Artificial intelligence (AI) for social innovation in health education: promoting health literacy through personalized ai-driven learning tools – a systematic review Authors: Dina Mansour Tbaishat, Maha Waleed Elfadel Published: 2025 ABSTRACT AbstractBackgroundArtificial Intelligence (AI) is transforming health education by enabling personalized, adaptive, and scalable approaches that may enhance aspects of health literacy. Despite rapid adoption, comprehensive synthesis of AI tools’ impact on health literacy as social innovation is limited. Understanding these effects guides educators, developers, and policymakers in designing potentially effective, inclusive, and ethical AI interventions. This review examines generative AI models, chatbots, and adaptive learning systems in supporting health literacy globally.MethodsA systematic review was conducted following PRISMA guidelines. Literature was identified primarily through PubMed/Medline, Scopus, and ScienceDirect. Connectedpapers.com was used exclusively as a citation chasing tool, performing both backward and forward reference searches to identify thematically linked studies not captured by database searches. All records retrieved via Connected Papers were subjected to the same eligibility criteria as database-sourced studies, covering publications from 2000–2025. A total of 75 peer-reviewed empirical and theoretical studies focusing on AI tools for health literacy and social innovation were included. Titles, abstracts, keywords, and full texts were screened using predefined criteria. Data were managed and de-duplicated using Zotero. Screening and eligibility decisions were recorded in Excel spreadsheets. Thematic synthesis was conducted manually. PRISMA 2020 and PRISMA-S checklists were used to ensure transparent reporting.ResultsAI research in health education was minimal until 2020 but rose sharply from 2021, peaking in 2023–2024 with generative AI (e.g., ChatGPT). Of the 75 included studies, 68 (90.7%) were co-authored by two or more researchers, 54 (72.0%) were published as Open Access, and review articles dominated with 41 studies (54.7%), while empirical research was limited, highlighting moderate to weak evidence. Research focused on personalized AI tools and learning effectiveness, with limited exploration of ethics, technical barriers, or social innovation. Findings suggest that AI interventions may improve readability, metacognitive engagement, cultural accessibility, and learner autonomy in the short term, particularly when multifaceted. However, evidence for long-term behavior change and real-world impact is sparse, indicating caution in generalizing results. Challenges include algorithmic bias, digital inequity, and lack of transparency, emphasizing the need for inclusive, equity-driven AI strategies.ConclusionAI-powered tools have potential to support health literacy and learner-centered innovation, while contributing to social impact. Multifaceted, adaptive interventions may offer greater benefits than single-tool approaches. Findings provide preliminary guidance for standardized training, AI literacy integration, and policy frameworks, while acknowledging the current limitations in evidence, generalizability, and long-term outcomes.Keywords: Artificial intelligence (AI), Health education, Health literacy promotion, Personalized learning tools, Digital health education, Systematic literature review Background In recent years, artificial intelligence (AI) has emerged as a powerful catalyst for innovation, particularly within healthcare and education [1]. As a defining pillar of the Fourth Industrial Revolution, AI enhances operational efficiency and enables personalized, data-driven interventions [2, 3]. While considerable attention has been devoted to AI’s role in clinical diagnostics and administrative automation [4, 5], its pedagogical applications in health education remain comparatively underexplored [6]. Increasingly, AI is being recognized not merely as a technological tool but as a driver of social innovation, capable of expanding access to reliable health information, supporting individualized learning pathways, and advancing health literacy across diverse populations [7, 8]. Recent advances in personalized AI technologies, such as intelligent tutoring systems [9], conversational health agents [10, 11], wearable health monitors [8], and AI-driven diagnostic tools, including PathAI and Zebra Medical Vision [12], have generated new opportunities for promoting health literacy through interactive and multilingual content delivery. These innovations have proven particularly effective in enhancing knowledge retention, facilitating early intervention, and empowering individuals to engage in preventive self-care [13]. Social Innovation (SI) is conceptualised as new or improved strategies, practices, and models that address unmet social needs and generate positive, equitable, and sustainable societal change. In health education, SI emphasises inclusivity, accessibility, and the empowerment of individuals and communities through tools that reduce health disparities and enhance health literacy. Positioning AI as a form of SI allows examine not only the functional capabilities of AI-based learning tools but also their broader societal impact. This includes their potential to democratize access to health information, support culturally responsive learning, and address systemic inequities in digital health education [10]. Global partnerships between major technology firms and healthcare organizations, such as Microsoft Azure’s collaborations with Isomorphic Labs, exemplify the growing commitment to deploying AI-powered education platforms on a broader scale [14, 15]. Collectively, these initiatives reflect a paradigm shift towards accessible, context-sensitive, and user-centered health education models, further reinforcing AI’s transformative role in fostering socially impactful learning environments [16]. Building on the foundational role of AI in health education, emerging technologies now offer unprecedented opportunities to enhance health literacy through personalized, multilingual, and culturally responsive learning experiences [9, 10]. AI-powered tools, including intelligent tutoring systems, health-focused chatbots, wearable health devices, and diagnostic platforms, have demonstrated strong potential in enhancing knowledge retention, facilitating early diagnosis, and promoting self-care, particularly among digitally underserved populations [8, 13]. These tools leverage real-time data to deliver interactive and adaptive content tailored to diverse user needs. AI plays a growing role in addressing systemic health inequities by adapting educational content to different languages, literacy levels, and cultural backgrounds [17]. Through features such as voice-assisted interfaces and context-aware feedback mechanisms, these technologies facilitate comprehension and engagement among users with varying levels of digital proficiency [8]. Beyond information delivery, AI actively shapes health behavior through data-driven messaging and intelligent recommendation systems that support informed decision-making and preventative care [16]. Interactive chatbots and virtual assistants also provide ongoing motivation through timely, personalized guidance [18]. Building on these advancements, recent research highlights the emergence of AI-intelligentized assistants, systems that extend beyond content delivery to provide empathetic, contextual, and adaptive interactions. These assistants include AI mental health companions and virtual health coaches guiding users in lifestyle, nutrition, and preventive care [19]; conversational AI chatbots for interactive guidance [20]; emotionally intelligent systems that respond to user emotions and sustain engagement [21]; adaptive learning platforms tailored to individual knowledge levels; and AI-powered wearable devices providing real-time feedback [22]. Their use in health education and behavioral promotion is further supported by systematic reviews of AI-driven virtual assistants in physical activity and preventive health contexts. Collectively, they enhance health literacy, support engagement, and promote equitable access to health education, particularly for vulnerable or digitally underserved populations. However, realizing the full potential of these innovations requires sustained attention to ethical safeguards, algorithmic transparency, and user-centered design to foster trust, inclusivity, and responsible adoption [23, 24]. Despite notable progress in AI-driven health education, several structural and ethical barriers persist, constraining its effective and equitable deployment. Chief among these is the persistent digital divide, which limits access to AI-based educational tools among marginalised populations [7, 25]. Disparities in digital literacy and baseline health knowledge further shape users’ engagement with, and trust in, AI-generated content [26, 27]. Ethical challenges, including data privacy, algorithmic bias, and misinformation, complicate the integration of AI within educational contexts [28, 29]. Exacerbating these challenges is the lack of standardised evaluation frameworks and the limited empirical evidence on the educational outcomes of AI interventions [12, 23]. While AI technologies have shown considerable promise in areas such as diagnostics and predictive modelling [30], their educational utility, particularly in influencing health behaviours and promoting literacy, remains under-examined. Few studies have systematically assessed the effectiveness, ethical viability, and contextual relevance of AI-powered educational interventions. Given the growing reliance on digital platforms for the dissemination of health information, there is a pressing need for rigorous research that investigates the role of AI in public health education. In response, this study undertakes a systematic examination of AI applications in health education, with particular emphasis on their capacity to enhance health literacy. Research aims Following the PRISMA (Preferred Reporting Items for Systematic Reviews) guidelines [31], this study seeks to generate actionable insights and policy-relevant recommendations that facilitate the effective integration of artificial intelligence (AI) technologies into health education tools. The ultimate goal is to offer scalable, personalized, and context-sensitive solutions for promoting health literacy across diverse populations. This review adopts the perspective of Social Innovation (SI), understood as a transformative approach that extends beyond technological advancement to create positive societal outcomes. The study not only evaluates the educational effectiveness of AI-driven health tools but also highlights their potential societal benefits, including improving access, addressing disparities, and empowering diverse learner populations. The review is structured around the following four core objectives: RO1. To identify the most widely adopted AI-driven personalized learning tools in health education, with an emphasis on their functionalities, implementation contexts, and pedagogical characteristics. RO1. To identify the most widely adopted AI-driven personalized learning tools in health education, with an emphasis on their functionalities, implementation contexts, and pedagogical characteristics. RO2. To evaluate the effectiveness of these tools in enhancing health literacy by synthesizing findings from experimental, applied, and empirical studies. RO2. To evaluate the effectiveness of these tools in enhancing health literacy by synthesizing findings from experimental, applied, and