Ethical approval and consent to participate were not applicable as this study is a systematic scoping review that did not involve human participants or use identifiable personal data.

This systematic scoping review was conducted following the stringent guidelines of PRISMA-ScR, with search transparency adhering strictly to the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses – Search extension) [28] protocol to ensure full transparency and reproducibility.

A comprehensive search was executed across 11 electronic databases: Web of Science, Scopus, PubMed, Medline, CINAHL, Cochrane, ACM Digital Library, IEEE Xplore, ScienceDirect, APA PsycInfo, and Google Scholar. The final search date was July 28, 2025.

To ensure comprehensive coverage, particularly of the multimodal focus, the search terms were meticulously refined and expanded to include concepts covering underlying models and various modalities of interaction. The search encompassed three core conceptual areas, which are (1) GenAI technology (including terms such as “Generative Artificial Intelligence,” “LLM,” “Diffusion Model,” “Image Generation,” “ChatGPT,” “Multi-modal,” and “Voice Assistant”), (2) geriatric population (“older adults” and “elderly”), and (3) health or application (“health” and “health management”). All searches were restricted to English-language articles and document types of research articles or conference proceedings. The inclusion of Google Scholar was necessary to capture cutting-edge, nonindexed, or difficult-to-locate conference papers from the computer science domain. The specific search formulas incorporating all conceptual areas for each database are summarized in Table 1. The PRISMA-ScR and PRISMA-S checklists are provided in Checklists 1 and 2.

A detailed data extraction template was developed and refined after piloting on 5 articles. To support the multimodal and geriatric focus of this review, the extracted items were expanded to include author, year, country, research method type, specific GenAI technology and underlying model (eg, LLM), input and output content modalities (eg, text, voice, image, and sensor data), specific health application domain, target population characteristics (eg, mean age, cognitive status, level of frailty, and comorbidity), empirical impact metrics, potential, and future trends. All data were extracted by the 2 independent reviewers (TL and YTL), with a final cross-check performed for consistency.

A critical appraisal of the methodological quality of every included study was performed using the Mixed Methods Appraisal Tool (MMAT) [29]. The MMAT is a validated instrument specifically designed to assess various study designs, quantitative, qualitative, and mixed methods in systematic and scoping reviews. The assessment explicitly documented whether studies accounted for inherent GenAI model biases (eg, “hallucination”) or addressed the cultural appropriateness of their application in diverse older adult populations. The detailed MMAT scoring results were then used to inform the discussion regarding the certainty and strength of the synthesized evidence. The methodological quality of the 28 included studies [30-57] was assessed using the MMAT. Overall, the studies demonstrated moderate to high methodological quality, with strong performance in the consistency of the research question and design, as well as the clarity of data collection methods. However, the appraisal revealed deficiencies in 2 key areas, which collectively serve as crucial limitations for the current body of evidence in Table 2.

The findings were systematically collected, summarized, and analyzed using descriptive statistics to characterize the trends and features of the included articles (eg, temporal growth, technology distribution, and geographical concentration). Subsequently, a narrative synthesis was performed using these descriptive findings to rigorously address the research questions raised in the review. This process involved synthesizing the empirical impact reported by the studies, identifying recurring themes in application and challenge, and mapping the evidence across different geriatric health domains. All authors verified the final synthesis and interpretation.

According to Tricco and colleagues [58], and considering the peculiarity of the scoping review, we did not appraise the methodological quality or risk of bias of the included papers.

The systematic search across the 11 electronic databases initially yielded a total of 69,074 records. The selection process, detailed in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram (Figure 1), began with a thorough cleaning stage where 14,147 duplicate records were removed. An additional 32,206 records were marked as ineligible by automation tools, and 22,721 records were removed for other nonacademic reasons, specifically including 1354 articles published in other languages and 21,367 classified as review articles, theses, nonacademic publications, or book chapters. The title and abstract screening then commenced with 1358 unique records. This phase resulted in the exclusion of 1250 records, with the main reasons being studies focusing on populations younger than 60 years of age (n=37), research outside the field of older adults’ health care (n=424), and studies focusing only on older adults’ perceptions or attitudes without actual empirical application (n=863). This rigorous screening left 108 reports sought for full-text retrieval. Of these, 55 reports could not be retrieved, leaving 53 reports assessed for eligibility. The full-text assessment resulted in the final exclusion of 25 reports due to the lack of a complete research process and methods (n=15), a sole focus on technical performance (n=7), or unclear data reporting (n=3). Ultimately, a final set of 28 articles [30-57] met all inclusion criteria and were included in the scope of this systematic review (refer to Table 3 for article characteristics).

