcare and can inform future curricular design and interdisciplinary training. Further research should test its scalability and relevance beyond the health sector. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2. Supplementary Material 3. Supplementary Material 3. Supplementary Material 4. Supplementary Material 4.