empirical studies. RO3. To investigate the ethical and technical challenges associated with AI applications in health education, and to propose evidence-based recommendations to address these concerns. RO3. To investigate the ethical and technical challenges associated with AI applications in health education, and to propose evidence-based recommendations to address these concerns. RO4. To explore current research gaps in the deployment and evaluation of AI-powered interventions for health education. RO4. To explore current research gaps in the deployment and evaluation of AI-powered interventions for health education. The structure of this paper is as follows: next section outlines the research methodology; then descriptive results and a thematic synthesis of the selected studies are presented; the following section offers a critical discussion of key findings, identifies research gaps, and outlines directions for future inquiry; and finally the last section concludes with the main takeaways and implications for AI-enhanced health education. Method To ensure methodological rigor and academic transparency, a Systematic Literature Review (SLR) was conducted, guided by the structured methodology proposed by Behera et al. [32]. This approach was selected for its comprehensive, replicable, and transparent framework for identifying, analyzing, and synthesizing relevant literature on AI applications in health education. Concurrently, the PRISMA 2020 statement was adhered to document the review process and enhance clarity and transparency in reporting. The combined application of these frameworks ensured a structured and rigorous methodology for identifying, screening, and synthesizing research focused on AI-driven tools in health education and literacy. The SLR was implemented in four principal stages: (1) planning and protocol design; (2) identification and screening; (3) eligibility assessment using predefined inclusion and exclusion criteria; and (4) data extraction, analysis, and synthesis. Step 1: planning and conducting the review protocol design In the initial phase of the study, a comprehensive and structured search strategy was developed following the Systematic Literature Review (SLR) framework proposed by Behera et al. [32]. The review protocol guided the identification, selection, and organization of studies on the use of artificial intelligence (AI) in health education and literacy, with a particular emphasis on social innovation. The search was conducted across three leading academic databases, Scopus, ScienceDirect, and PubMed, which are widely recognized for their coverage of peer-reviewed literature in the fields of health, education, and computer science [33]. The search strategy employed a well-defined set of keywords aligned with the study’s core objectives. These included terms such as “Artificial Intelligence”, “Personalized AI-Driven Learning Tools”, “Health Education”, “AI in Education”, “Health Promotion”, “Health Literacy”, “Digital Health”, “Personalized Health Education”, “Social Innovation”, “AI for Social Innovation”, “Ethical Issues”, “Personalized Learning Tools”, and “Chatbots”. Boolean operators (AND, OR) were applied to combine keywords effectively, and the search was restricted to the title, abstract, and keyword fields to maximize relevance and precision. Subject filters were applied within each database as follows: Scopus: Computer Science, Medical Sciences, Health Professions, and Social Sciences. Scopus: Computer Science, Medical Sciences, Health Professions, and Social Sciences. ScienceDirect: Health Sciences and Education. ScienceDirect: Health Sciences and Education. PubMed: Medical Education and Digital Health Technologies. PubMed: Medical Education and Digital Health Technologies. The search was limited to studies published between January 2000 and June 2025, ensuring the inclusion of both foundational and contemporary research. This timeframe allowed for a holistic understanding of the evolving integration of AI in health education and its alignment with digital transformation trends. Although AI technologies in health and informal education date back several decades, practical adoption of AI tools within formal health education settings accelerated in the early 2000s, driven by advancements in intelligent tutoring systems, adaptive learning, and natural language processing capabilities. Documented uptake in medical curricula began emerging after 2000, as highlighted by Li and Wu [34], rendering this period an appropriate baseline for the review. This rigorous search protocol ensured that the included studies were aligned with the review’s conceptual scope and methodological rigour, enabling a focused and replicable analysis. To enhance transparency, documentation and reproducibility, complete search strings for each database were documented according to PRISMA-S guidelines. The final searches were conducted between May 10 and June 15, 2025. A detailed record of all search strings, field restrictions, and limits is included in Appendix A (PRISMA-S Table). Step 2: screening and selection process Following the initial search, data were managed and de-duplicated using Zotero, a widely recognized open-source reference management tool that facilitates transparent data handling, supports de-duplication of bibliographic entries, and enables structured organization, tagging, and collaborative screening. Using Zotero’s duplicate detection feature, 153 duplicate records were automatically identified and removed. Screening and eligibility decisions were recorded in Excel spreadsheets, and thematic synthesis was conducted manually. Additionally, 98 records were automatically excluded based on pre-applied database filters, including non-English publications, conference abstracts, editorials, book chapters, and duplicate records detected through automated de-duplication tools, resulting in 820 records subjected to title and abstract screening. The screening process was conducted in two structured phases: Initial screening Titles and abstracts were assessed to exclude irrelevant studies, particularly those unrelated to AI in health education, health literacy, or personalized AI-driven learning tools. Full-text review Remaining articles were evaluated in full against the inclusion and exclusion criteria outlined in Step 3. Qualitative analytical framework To ensure transparency, validity, credibility, replicability, and interpretive rigor, the included studies underwent a structured thematic analysis. Two reviewers independently coded qualitative data extracted from the studies, identifying recurring patterns, themes, and subthemes relevant to AI-driven health literacy interventions. Discrepancies in coding were resolved through discussion until consensus was reached, with a third senior reviewer consulted when needed to arbitrate. This process ensured systematic, transparent, and credible analysis, guiding the descriptive synthesis while enhancing the validity of the qualitative appraisal. The screening and data extraction were conducted systematically: titles and abstracts were assessed against predefined inclusion/exclusion criteria, full texts were evaluated for eligibility, and data were extracted using a structured matrix. All reviewers underwent training in systematic review methodology, including calibration on a pilot sample to ensure consistency. Discrepancies at any stage were resolved through discussion, and a third senior reviewer acted as an arbiter when consensus could not be reached. This dual-review process ensured methodological rigor and transparency, with minor subjective judgments representing a potential limitation. Following these procedures, the included studies underwent qualitative appraisal for methodological quality. To ensure transparency and interpretive rigor, the methodological quality of the included studies was appraised qualitatively rather than through standardized scoring systems. Given the heterogeneity of designs (descriptive, survey-based, and quasi-experimental), formal risk-of-bias instruments (e.g., JBI, RoB 2, AMSTAR 2) were not applied. Instead, two reviewers independently examined each study for methodological clarity, transparency of data collection and analysis, and alignment with the review objectives. Discrepancies were discussed until consensus was achieved. This qualitative appraisal guided the descriptive analysis and thematic synthesis phases by assigning greater interpretive weight to studies demonstrating robust design and transparent reporting. To ensure comprehensive coverage and mitigate potential keyword or disciplinary bias, the Connected Papers platform was used on 15 June 2025 as a citation chasing tool, performing both backward and forward reference searches to identify thematically linked studies not captured by database searches. Five seed papers were selected based on high citation impact and relevance to AI-driven health education and literacy [1, 8, 39, 95, 105]. For each seed paper, both forward and backward citation tracing were conducted using the co-citation graph, yielding eight additional records. All records were manually screened against the predefined inclusion/exclusion criteria. After full-text evaluation, four studies met al.l criteria and were included in the final synthesis [35–38], while four were excluded due to insufficient methodological detail or lack of educational focus. All titles, screening notes, and inclusion/exclusion decisions are documented in the screening log (Appendix B, CSV format) to ensure transparency and replicability (see Figs. 1, 2, 3 and 4). Connectedpapers.com network visualization – seed paper: Love et al. (2025) [35] Connectedpapers.com network visualization – seed paper: Chun et al. (2025) [36] Connectedpapers.com network visualization – seed paper: Dzangare and Gulu (2025) [37] Connectedpapers.com network visualization – seed paper: Fitzpatrick, 2023 [38] In each Connected Papers network diagram, the studies positioned on the right-hand side represent those included in the final synthesis, while the studies displayed on the left-hand side are related works that did not meet the inclusion criteria but were identified as thematically relevant during citation chasing. This visual distinction clarifies the relationship between included and excluded studies and improves the interpretability of the search and selection process. Step 3: inclusion and exclusion criteria To ensure methodological consistency and relevance, a clear set of inclusion and exclusion criteria was applied during the full-text screening phase. Inclusion criteria Studies were eligible for inclusion if they were published in peer-reviewed academic journals ranked Q1 or Q2 according to Scimago Journal Rankings (scimagojr.com). This restriction was applied to ensure the inclusion of studies with high methodological rigor, validated research designs, and robust peer-review processes. Focusing on Q1 and Q2 journals prioritizes reliable and high-impact evidence, enhancing the credibility and consistency of the review findings. Only articles written in English and published between January 2000 and June 2025 were considered, capturing both foundational and contemporary developments in AI. Eligible studies focused on the application of artificial intelligence (AI) in health education or the promotion of health literacy through personalized AI-driven learning tools. Research addressing ethical, technological, or accessibility challenges within AI-enabled health education was also included. Accepted study types comprised empirical studies, conceptual frameworks, mixed-methods research, and systematic or scoping