The final analysis incorporated 28 studies [30-57], with trends in publication and regional distribution characteristics of GenAI research in older adults’ health examined (Figures 2 and 3). Temporally, the studies exhibit a clear exponential growth trajectory—the annual publication rate averaged only 1.25 articles (SD 0.50) from 2019 to 2022, which then surged to 12 in 2024 and 11 in 2025 (overall mean 4.67, SD 5.32). This recent output accounts for more than 82% (23/28) of the total publications, indicating that GenAI applications are in a phase of rapid emergence driven by breakthroughs in GenAI technology, the escalating global aging crisis, and strategic policy emphases on digital health.

Geographically, research output demonstrates significant regional concentration. The United States (n=4) [42,43,46,50] and the China region (China: n=5 [32-34,55,57] and Hong Kong: n=2 [48,56]) collectively contributed 11 studies (11/28, 39.2%), establishing a clear leading position aligned with their synergistic advantages in cutting-edge research and development and vast older adult market demand. Following these leaders are countries or regions with specific aging pressures or policy interests, including Japan (n=2) [30,49], South Korea (n=2) [35,45], Taiwan (n=2) [39,40], Saudi Arabia (n=2) [44,52], and Thailand (n=2) [51,53]. European welfare states and developing nations also contribute to the literature, highlighting the diverse, context-specific demands for GenAI in addressing global aging challenges.

Methodologically, the studies predominantly used quantitative approaches (n=16) [31,33,36,37,39,40,44,45,47,51-57], serving as the primary evaluation framework to validate tool efficacy through statistical techniques like pre-post control designs and correlation tests. These were complemented by mixed methods research (n=7), which integrated quantitative tests with qualitative interview techniques for comprehensive data cross-validation, and qualitative studies (n=5) [32,34,43,46,50], which prioritized in-depth exploration of user subjective experiences, perceived barriers, and needs through thematic analysis and observational protocols (Table 4).

Regarding technological type (Table 5), general GenAI systems (basic chatbots; n=13) remain the most frequent single category, serving as the foundational technology for health information interaction and simple task automation. These basic systems are complemented by vertically integrated applications, which expand technical boundaries through GenAI+multimodal/specific application (n=14), encompassing systems for assisted living, multimedia generation, and therapeutic integration. The core value of GenAI technologies lies in multimodal interaction support and personalized service generation (n=3), where studies have integrated voice assistants with chatbots and leveraged visual question-answering models to reduce operational barriers and enable more natural conversational flows. However, the current technological landscape exhibits a homogenization tendency, with over one-third of the studies concentrating on basic health consultations, underscoring the need for future research to broaden GenAI’s functional boundaries in older adults’ health, validating cutting-edge directions, such as affective computing and ethical risk control.

The data interaction patterns demonstrate a clear multimodal input-output integration trend, built upon a text-based foundation (Table 6). Text (n=6) remains the most universally accessible interaction medium, but its relatively low frequency suggests that GenAI applications heavily rely on multimodal combinations to meet the diverse needs of older users. This is primarily due to the text’s low technical barrier and alignment with older adults’ communication preferences. Beyond text, studies integrate alternative input modalities such as voice (n=8), images (n=10), and physiological or environmental parameters (n=2) to enhance monitoring accuracy and address complex health issues. For output, text is the primary feedback modality, yet multimodal outputs serve as crucial supplements: music or audio (n=2) for mood regulation, images or videos (n=5) for visualizing complex health information, and gamified text (n=2) for cognitive training. Although multimodal applications are concentrated in specific scenarios, the overall model is text-centric, supplemented by multimodal elements (eg, images, voice, and physiological data) to meet older users’ needs for visual preferences, operational ease, and emotional support.

The applications of GenAI primarily concentrate on 4 core domains—neurocognitive health, mental health, comprehensive health management, and specific disease interventions (Table 7). Comprehensive health management (n=12) covers a broad foundational range, including health knowledge acquisition, physical activity, and fall prevention. These applications use GenAI primarily for information dissemination and behavioral reminders. Neurocognitive-related health issues (n=8) dominate the landscape, encompassing subdomains, such as Alzheimer disease and related dementias, memory, and speech disorders. These applications leverage GenAI’s multimodal and personalized capabilities to address age-associated cognitive decline through tailored interventions, emphasizing early prevention and functional maintenance. Mental health support (n=6) represents another high-frequency area, spanning depression, stress, and general well-being. GenAI excels here by simulating empathetic dialogue (affective computing) and delivering standardized psychological protocols via multimodal formats, often forming a “cognitive-emotional” synergistic intervention model. Specific disease interventions (n=2; eg, management of patients with cancer and medication management) remain rare, likely due to the complexity of disease mechanisms and stringent data privacy requirements. This distribution reflects a strategic focus on addressing urgent neurocognitive and mental health needs while promoting broad-based health maintenance.