reviews that contributed meaningfully to understanding how AI technologies, such as chatbots, machine learning algorithms, large language models (LLMs), or adaptive learning systems, are designed and implemented to enhance personalized digital health education. Exclusion criteria Studies not written in English, not peer-reviewed, or failing to meet standards of academic quality and transparency were excluded. In line with Tranfield et al. [39], conference proceedings, book chapters, editorials, theses, dissertations, and all forms of grey literature were also excluded. The exclusion of grey literature was deliberate to ensure methodological rigor, reliability, and transparency, as such sources often lack formal peer review and standardized reporting, which may introduce bias, limit replicability, and reduce overall evidence quality. Additionally, studies focusing exclusively on clinical applications of AI (e.g., diagnostics, decision support systems, or treatment planning) without an explicit educational component were excluded. Articles lacking sufficient methodological detail, those unavailable in full text, or those published outside the defined timeframe (2000–2025) or in journals not ranked Q1 or Q2 were also excluded. Ultimately, only studies that fully aligned with the review’s central focus on AI-enabled health education and literacy were retained for final analysis. A total of 1,071 records were initially retrieved from Scopus, ScienceDirect, and PubMed. Following the removal of 153 duplicates and the exclusion of 98 entries based on language and document type, 820 records remained for title and abstract screening. After a rigorous full-text eligibility assessment in accordance with the defined criteria, 75 studies were ultimately included in the final synthesis, 71 identified through database searches and four through Connectedpapers.com. The detailed search criteria are summarized in Table 1. Table 1Search criteria used in the systematic literature reviewCategoryDetails Databases Searched Scopus, ScienceDirect, PubMed Date of Last Search Between May 10 and June 15, 2025 Search Fields Article title, Abstract, Keywords Keywords (“Artificial Intelligence” OR AI OR “Personalized AI-Driven Learning Tools”) AND (“Health Education” OR “Health Literacy” OR " Digital Health Education " OR “Personalized Health Education”) AND (“Social Innovation” OR “AI For Social Innovation”) AND (“Ethical Issues”) AND (“AI in Education “) AND (“Chatbots”) AND (“Health Promotion”) AND (“Personalized Learning Tools”) Boolean Operators “AND”, “OR” Document Type Case Study, Review article, Quantitative study, Qualitative Study, Systematic Literature Review, Opinion Article. Subject Areas Computer Science (Artificial Intelligence & Computer Science Applications), Social Sciences (Education, Health (Social Science) & Development), Medicine (Public Health, Environmental and Occupational Health), Health Professions. Time Frame January 2000 - June 2025. Inclusion Criteria - Peer-reviewed journals (Q1 + Q2).- English language.- Studies targeting AI applications in health education or promotion of health literacy through personalized AI-driven tools.-Addressed ethical, technical, or accessibility challenges.- Empirical/conceptual/systematic frameworks.- Included empirical findings, conceptual frameworks, or systematic reviews. Exclusion Criteria - Non-English.- Non-peer-reviewed-Conference papers, Conference review, book, book chapters, editorials, dissertations, or any grey literature.- Clinical/diagnostic-only focus.- Insufficient methodological detail.- Unavailable full text.- Studies unrelated to health education or AI-enabled learning tools.- General education without health context. Search criteria used in the systematic literature review - Peer-reviewed journals (Q1 + Q2). - English language. - Studies targeting AI applications in health education or promotion of health literacy through personalized AI-driven tools. -Addressed ethical, technical, or accessibility challenges. - Empirical/conceptual/systematic frameworks. - Included empirical findings, conceptual frameworks, or systematic reviews. - Non-English. - Non-peer-reviewed -Conference papers, Conference review, book, book chapters, editorials, dissertations, or any grey literature. - Clinical/diagnostic-only focus. - Insufficient methodological detail. - Unavailable full text. - Studies unrelated to health education or AI-enabled learning tools. - General education without health context. A detailed PRISMA 2020 flow diagram (Fig. 5) was employed to illustrate the entire screening and selection process. The diagram reports the number of studies identified, screened, and excluded at each stage, along with the reasons for exclusion (e.g., insufficient methodological detail, absence of educational focus, or duplication). This ensures full transparency regarding how the final 75 studies were selected for inclusion in the review. A full list of included studies and their key characteristics is provided in Supplementary Appendix B. PRISMA 2020 Flow Diagram of Study Selection.(Source: Authors, 2025) Step 4: data extraction, analysis, and synthesis Following the final selection of eligible studies, essential information was systematically extracted and organized into a structured data matrix. Extracted data encompassed authorship, publication year, country of origin, study objectives, research methodology (qualitative, quantitative, mixed-methods, or review), target population, educational setting, type of AI application, and reported outcomes related to health education and health literacy. Special attention was given to ethical, technical, and social issues addressed in the studies. This dataset formed the foundation for a thematic analysis, through which the research team identified recurrent patterns, emerging insights, and notable gaps in the current literature. The thematic synthesis was structured around four core dimensions: (1) types and features of personalized AI-driven learning tools; (2) their effectiveness in improving health literacy; (3) ethical, technical, and accessibility challenges; and (4) research gaps and methodological limitations. The entire synthesis process was conducted in accordance with PRISMA 2020 guidelines, ensuring transparency, methodological rigor, and replicability. The results of this synthesis are presented in the subsequent Results section. Results Descriptive analysis The descriptive analysis of publishing trends on Artificial Intelligence (AI) in health education and healthcare systems between 2000 and 2025 reveals a clear temporal evolution in research focus. Between 2000 and 2019, fewer than five relevant papers were published per year, indicating that AI had not yet been conceptualized as an instructional or pedagogical tool in health education. A substantial increase emerged from 2021 onward, with publication activity peaking in 2024 (see Fig. 6). This surge aligns with the global adoption of generative AI technologies, particularly AI chatbots and intelligent tutoring systems, across academic and clinical education contexts. Growth trend of publications on artificial intelligence in health education (2000–2025). Source: Author’s analysis based on data retrieved from the Scopus database Temporal distribution of publications The analysis of 75 eligible studies (Fig. 7) shows that no studies met the review criteria before 2020, with the first qualified papers appearing that year. A sharp increase was observed during 2023–2025, accounting for 85% (n = 64) of the total studies. The yearly distribution indicates a clear growth in research activity over time. Only 2.67% of the studies were published in 2020 and 4% in 2021, reflecting limited early attention to the topic. The proportion increased to 8% in 2022 and 16% in 2023, showing gradual academic engagement. The peak occurred in 2024, with 40% of all studies published that year, demonstrating strong research momentum. Although the share slightly declined to 29.33% in 2025, it still reflects a high level of productivity compared to earlier years. This trend corresponds to the emergence of tools such as ChatGPT, signaling the early phase of AI integration in health curricula, primarily through pilot initiatives and adaptive learning platforms. These findings suggest that AI supports learner engagement, personalized feedback, and skill acquisition in clinical reasoning and decision-making. From a psychological perspective, technological readiness and perceived usefulness appear to enhance both institutional adoption and student motivation, reflecting a shift toward learner-centered pedagogy that fosters autonomy and adaptive skill development. Distribution of Included Studies by Year (n = 75). (Source: Authors, 2025) Authorship patterns and collaborative dynamics The analysis of authorship patterns among the 75 included studies reflects a clear tendency toward collaborative research (see Table 2). Specifically, 90.7% (n = 68 of 75) of publications were co-authored by two or more researchers, whereas only 9.3% (n = 7 of 75) were single-author studies. This predominance of multi-author works suggests that AI research in health education is inherently interdisciplinary, often requiring expertise across health sciences, educational theory, artificial intelligence, computer science, and data analysis. Table 2Authorship pattern (n = 75)Authorship TypeNo. of PublicationsPercentageMultiple Authors6890.7%Single Author79.3%Total75100% Authorship pattern (n = 75) Such collaboration may positively influence the design and implementation of AI-driven educational interventions, as studies involving diverse expertise tend to produce more comprehensive and pedagogically sound health literacy tools. Multi-author studies frequently report improved learner engagement, enhanced comprehension of complex health topics, and more effective integration of AI tools into curricula, compared to single-author studies. Additionally, team-based scholarship enables robust validation of AI models, addressing both technical reliability and ethical considerations. The limited number of individual-authored studies highlights potential challenges in visibility and independent innovation, particularly in multi-institutional contexts. General, these patterns underscore the importance of interdisciplinary collaboration for advancing evidence-based AI interventions in health education and their psychological outcomes. Journal quality and source distribution Analysis of the publication sources indicates that the majority of the 75 included studies were published in high-impact journals, with 72% (n = 54 of 75) appearing in Q1 journals and 28% (n = 21 of 75) in Q2 journals, based on Scimago Journal Rankings (see Table 3). The research primarily addressed medical education, digital health, and interdisciplinary care, with BMC Medical Education (n = 12) and JMIR Medical Education (n = 6) being the most frequent outlets. Table 3Distribution of publications by journal and quartile ranking (n = 75)JournalN. PublicationsQuartileAcademic Medicine1Q1Annals of Medicine1Q1Annals of Medicine and Surgery1Q1BMC Medical Education12Q1BMC Medical Informatics and Decision Making1Q1BMC Public Health1Q1Clinical Anatomy1Q1Computers and Education: Artificial Intelligence2Q1European Urology Open Science1Q1Frontiers in Digital Health2Q1Frontiers in Medicine1Q1Frontiers in Public Health2Q1Health Promotion Practice1Q1Humanities and Social Sciences Communications1Q1IEEE Access2Q1International Journal of Integrated Care1Q1JMIR Aging1Q1JMIR Human Factors1Q1JMIR Medical Education6Q1Journal of General Internal Medicine1Q1Journal of Medical Internet Research4Q1Journal of Racial and Ethnic Health Disparities1Q1Library Hi Tech News1Q1Npj Digital Medicine1Q1Nurse Education in Practice1Q1Nursing Outlook1Q1PLOS ONE2Q1Scandinavian Journal of Caring Sciences1Q1Technology in Society1Q1Urologic Oncology: Seminars and Original Investigations1Q1Total 54 (72%) Advances in Medical Education and Practice1Q2AI (Switzerland)1Q2Applied Sciences (Switzerland)1Q2British Dental Journal1Q2Ciencia e Saude Coletiva1Q2Digital Health1Q2Frontiers in Oncology1Q2Health Science Reports1Q2Healthcare (Switzerland)3Q2Information Development1Q2Internal Medicine Journal1Q2JMIR Formative Research3Q2Journal of Education and Health Promotion2Q2Multimodal Technologies and Interaction1Q2Narra J1Q2Pediatric Dermatology1Q2Total 21 (28%) Distribution of publications by journal and quartile ranking (n = 75) This distribution demonstrates that AI in health education is largely investigated in journals with rigorous peer-review standards and high methodological quality. High-impact publications provide robust evidence that AI-driven interventions can enhance health literacy, improve learner engagement, and support adaptive learning processes. For example, studies in BMC Medical Education (n = 12) reported measurable gains in students’ comprehension of complex health topics using AI-powered chatbots or personalized learning platforms. Similarly, JMIR Medical Education studies (n = 6) documented improvements in metacognitive skills, self-directed learning, and autonomous problem-solving among health profession students exposed to AI-based tools. These findings underscore the practical implications for curriculum designers and educators: integrating AI tools into health education can strengthen both cognitive and behavioral outcomes while supporting psychological constructs such as learner motivation, engagement, and self-efficacy. Access type distribution Among the 75 reviewed studies, 54 studies of 75 (72.0%) were published as Open Access, 20 studies of 75 (26.7%) were accessible only through institutional subscriptions, and 1 study of 75 (1.3%) was available via short-term access (see Table 4). No studies were excluded due to incomplete accessibility. Table 4Distribution of included studies by access type (n = 75)Access TypeNo. of PublicationsPercentageOpen Access (available and downloaded)5472.0%Closed Access (institutional access only)2026.7%Short-term open access11.3%Not available (excluded based on criteria)00.00%Total75100% Distribution of included studies by access type (n = 75) The high proportion of Open Access publications enhances the dissemination of AI-driven educational tools, allowing educators, students, and healthcare practitioners to access research findings without subscription barriers. This broader accessibility may foster greater engagement with AI-based learning technologies, support evidence-informed curriculum design, and encourage the adoption of innovative teaching strategies. For example, Open Access studies were more frequently cited in subsequent research, demonstrating their influence on educational practice and the development of health literacy interventions. These findings highlight the importance of accessibility in supporting widespread adoption and implementation of AI tools in health education settings, promoting both equitable learning opportunities and informed pedagogical decision-making. Types of included studies The distribution of the 75 reviewed studies by document type indicates that AI-driven health literacy in health education remains an emerging research field. Review-based studies dominated the literature, accounting for 29 out of 75 studies (38.6%; see Table 5), highlighting an initial focus on synthesizing existing evidence rather than generating new empirical data. This prevalence underscores the ongoing need for evidence-based guidance on integrating AI tools into health education curricula. From a psychological perspective, the limited number of qualitative and mixed-methods studies suggests that user experiences, learner motivation, and cognitive engagement remain underexplored. Table 5Distribution of studies by paper type (n = 75)Document TypeNo. of StudiesPercentageCase Study45.33%Original Research4256%Scoping Review34%Systematic Literature Review2533.33%Structured Literature Review11.33%Total 75 100% Distribution of studies by paper type (n = 75) Original research studies comprised 42 of the 75 studies (56%), typically assessing outcomes such as readability, comprehension, and short-term knowledge retention following AI-based interventions. Among the remaining studies, systematic literature reviews accounted for 25 studies (33.33%), structured literature reviews and scoping reviews each comprised 4 studies (5.33%), and case studies represented 4 studies (5.33%). These numbers reveal a significant gap in understanding context-specific challenges, learner perceptions, and real-world implementation. General, the findings suggest that while AI tools demonstrate potential for enhancing learning outcomes, the predominance of review-based studies and the limited empirical evidence highlight the need for future research employing rigorous quantitative, qualitative, and mixed-methods approaches. Such research should aim to comprehensively evaluate both the educational and psychological impacts of AI interventions, particularly in authentic classroom and clinical settings. Thematic landscape: conceptual priorities and gaps The thematic synthesis of the 75 included studies highlights key conceptual priorities and gaps in AI-driven health education research (see Table 6). Personalized AI-driven learning tools were reported in 43 of 75 studies (57%) to enhance students’ comprehension of health topics, demonstrating measurable improvements in metacognitive engagement, knowledge retention, and learner autonomy. Adaptive learning systems further increased motivation and self-directed learning, consistent with cognitive and behavioral theories in health education. Table 6Thematic frequency matrix (n = 75)Author (Year)Theme/TopicPersonalized AI ToolsHealth EducationHealth LiteracyEffectiveness\ OutcomesEthical IssuesTechnical ChallengesDigital Health IntegrationMachine Learning ApplicationsSocial InnovationAbou Hashish & Alnajjar (2024) [40]✔✔✔✔✔Adegboye (2024) [41]✔✔✔✔✔✔✔Ajalo et al. (2025) [42]✔✔✔✔✔✔✔Alkhaaldi et al. (2023) [43]✔✔✔✔✔✔Almarzouki et al. (2025) [44]✔✔✔✔✔✔Aloi et al. (2025) [45]✔✔✔✔✔✔✔AlZaabi & Masters (2025) [46]✔✔✔✔✔✔Asiksoy (2025) [47]✔✔✔✔✔✔✔Au et al. (2023) [48]✔✔✔✔✔✔✔Aydınlar et al. (2024) [49]✔✔✔✔✔Ayre et al. (2024) [50]✔✔✔✔✔✔✔Baumgärtner et al. (2024) [51]✔✔✔✔✔✔✔✔Boscardin et al. (2024) [52]✔✔✔✔✔✔Bukar et al. (2024) [53]✔✔✔✔✔Burns et al. (2024) [54]✔✔✔✔✔✔✔✔Charow et al. (2021) [55]✔✔✔✔✔✔✔Chen et al. (2024) [56]✔✔✔✔✔✔✔✔Chu et al. (2022) [57]✔✔✔✔✔✔✔Chun et al. (2025) [58]✔✔✔✔✔✔✔Dave & Patel (2023) [59]✔✔✔✔✔✔Duran et al. (2025) [60]✔✔✔✔✔✔✔Dzangare & Gulu (2025) [37]✔✔✔✔✔✔✔Feigerlova et al. (2025) [61]✔✔✔✔✔✔Fitzek & Choi (2024) [62]✔✔✔✔✔✔✔Fitzpatrick (2023) [38]✔✔✔✔✔✔✔✔Periaysamy et al. (2023) [63]✔✔✔✔✔✔✔✔Garcia Valencia et al. (2024) [64]✔✔✔✔✔✔✔✔García-López et al. (2023) [65]✔✔✔✔✔✔✔✔Gibson et al. (2024) [66]✔✔✔✔✔✔Griewing et al. (2024) [67]✔✔✔✔Halawani et al. (2024) [68]✔✔✔✔✔✔Hoelscher & Pugh (2025) [69]✔✔✔Jensen et al. (2021) [70]✔✔✔✔Jha et al. (2022) [71]✔✔Joseph et al. (2025) [72]✔✔✔✔✔Koch et al. (2025) [73]✔✔✔✔✔✔✔Kurniawan et al. (2024) [74]✔✔✔✔Lau et al. (2023) [75]✔✔✔Lawrence & Levine (2024) [76]✔✔Li et al. (2022) [77]✔✔Love et al. (2025) [35]✔✔✔✔Malek et al. (2023) [78]✔✔✔Maqbool et al. (2025) [79]✔✔✔✔Mertala et al. (2022) [80]✔✔Mheidly (2024) [81]✔✔✔✔Milne-Ives et al. (2020) [82]✔✔✔✔Mittal et al. (2024) [83]✔✔✔✔Mohammed et al. (2024) [84]✔✔✔✔Will et al. (2025) [85]✔✔✔✔✔✔✔Mousavi Baigi et al. (2023) [86]✔✔Naamati-Schneider (2024) [87]✔✔✔Naseer et al. (2025) [88]✔✔✔✔Ng et al. (2022) [89]✔✔✔✔Peres (2024) [24]✔✔✔✔Rädel-Ablass et al. (2025) [90]✔✔✔✔✔✔Rincón et al. (2025) [91]✔✔✔✔✔✔✔Rodger et al. (2025) [92]✔✔✔✔Ruiz-Núñez et al. (2024) [93]✔✔✔✔✔Saigí-Rubió et al. (2024) [94]✔✔✔✔✔✔Sallam (2023) [23]✔✔✔✔✔✔✔✔✔Sallam et al. (2023) [95]✔✔✔✔✔Sapci & Sapci (2020) [96]✔✔✔✔✔✔Secinaro et al. (2021) [97]✔✔✔✔✔✔✔✔Singla et al. (2024) [98]✔✔✔✔✔✔Skryd & Lawrence (2024) [99]✔✔✔✔✔✔✔Soares et al. (2024) [100]✔✔✔✔✔✔Sridharan & Sequeira (2024) [101]✔✔✔✔✔✔Sumner et al. (2025) [102]✔✔✔✔✔✔✔Teng et al. (2022) [103]✔✔✔✔Waldock et al. (2024) [104]✔✔✔✔✔✔✔✔Wang & Li (2024) [105]✔✔✔✔✔✔✔✔✔Wubineh et al. (2024) [106]✔✔✔✔✔✔✔✔Yang et al. (2025) [107]✔✔✔✔✔✔Zahlan et al. (2023) [108]✔✔✔✔✔✔✔✔Nasra et al. (2025) [109]✔✔✔✔✔✔✔✔ Thematic frequency matrix (n = 75) Ethical and technical challenges were identified in 18 of 75 studies (24%), including algorithmic bias, limited AI literacy, and data privacy concerns. Digital inequities were reported in 15 of 75 studies (20%), emphasizing the importance of inclusive design and equitable access. These findings indicate that while AI tools can enhance cognitive and behavioral learning outcomes, careful implementation is required to address ethical, technical, and contextual factors. Despite the promising results, substantial gaps remain. Few studies employed qualitative or mixed-method designs to explore learners’ experiences, perceptions, or psychological responses to AI-based interventions. Similarly, there is limited evidence from low- and middle-income domains, suggesting that findings may not generalize globally. Finally, inconsistencies in reporting AI system architecture, training data, and pedagogical integration hinder meta-analysis and replication. General, integrating AI tools into health curricula presents clear opportunities to support tailored learning, metacognitive skill development, and culturally responsive education. Future research should prioritize empirical, cross-cultural, and longitudinal studies to evaluate both learning and psychological outcomes, while systematically addressing technical, ethical, and accessibility challenges. Thematic synthesis Based on a complete set of 75 published studies, the literature is categorized into three interrelated themes aligned with the study’s objectives: (1) AI-driven educational tools, (2) their effectiveness of AI tools in promoting health literacy, and (3) ethical and technical challenges in health education. AI-driven educational tools in health education The synthesis of 75 studies indicates that individual AI units in health learning can be widely classified into five main functional categories (see Table 7): (1) Large language model (LLM)-based simplifiers, (2) Domain-specific interactive agents, (3) Adaptive learning and recommendation systems, (4) AI-driven courses and storytelling tools, and (5) Multimodal AI-supported platforms. These units are implemented across different target groups, including students, health teachers, and the public, with the shared goal of adapting the educational material to match the learners’ reading skills and learning needs [64]. Table 7Thematic matrix of personalized AI-driven learning tools in health education (n = 75)Tool CategoryDescriptionExample Studies (Citation)Core FunctionsUsage ContextsLLM-based simplifiersTools based on large language models that rewrite complex health texts into simplified language suitable for general literacy levels. Sallam, 2023 [32]; Periaysamy et al., 2023 [63]; Halawani et al., 2024 [68]; Skryd & Lawrence, 2024 [99]; Waldock et al., 2024 [104]; Ayre et al., 2024 [50]; Will et al., 2025 [85].Simplifying content, preserving key messages, increasing accessibility for non-specialist audiencesClassroom settings, community outreach campaignsDomain-specific interactive agents (Chatbots)Chatbots specialized in specific health domains providing clear, interactive dialogues to reinforce health concepts. Kurniawan et al., 2024 [74]; Burns et al., 2024 [54]; Gibson et al., 2024 [66]; Asiksoy, 2025 [47]; Au et al., 2023 [48]; Halawani et al., 2024 [68]; Milne-Ives et al., 2020 [82]; Baumgärtner et al., 2024 [51].Delivering interactive answers, enhancing engagement, supporting self-directed learningCommunity awareness programs, student trainingAdaptive learning and recommendation systemsLearning systems that adapt in real time to learner progress by adjusting content sequencing and assessments; recommender systems provide bite-sized lessons based on user behavior. Periaysamy et al., 2023 [63]; Jha et al., 2022 [71]; Joseph et al., 2025 [72]; Waldock et al., 2024 [104]; Yang et al., 2025 [107]; Wangsa et al., 2024 [12]; Fitzpatrick, 2023 [38].Real-time content personalization, customized feedback, increasing learner retentionInformal learning, mobile health apps, daily health routinesAI-driven courses and storytelling toolsTools that generate personalized curricula from user data and enable creation of interactive health-related narratives, especially for youth and caregivers.Maqbool et al.,2025 [79]; Naseer et al., 2025 [88]; Rincón et al., 2025 [92]; Singla et al., 2024 [98]; García-López et al., 2023 [65]; Aloi et al., 2025 [45]; AlZaabi & Masters, 2025 [46]; Aydınlar, 2024 [49].Creating individualized education plans, enabling interactive storytelling, motivating participationCaregiver education, adolescent health programs, mental health educationMultimodal AI-supported platformsPlatforms that support creation of visual, audio, and textual public health content, including infographics, micro-lessons, and voice-guided health advice. Charow et al., 2021 [55]; Wubineh et al., 2024 [106]; Aloi et al., 2025 [45]; Fitzpatrick, 2023 [38]; Wangsa et al., 2024 [12].Producing multimodal content, enhancing interactivity, supporting digital learning, delivering daily push notificationsSchool health campaigns, social media outreach, elder adult educationMultilingual AI Translation ToolsIntelligent translation tools that simplify and localize health education content across multiple languages to support diverse linguistic populations. Ayre et al., 2024 [50]; Adegboye, 2024 [41];Simplified translation, cultural localization, expanding access to health contentMultilingual communities, cross-border health educationAI-powered Quiz GeneratorsTools that create personalized health quizzes with adaptive difficulty and automated feedback to support interactive assessment and learning. Adegboye, 2024 [41]Designing personalized assessments, automated feedback, adaptive difficultyAcademic education, health awareness, self-assessment Thematic matrix of personalized AI-driven learning tools in health education (n = 75) Periaysamy et al., 2023 [63]; Jha et al., 2022 [71]; Joseph et al., 2025 [72]; Waldock et al., 2024 [104]; Yang et al., 2025 [107]; Wangsa et al., 2024 [12]; Fitzpatrick, 2023 [38]. Charow et al., 2021 [55]; Wubineh et al., 2024 [106]; Aloi et al., 2025 [45]; Fitzpatrick, 2023 [38]; Wangsa et al., 2024 [12]. Large language model (LLM)-based simplifiers: These tools rewrite complex health texts in more accessible formats, improving readability for wider target groups. For example, ChatGPT-4.0 reduced the reading level for information on kidney donation from grade 9.6 to 4.3, maintaining essential information while simplifying syntax [23, 63, 68, 99, 104]. Other studies, such as Ayre et al. [50] and Will et al. [85], noted that simplified outputs maintained a substantial proportion of key information while reducing syntactic complexity. Domain-specific interactive agents: AI chatbots provide condensed learning experiences that reinforce health concepts [54, 66, 74]. For instance, Lucy Liverbot, designed for liver health, was positively rated for user confidence and understanding [47, 48, 68, 82]. Similarly, Praska Chatbot improved men’s comprehension of prostate screening and contributed to broader social and educational initiatives [51]. Adaptive learning and recommendation systems: These systems tailor material distribution in real-time according to student progress and interaction patterns. They support individualized sequencing, adaptive assessment, and micro-text delivery via mobile platforms, often enhancing engagement and performance [12, 38, 63, 71, 72, 104, 107]. AI-driven courses and storytelling tools: These tools enable personalized educational plans and narrative-based learning experiences [79, 88, 91, 98]. For example, García-López et al. [65] developed a system aligning geriatric health education with nursing standards (NANDA, NOC, NIC). Interactive storytelling tools, such as Twine, engaged adolescents in co-creating health narratives [41, 46, 49]. Multimodal AI-supported platforms: These include infographic creation tools (Visme AI, Canva AI), voice-interactive assistants (Alexa Health Skills, Google Assistant), multilingual translation applications (DeepL, Google Health Translate), and AI-based quiz generators (Quizgecko, AI-QuizBot). Collectively, these units supported health learning campaigns in areas such as hygiene, nutrition, weight management, and disease prevention. Their efficiency varies depending on context and target population [41, 50, 55, 106]. These AI units have been implemented across diverse groups, including students, health teachers, and the general public, with the shared goal of adapting educational content to learners’ reading skills and learning needs [64]. Overall, the evidence suggests that AI-driven tools enhance learner engagement, comprehension, and metacognitive skills, particularly in preventive health education. While promising, these technologies typically complement rather than replace traditional educational or clinical guidance, and effectiveness may vary across contexts. This thematic categorization provides a clear framework for understanding how AI can be integrated into health education, guiding curriculum designers and educators in developing personalized, learner-centered interventions that also consider ethical, technical, and accessibility challenges. Effectiveness of AI tools in promoting health literacy Analysis of the 75 reviewed studies indicates that AI tools can enhance health and digital literacy, addressing six main areas related to information access, personalization, reflective thinking, comprehension, engagement, and skill development (see Table 8). Access and comprehension: AI tools, such as large language models (LLMs), simplify complex medical texts into clear and accessible language without losing essential content, thereby bridging the gap between specialized knowledge and learners [50]. For example, Prosca, designed for prostate cancer patient education, improved understanding and encouraged patients to make informed decisions [51]. Multilingual applications, such as Spanish-language implementations, preserved key messages while adapting content for local comprehension [60]. Access and comprehension: AI tools, such as large language models (LLMs), simplify complex medical texts into clear and accessible language without losing essential content, thereby bridging the gap between specialized knowledge and learners [50]. For example, Prosca, designed for prostate cancer patient education, improved understanding and encouraged patients to make informed decisions [51]. Multilingual applications, such as Spanish-language implementations, preserved key messages while adapting content for local comprehension [60]. Reflective thinking and metacognitive skills: AI-driven interactive platforms support deeper engagement with health knowledge. Preclinical medical students reported increased preparedness for emergency situations, reflecting enhanced critical thinking and metacognitive engagement [46]. AI-backed digital environments also increased user confidence in utilizing electronic health resources [40, 43]. Reflective thinking and metacognitive skills: AI-driven interactive platforms support deeper engagement with health knowledge. Preclinical medical students reported increased preparedness for emergency situations, reflecting enhanced critical thinking and metacognitive engagement [46]. AI-backed digital environments also increased user confidence in utilizing electronic health resources [40, 43]. Personalized learning and emotional support: Adaptive AI systems tailor learning according to individual performance and interests. Some platforms, such as Lucilivarbot, provided limited emotional support while guiding patients through educational content, enhancing personalized learning experiences [48]. Personalized learning and emotional support: Adaptive AI systems tailor learning according to individual performance and interests. Some platforms, such as Lucilivarbot, provided limited emotional support while guiding patients through educational content, enhancing personalized learning experiences [48]. Cultural and linguistic adaptation: AI tools that consider cultural and linguistic contexts make educational content more understandable and engaging. Current evidence remains limited, and further research is needed to evaluate their effectiveness across broader populations [29, 37, 41]. Cultural and linguistic adaptation: AI tools that consider cultural and linguistic contexts make educational content more understandable and engaging. Current evidence remains limited, and further research is needed to evaluate their effectiveness across broader populations [29, 37, 41]. Social interaction and engagement: Involving users in the co-design of digital tools has shown positive effects on engagement and learning. However, these results are often constrained by small sample sizes or specific populations [44, 54]. Social interaction and engagement: Involving users in the co-design of digital tools has shown positive effects on engagement and learning. However, these results are often constrained by small sample sizes or specific populations [44, 54]. Potential as educational resources: Well-designed AI tools can simplify health information, increase user engagement, and accommodate cultural diversity. Effectiveness varies depending on design, implementation, and target population [52, 64, 106]. Potential as educational resources: Well-designed AI tools can simplify health information, increase user engagement, and accommodate cultural diversity. Effectiveness varies depending on design, implementation, and target population [52, 64, 106]. Table 8Thematic matrix of included studies on the effectiveness of AI tools in promoting health literacy (n = 75)ThemeConcise Description of What the Literature ShowsStudies1. Making health content easy to read and rememberLarge-language models and specialised chatbots turn dense medical prose into plain, memorable language, raising readability while preserving core meaning. Ayre et al., 2024 [50]; Baumgärtner et al., 2024 [51]; Duran et al., 2025 [56]2. Building metacognitive skills in academic settingsAI curricula prompt learners to reflect on AI’s limits and potential, boosting confidence in navigating digital health resources and fostering deeper self-regulated learning. AlZaabi & Masters, 2025 [46]; Alkhaaldi et al., 2023 [43]; Abou Hashish & Alnajjar, 2024 [40]3. Personalising learning paths and emotional supportAdaptive engines and conversational agents curate next-step content in real time and provide empathic feedback, increasing trust and sustaining engagement—especially for chronic-disease education. Au et al., 2023 [48]4. Embedding culture and language for wider reachAI translation and culturally tuned dialogue systems make health messages relatable across languages and contexts, helping diverse communities evaluate and act on online information. Marzo, 2024 [29]; Adegboye, 2024 [41]; Garcia Valencia et al., 2024 [64]5. Catalysing social innovation and shared ownershipCo-design with target users (e.g., adolescents) transforms learners from passive recipients to active creators, markedly elevating engagement and knowledge gains. Aloi et al., 2025 [45]; Charow et al., 2021 [55]6. From novelty to strategic assetWhen guided by ethical safeguards and curricular alignment, AI tools simplify texts, personalise learning, respect culture, and cultivate critical appraisal skills—laying the groundwork for a health-literate citizenry. Boscardin et al., 2024 [52]; Wubineh et al., 2024 [106].Note: Themes were derived through a thematic synthesis of all 75 included studies; references listed are illustrative exemplars rather than exhaustive citations for each theme Thematic matrix of included studies on the effectiveness of AI tools in promoting health literacy (n = 75) Note: Themes were derived through a thematic synthesis of all 75 included studies; references listed are illustrative exemplars rather than exhaustive citations for each theme The evidence suggests that AI tools can complement traditional health education by enhancing accessibility, comprehension, engagement, and personalized learning, while also supporting metacognitive skill development. Nevertheless, the unevenness of evidence across contexts highlights the need for continued research to validate their broader educational and psychological impacts. Ethical and technical challenges in health education A thematic synthesis of 75 reviewed studies reveals that while AI-powered educational systems hold significant promise for advancing health and digital literacy, their integration introduces complex ethical and technical challenges that require structured governance and continuous ethical monitoring. Ethical Challenges: Four primary ethical dimensions emerge across the reviewed literature, algorithmic bias, transparency, data privacy, and digital inequity, each influencing the fairness, reliability, and inclusivity of AI-driven health education (see Table 9). In addition, broader ethical considerations involve responsible handling of sensitive learner data, risks of over-reliance on AI outputs, potential for inappropriate guidance, and the need for ethical oversight in AI-mediated interactions [110, 111]. Table 9Thematic matrix of included studies for ethical and technical challenges in AI-Driven health educationThemeDescriptionReferencesAccuracy and ReliabilityAI technologies may produce inaccurate or unreliable information due to algorithmic bias, contextual insensitivity, or over-reliance by users on unverified AI outputs for therapeutic support. Such dependence may lead to inappropriate guidance, incorrect interpretations, or even harmful recommendations. Emotional or empathetic responses generated by AI can further amplify user trust in potentially flawed outputs. Sub-challenges: lack of human oversight, propagation of misinformation, and limited content validation mechanisms. Burns et al., 2024 [54]; Adegboye, 2024 [41]; Chun et al., 2025 [36]; Hoelscher & Pugh, 2025 [69]; Wubineh et al., 2024 [106]; Wang & Li, 2024 [105]; Love et al., 2025 [35].Transparency and ExplainabilityLearners’ difficulty in understanding how AI systems generate responses reduces trust and critical engagement. Ethical challenges include the “black-box” nature of algorithms, insufficient explanation of reasoning processes, and lack of user awareness regarding AI limitations. Sub-challenges: unclear accountability, absence of traceable decision pathways, and limited disclosure of model biases or data sources. Chen et al., 2024 [56]; Boscardin et al., 2024 [52]; Dzangare & Gulu, 2025 [37].Data Privacy, confidentiality and ConsentHandling of sensitive behavioral and health-related data raises major ethical concerns, particularly when consent mechanisms or anonymization are unclear. Sub-challenges: misuse of personal or psychological data, vulnerability to breaches, inadequate user control over shared information, and lack of institutional data governance. Adegboye, 2024 [41]; Wubineh et al., 2024 [106]; Chun et al., 2025 [36]; Naseer et al., 2025 [88]; Mittal et al., 2024 [83]; Love et al., 2025 [35]; Fitzpatrick, 2023 [38]; Wang & Li 2024 [105].Digital EquityUnequal access to AI-driven tools due to disparities in digital infrastructure, literacy, and affordability can deepen learning gaps. Sub-challenges: socio-economic barriers, unequal access to adaptive AI technologies, and exclusion of digitally underprepared learners. Hoelscher & Pugh, 2025 [69]; Wubineh et al., 2024 [106]Interoperability and IntegrationThe limited compatibility of AI systems with existing educational or health curricula hinders widespread adoption. Ethical challenges include not ensuring safe, appropriate, contextually reliable publication that is incompatible with learning management systems and institutional standards. Sub-challenges: fragmented systems design, inconsistent learning outcomes, and lack of technical standardization. Feigerlova et al., 2025 [61]Cultural and Linguistic RelevanceAI tools often fail to reflect local languages, social norms, or cultural health beliefs, leading to bias and limited usability. Ethical responsibilities involve ensuring cultural sensitivity, inclusivity, and contextual adaptation. Sub-challenges: ethnocentric design, linguistic oversimplification, exclusion of marginalized cultural perspectives, and failure to adapt educational materials for diverse learning environments. Marzo, 2024 [29]; Duran et al., 2025 [60]Faculty Readiness and TrainingEducators may lack the technical and ethical competencies needed to integrate AI effectively. Challenges include insufficient training, weak institutional support, and uncertainty in managing AI-generated outputs responsibly, misuse due to limited literacy, ethical blind spots, and a lack of professional development pathways and security orientation. AlZaabi & Masters, 2025 [46]; Ajalo et al., 2025 [42]; Wubineh et al., 2024 [106]; Naamati-Schneider, 2024 [87].Co-Design and Ethical FrameworksEngaging learners and educators in co-creation and adopting structured ethical models (e.g., RUBRICC) ensures inclusive and safe AI deployment. Sub-challenges: insufficient stakeholder participation, absence of ethical oversight committees, weak continuous monitoring and evaluation, and lack of transparent governance mechanisms. Aloi et al., 2025 [45]; Wang & Li, 2024 [105]; Naamati-Schneider, 2024 [87]. Thematic matrix of included studies for ethical and technical challenges in AI-Driven health education Adegboye, 2024 [41]; Wubineh et al., 2024 [106]; Chun et al., 2025 [36]; Naseer et al., 2025 [88]; Mittal et al., 2024 [83]; Love et al., 2025 [35]; Fitzpatrick, 2023 [38]; Wang & Li 2024 [105]. First, algorithmic bias occurs when datasets used to train AI models reflect unbalanced representations of populations, resulting in unequal learning outcomes and potential reinforcement of social or cultural stereotypes. Such bias can distort learners’ understanding of health information or marginalize underrepresented groups, emphasizing the need for transparent data auditing and bias-mitigation strategies [41, 54, 105]. Second, transparency and explainability remain critical but underdeveloped. Many AI applications function as “black boxes,” providing outputs without revealing how decisions or recommendations are generated. This opacity limits accountability and erodes learner trust, particularly in sensitive health education contexts where clarity of reasoning is essential [37, 52, 56]. Third, data privacy and consent represent major ethical concerns. AI learning systems often rely on continuous data collection, tracking students’ behavior, performance, and even emotional responses, to personalize feedback. Without robust anonymization, consent mechanisms, or institutional data governance, learners’ personal information may be vulnerable to misuse or breaches [36, 41, 88, 106]. Finally, digital inequity remains a structural ethical issue. Students in low-resource environments or with limited digital literacy may be excluded from AI-enhanced learning, thereby reinforcing existing educational inequalities. Equitable access to AI infrastructure and localized support is thus essential for inclusive adoption [69, 106]. Technical Challenges: On the technical front, issues such as systems interoperability, contextual adaptation, and faculty readiness constrain sustainable implementation. AI tools often operate in isolation from institutional learning management systems, making scalability difficult [61]. Moreover, lack of localization, such as failure to align with cultural norms, language diversity, or regional health priorities, reduces pedagogical relevance [29, 60]. Faculty members may also face challenges due to inadequate training in AI ethics, digital literacy, and oversight mechanisms [42, 46, 87]. To address these interconnected challenges, scholars have proposed co-design frameworks and structured ethical models (e.g., RUBRICC) to enhance transparency, accountability, and participatory governance in AI-mediated education [45, 87, 105]. Adopting such models can help institutions establish ethical review committees, promote human oversight, and ensure continuous evaluation of AI-driven educational systems. Discussion By triangulating tool typologies, outcome data, and cross-cutting challenges, this thematic synthesis substantiates AI’s dual role as both a pedagogical accelerator and a driver of social innovation (SI) in health education. This section integrates SI explicitly into the analysis of AI-driven health literacy interventions, and research gaps, future agendas, and study limitations are articulated in depth. Observed research gaps A critical mapping of the 75 studies reviewed shows that scholarship on AI‑enabled health‑literacy interventions remains heavily skewed toward clinical and tertiary‑education contexts, with comparatively sparse attention to community or school‑based learning environments. More than half of the empirical papers evaluate tools inside medical, nursing, or allied health curricula, AI‑ready scales for medical students [46], and digital‑literacy audits in health‑science faculties [49], whereas only a handful explore population‑level or public‑library implementations, such as adolescent co‑design workshops [45] or cross‑sectional surveys of lay users of online health information. This concentration leaves a sizeable gap in understanding how AI tools perform in non‑formal education settings, particularly for adults with limited schooling or low baseline e-health literacy skills. From a social innovation perspective, this limits insights into how AI can generate equitable, community-centered benefits beyond formal education. Even within higher education, most studies center on formative perceptions rather than summative learning outcomes; Feigerlova et al. [61] found that only 12 papers reached Kirkpatrick level-2 knowledge testing, and none assessed behavioral transfer into real world self-care. The geographical distribution of evidence is likewise imbalanced. Researchers affiliated with the United States, the United Kingdom, and Australia authored nearly 60% of the sampled papers, whereas contributions from low and middle-income countries (LMICs) are markedly under represented [42, 106]. This disparity limits the generalizability of findings to domains where infrastructural constraints, language plurality, and cultural factors shape both digital access and health information needs. For example, while Spanish-language readability experiments demonstrate promising gains using ChatGPT [60], few studies interrogate AI adaptation for Arabic, African, or South-Asian contexts, despite evidence that linguistic tailoring drives engagement [29]. Moreover, only two investigations explicitly analyzed gender or rural-urban differentials in AI-mediated learning outcomes [40, 62], signaling a wider neglect of equity-oriented metrics critical to socially innovative outcomes. Methodologically, the corpus leans toward cross-sectional surveys and short usability pilots, with limited longitudinal or randomized-controlled evaluations capable of attributing causal change in health-literacy competencies. Burns et al. [54] note that conversational agents can raise immediate comprehension, yet neither tracks sustained knowledge retention nor behavior modification. Similarly, prototype studies such as Lucy Liver Bot focus on qualitative user satisfaction without parallel objective literacy assessments [48]. Only one RCT—PROSCA for prostate cancer education, measured reduction in information needs over time, but even this trial omitted standardized health literacy scales [51]. An additional gap concerns theoretical anchoring: while several authors invoke “AI literacy” frameworks [52, 69], few integrate established health literacy models or behavior change theories, hindering comparative synthesis across studies. From a social innovation lens, this gap constrains the understanding of how AI interventions can systematically produce equitable, socially beneficial outcomes. Finally, technical ethical dimensions receive descriptive rather than empirical treatment. Privacy, algorithmic bias, and explainability are frequently mentioned as risks [53, 56], yet rigorous audits of data‑handling practices or bias mitigation strategies are scarce. Only Chun et al. [36] empirically linked digital equity barriers to differential learning outcomes in an AI textbook intervention, underscoring the need for finer-grained analytics on how connectivity, device capability, and user experience design mediate effectiveness. Collectively, these gaps reveal an urgent research agenda: move beyond perception studies toward longitudinal, theory‑grounded, and contextually diverse evaluations that quantify how AI tools translate into durable, equitable gains in health literacy, while explicitly considering social innovation implications to maximize community-level and societal impact. Future research directions Future directions- theory Recent work shows that AI‑driven health‑literacy interventions are advancing faster than the theoretical frameworks that should guide their design and evaluation. Many studies still default to generic acceptance models such as TAM or UTAUT, without adapting them to the nuances of adaptive dialogue, cultural tailoring, or algorithmic co‑creation that typify modern AI tools [52, 56]. Future research must therefore develop, or extend, domain‑specific theories that integrate constructs from digital‑health literacy, critical AI literacy, and behavior change science. For instance, integrating Nutbeam’s three‑level literacy model with AI Literacy concepts such as explainability and data provenance could clarify which tiers of competence are most sensitive to chatbot‑mediated instruction [37, 45]. Second, current evaluations rarely capture the metacognitive and affective processes triggered by conversational agents. Studies report gains in “perceived readiness” or “readability,” yet seldom theories how adaptive feedback loops foster reflection, self‑efficacy, or trust [46, 48]. Future models should therefore embed constructs from Self-Regulated Learning and Social Cognitive Theory to explain why tools like Lucy Liver Bot simultaneously inform and reassure patients living with chronic liver disease [48]. Such integration would move the field beyond surface metrics toward richer causal accounts of how AI personalization strengthens health agency. Finally, there is a theoretical blind‑spot regarding equity mechanisms. Although many authors mention “digital divide” or “algorithmic bias,” few specify pathways by which AI either widens or narrows disparities in health knowledge [60, 106]. Building on Capability Theory and Intersectionality, scholars could theories how language localization, cultural narratives, and participatory co‑design (e.g., adolescents scripting their own DHL storylines) reconfigure informational capabilities among marginalized groups [41, 45]. Testing such frameworks will enable researchers to predict when AI adaptation leads to genuine empowerment versus mere technological novelty. Future directions- social and behavioral context Empirical coverage of AI‑enabled health‑literacy programs remains heavily skewed toward high‑income, English‑speaking settings. Half of all trials and quasi‑experimental studies in this review were conducted in the USA or UK, yet LMIC scholars contributed fewer than 10% of Q1‑level publications [37]. Addressing this imbalance requires context‑sensitive investigations that map infrastructural, linguistic, and cultural determinants of AI uptake. For example, Ugandan medical students reported enthusiastic adoption of ChatGPT for coursework, but also highlighted older peers’ lower usage and the absence of formal institutional guidance [42]. Similar contextual audits should be replicated across diverse educational tiers, secondary schools, community colleges, and vocational nursing programs to identify where low bandwidth chatbots or SMS‑based AI agents might outperform data heavy large language model (LLM) interfaces. Another pressing contextual gap concerns minority languages and culturally specific health beliefs. Readability studies show that LLMs can reduce Spanish patient‑education texts to a sixth-grade level [60], yet few tools address Indigenous or sign language communities. Social media and digital platforms increasingly shape health literacy and AI-mediated education, especially among younger populations. Platforms such as Facebook, Twitter, and Instagram serve as channels for health information dissemination, peer learning, and engagement with AI-driven educational tools. Integrating AI interventions with these platforms may enhance accessibility, foster collaborative learning, and reinforce the adoption of evidence-based health practices. Recognizing this interaction underscores a complementary pathway for AI-supported health literacy initiatives beyond formal educational settings and highlights the importance of considering social and behavioral contexts in future research [112]. Future work should test multimodal and multilingual AI systems that generate voice, text, and pictorial explanations aligned with local idioms, dietary norms, and health practices [29, 64]. Moreover, the ethics of data sourcing must be scrutinized: if minority dialects are underrepresented in training data, personalization engines risk perpetuating knowledge gaps rather than closing them [68]. Finally, contextual research should examine institutional ecosystems. Nursing frameworks such as NURSES‑AI emphasize that workplace culture, leadership advocacy, and EHR integration determine whether frontline staff actually use AI literacy aids [69]. Pilot studies in municipal healthcare centers show that tools like the Conversational Health Literacy Assessment Tool only sustain usage when providers have clear action pathways after assessing literacy needs [70]. Future implementation studies must therefore track organizational readiness, reimbursement models, and policy incentives that modulate AI adoption across public health, academic, and community settings. Future directions- methodology Methodologically, the evidence base remains dominated by cross‑sectional surveys and single‑site pilots. Only four randomized controlled trials (RCTs) were identified in the entire corpus, two of which focused on cancer chatbots [51, 66], and a few studies reported follow‑up beyond three months. Future research should priorities multicenter RCTs that measure not only immediate literacy gains but also medium‑term behavior change (e.g., screening uptake, medication adherence) and downstream clinical outcomes [61]. Second, richer mixed‑methods designs are needed. Quantitative readability scores (SMOG, Flesch) tell us little about how users negotiate AI advice within their social networks. Ethnographic observations and think‑aloud protocols could uncover interpretive pitfalls or trust dynamics invisible to survey metrics [47, 54]. Combining learning analytics with qualitative diaries would also clarify why adaptive sequencing boosts self‑regulated learning in some cohorts but not others [36]. Data transparency and reproducibility present additional methodological imperatives. Only 11 of 75 studies (14.7%) in this review disclosed training‑set provenance or hyper‑parameter details, limiting transparency, reproducibility, and the ability to conduct meta‑analysis or bias auditing. Journals should mandate AI Transparency Statements that specify model architecture, data lineage, and versioning. Researchers, meanwhile, can adopt open‑source evaluation pipelines such as REF‑AI and PEMAT‑AI to benchmark accuracy, actionability, and equity across competing chatbots [66]. Finally, future evaluations must operationalize fairness metrics. Algorithmic audits should stratify outcomes by gender, age, literacy baseline, and socioeconomic status to detect unintended disparities [37]. Simulation studies that stress test models under adversarial or low‑resource inputs could preempt real‑world failures in marginalized contexts [36]. By embedding equity checkpoints in study protocols, researchers will not only satisfy emerging regulatory standards but also ensure that AI‑enabled health‑literacy programs fulfil their promise of inclusive digital transformation. Practical recommendations for educators and AI implementers (SI Lens) Building on the thematic synthesis of 75 studies, AI-powered tools clearly offer substantial potential to enhance health literacy, while simultaneously serving as mechanisms for social innovation, improving access, reducing disparities, and empowering diverse learners. To translate these insights into actionable practice, Table 10 propose a structured checklist that links each domain to societal impact: Table 10Practical checklist for educators and AI implementers in health literacy (SI-Oriented)DomainKey ActionsRationale/EvidenceSocial Innovation ImpactExamplesAI Transparency & Explainability- State model/version (e.g., ChatGPT-4.0)- Describe training data and limitations- Disclose algorithmic pathways- Provide clear usage instructionsReduces “black-box” effect, builds learner trust [37, 52, 56]Enhances trust, supports informed societal use of AI.Chatbots for health Q&AData Privacy & Consent- Inform users about data collection- Implement anonymization & secure storage- Obtain explicit consent for trackingProtects learners’ data and ensures ethical compliance [36, 41, 88, 106]Protects vulnerable learners, ensures ethical societal impact.Personal health appsDigital & Cultural Equity- Localize content for languages- Adapt to low-bandwidth/offline contexts- Ensure inclusivity for marginalized groupsMitigates digital inequity and broadens access [29, 41, 69]Empowers underserved communities, reduces digital divides.Multilingual health modulesAlgorithmic Fairness- Conduct bias audits- Monitor outcomes for underrepresented groups- Adjust models to reduce disparitiesEnsures equitable learning outcomes, avoids stereotypes [41, 54, 105]Promotes social justice and fairnessInclusive AI content recommendationIntegration & Faculty Readiness- Align AI with curricula/LMS- Train educators in AI literacy/ethics- Support AI supervisionFacilitates adoption, enhances pedagogical relevance [42, 46, 61, 83]Ensures societal benefits extend from classroom to community.AI-assisted lesson planningCo-Design & Ethical Oversight- Engage learners/educators in development- Establish governance & ethical review- Continuous feedback loopsPromotes participatory, accountable AI deployment [45, 87, 105]Fosters community ownership, empowerment.Co-created AI health platformsMonitoring & Evaluation- Track short-term learning gains and long-term behavior- Use mixed-methods evaluation- Report findings openlyEnables evidence-based improvement [36, 61, 66]Measures societal impact, supports continuous improvement.Evaluation of health literacy programs Practical checklist for educators and AI implementers in health literacy (SI-Oriented) - State model/version (e.g., ChatGPT-4.0) - Describe training data and limitations - Disclose algorithmic pathways - Provide clear usage instructions - Inform users about data collection - Implement anonymization & secure storage - Obtain explicit consent for tracking - Localize content for languages - Adapt to low-bandwidth/offline contexts - Ensure inclusivity for marginalized groups - Conduct bias audits - Monitor outcomes for underrepresented groups - Adjust models to reduce disparities - Align AI with curricula/LMS - Train educators in AI literacy/ethics - Support AI supervision - Engage learners/educators in development - Establish governance & ethical review - Continuous feedback loops - Track short-term learning gains and long-term behavior - Use mixed-methods evaluation - Report findings openly This checklist translates research insights into actionable steps for educators, developers, and policymakers. Each action not only enhances learning outcomes but also strengthens equity, inclusion, and empowerment. For instance: AI transparency and fairness prevent bias and build trust. Data privacy measures protect vulnerable learners. Curriculum integration and faculty training ensure responsible use. Participatory co-design and monitoring foster community ownership and long-term societal impact. By following this framework, health education initiatives can leverage AI not only for cognitive gains but also to generate meaningful societal transformation, ensuring that technology serves learners and communities equitably and ethically. Limitations This study has several limitations that should be considered when interpreting the findings. First, a language bias may exist, as only English-language publications were included. This linguistic restriction has led to the exclusion of relevant contributions from non-English-speaking domains, especially in contexts where culturally embedded practices in health education could offer valuable insights. Future reviews should therefore consider incorporating multilingual sources to enhance inclusivity and global representation. Second, the review focused primarily on publications since 2000, and the scope of data sources was restricted to high-ranking (Q1–Q2) journals indexed in PubMed/Medline, Scopus, ScienceDirect, and Connectedpapers.com. This selection was intended to enhance the reliability and quality of the review by including studies with rigorous methodologies, high impact, and robust peer review. However, this approach may have excluded relevant perspectives or emerging evidence from lower-ranked (Q3/Q4) journals, grey literature, or earlier periods. Consequently, some perspectives or insights may be underrepresented, which could slightly limit the comprehensiveness of the conclusions and represent an opportunity for future research. It should be explicitly noted that statements about the predominance of studies in Q1/Q2 journals reflect this inclusion criterion, rather than an inherent characteristic of the research field. Third, all screening and data extraction steps were conducted independently by two reviewers, with any discrepancies resolved through discussion and arbitration by a senior reviewer. While this dual-review process and consensus arbitration aimed to minimize bias, minor subjective judgments may still have influenced inclusion decisions. Inter-rater reliability was not formally quantified using statistical measures (e.g., Cohen’s kappa), which represents a potential limitation of the study. Furthermore, although all included studies were appraised for methodological clarity and transparency, formal, standardized quality assessment tools (such as JBI, RoB 2, or AMSTAR 2) were not employed. This may limit cross-study comparability and represents an additional methodological limitation of this review. Additionally, this review was not prospectively registered on platforms such as PROSPERO or OSF. The absence of a registered protocol represents a methodological limitation in terms of transparency and pre-specification of study procedures, which may affect replicability and confidence in the selection and synthesis processes. Fourth, literature was identified primarily through PubMed/Medline, Scopus, and ScienceDirect, with searches restricted to title, abstract, and keyword fields to maximize relevance and precision. While this approach ensured focused retrieval of studies closely aligned with the review objectives, we acknowledge that limiting the search to these databases may have excluded additional education- and health-focused publications indexed in ERIC, PsycINFO, CINAHL, or Cochrane. This represents a potential limitation and an opportunity for future research. Fifth, the predominance of review-based research over empirical and mixed-method studies limits the depth of understanding regarding the real-world application and behavioral outcomes of AI-driven health education. Many included studies synthesized secondary evidence rather than generating primary data, which reduces the ability to draw robust conclusions about causality, implementation feasibility, and long-term learning impact. Future studies should prioritize experimental or longitudinal research in authentic educational settings. Sixth, there remains a contextual gap in representation. Most of the reviewed studies originated from high-resource countries, leaving limited empirical evidence from low- and middle-income domains. This imbalance restricts the generalizability of findings, particularly regarding infrastructural readiness, faculty competence, and learner access to AI tools in resource-constrained environments. Seventh, methodological and conceptual limitations arise from the reliance on keyword-based thematic mapping. This approach may overlook studies that employ alternative terminologies or interdisciplinary frameworks, thereby underrepresenting important subthemes such as ethical reasoning, digital equity, and culturally adaptive pedagogies. Lastly, technical and ethical constraints intrinsic to AI integration, such as data privacy, algorithmic bias, and institutional readiness, may have influenced reported outcomes and interpretations. These issues highlight the complex interplay between technology design, human factors, and educational ethics. Future research directions should address these boundaries by expanding inclusion criteria to non-English and grey literature sources; promoting empirical, cross-cultural, and large-scale investigations, particularly in low-resource settings; exploring longitudinal effects of AI adoption on learner outcomes and professional practice; and developing comparative frameworks integrating health education, psychology, and AI ethics to deepen theoretical and applied understanding. Conclusion This article systematically synthesizes the current landscape of AI-supported tools in health education, focusing on their potential contributions to promoting health literacy and skills. By reviewing 75 studies across various educational and technical contexts, findings indicate trends rather than definitive outcomes: AI-operated applications, particularly individual learning systems and conversational agents, may contribute to improved accessibility, student engagement, and understanding of health information in certain populations. The interdisciplinary nature of the field is evident across various publication types, highlighting growing interest in the intersection of AI, education, and digital health. However, these indications should be interpreted with caution, as evidence is uneven and largely descriptive. The review identifies ethical and technical challenges, including reliability of materials, digital inequities, and limited AI-literacy support. Many studies are short-term or observational, with few rigorous longitudinal or experimental designs, limiting strong conclusions regarding the impact of AI interventions. This review has several limitations affecting generalizability. Restricting inclusion to English-language, peer-reviewed publications may introduce linguistic and selection biases, potentially excluding relevant studies from non-English contexts, grey literature, or lower-tier journals. Variability in terminology, study design, and methodological rigor may also have led to the exclusion of pertinent evidence. Future research should adopt broader inclusion criteria, incorporate non-English sources, and evaluate AI interventions across diverse cultural and educational contexts. Emphasis on rigorous, longitudinal, and theory-informed studies will help clarify how AI tools could support health literacy more effectively and equitably. This review provides a foundational, cautious, and evidence-grounded synthesis to guide inclusive, ethically informed, and educationally meaningful AI interventions, while acknowledging the current limitations of available evidence. Acknowledgments Dual publications No, the results/data/figures in this manuscript have not been published elsewhere, nor are they under consideration (from you or one of your Contributing Authors) by another publisher. Declaration of generative AI and AI-assisted technologies in the writing process During this manuscript's preparation, the author(s) used ChatGPT, Scispace, QuillBot AI, ProWritingAid, and Grammarly to help produce the first draft, improve linguistic accuracy, and make sure the documents are free of errors in grammar, spelling, punctuation, and tone. Following the use of these resources, the author(s) thoroughly reviewed and edited the material as needed, and they take full responsibility for the publication's content integrity. Authors’ contributions Dina Mansour Tbaishat served as the corresponding author, secured the research grant, contributed to conceptualization, provided strategic guidance throughout the study, contributed to the refinement of the review methodology, and critically revised the manuscript for important intellectual content. Maha Waleed Elfadel led the conception and design of the review, conducted the literature search, performed data extraction and thematic synthesis, and drafted the manuscript. Both authors read and approved the final manuscript. Funding Through Zayed University’s Social Innovation Award 12096. Data availability This study is a systematic review; all data analysed during this study are extracted from the published articles cited in the manuscript. No new datasets were generated. Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate Not Applicable. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement This study is a systematic review; all data analysed during this study are extracted from the published articles cited in the manuscript. No new datasets were generated.