The research primarily targets 3 key cohorts. First, older adults with normal cognitive function constitute the main subjects for evaluating GenAI’s generalizability in basic health management and technology acceptance testing. Second, chronic disease populations and older adult patients form a critical cohort, focusing on GenAI’s role in addressing complex health demands like chronic disease management, postoperative rehabilitation, or acute condition prevention, which necessitates enhanced personalization and multimodal data integration. Third, mental health-focused populations are key subjects in specialized demand scenarios, assessing GenAI’s efficacy in emotional support and psychological intervention. The gradual inclusion of ethnic minority groups and populations with unique social attributes (eg, African American and Black adults) reflects growing attention to health equity, aiming to investigate GenAI’s adaptability and cultural relevance in diverse contexts. A small subset of studies also targets special health status populations, such as individuals with borderline cognitive impairment or those facing significant barriers to technology adoption.

This study, based on a systematic analysis of 28 articles [30-57], reveals the core characteristics and findings of GenAI applications in older adults’ health. Temporally, the research exhibits a growth trend over the past 2 years, reflecting intensifying global aging challenges and the rising maturity of GenAI technology. Geographically, research output is concentrated, with the United States and China leading the field by leveraging their advanced AI foundations and proactive smart older adults’ care policies. Methodologically, quantitative research is the dominant approach, supplemented by qualitative and mixed methods designs. The technology focus is on general GenAI systems, which form the core, expanding to specialized multimodal content generation and integrated applications complemented by interaction tools like voice assistants and visual question-answering models. Multimodal inputs (voice, images, and physiological parameters) and outputs (music, video, and gamified text) are increasingly used for personalized health services, demonstrating a technological shift from single-text to composite interactions. Health applications primarily cover neurocognitive health, mental health, and comprehensive health management, with the target populations ranging from ordinary older adults with normal cognition to those with chronic diseases and specific mental health concerns. The inclusion of ethnic minorities reflects growing attention to health equity and marginalized group needs.

GenAI demonstrates particular significance across 3 functional dimensions—cognitive preservation, mental well-being, and health management efficiency. In the cognitive domain, interventions like MusicTGA-HR and GenAI-based image recall therapy effectively improve frontal lobe function and memory performance, validating GenAI’s potential in maintaining cognitive reserve and delaying decline through personalized adaptation [30,43,46]. For mental health support, GenAI, through affective computing and empathetic dialogue, serves as a critical tool for alleviating loneliness and psychological stress, with general-purpose AI chatbots fostering self-disclosure and simulating the emotional warmth of human companionship [39,57]. Complementary AI-generated content, such as horticultural therapy integration, further expands psychological intervention scenarios, significantly reducing depressive symptoms [33,59]. Furthermore, in health management, LLM-based chatbots enhance accessibility by transforming complex medical information into simple, intuitive content, while integrating Internet of Things (IoT) sensor data for real-time tracking of critical health indicators and personalized interventions in chronic disease management [37,43,46]. Collectively, these applications establish a robust support system, advancing health equity by offering high-quality services to underserved groups and optimizing health care resource allocation by automating information delivery and monitoring [60].

The application of GenAI yields significant benefits in maintaining cognitive function, supporting mental health, and promoting health management autonomy, alongside facing adoption challenges [7]. Key benefits include personalized cognitive training (eg, text-to-music interventions) and psychological intervention via empathetic dialogue, which alleviates loneliness and fosters long-term self-reflection, leading to a “cognitive activation effect” [30,34,42,61]. GenAI also enhances self-efficacy by improving information accessibility and operational convenience, simplifying complex medical knowledge, and using real-time IoT data for timely health alerts [44,54]. GenAI also enhances self-efficacy by improving information accessibility and operational convenience, simplifying complex medical knowledge, and using real-time IoT data for timely health alerts [62,63]. Moreover, suboptimal system responsiveness and a lack of affective computing modules often trigger perceptions of machine coldness, leading to doubts about the reliability of AI-generated content [47,64]. Digital access barriers (small fonts and complex workflows) and concerns over privacy, algorithmic bias, and potential errors further deter adoption [65,66]. GenAI’s long-term potential requires robust policy and ethical frameworks, as well as advancements in voice interaction, multimodal fusion, and adaptive learning algorithms to overcome interaction bottlenecks and ensure service equity [36,67].

Future development trends for GenAI in older adults’ health will center on 4 key dimensions—technological deepening, interaction optimization, population adaptation, and ethical safeguards. Technologically, research must delve into the long-term effects and mechanisms of GenAI psychological interventions, expanding chatbot functionalities to integrate therapies like personalized art and mindfulness-based feedback for targeted emotional management [33]. Interaction optimization will drive the evolution from current text-based tools toward immersive multimodal experiences, combining voice, imagery, haptics, and olfaction, with designs prioritizing older adults’ physiological and psychological traits (eg, large-font interfaces and voice-input alternatives) to reduce cognitive load [68]. In terms of population adaptation, widespread adoption requires professional guidance, including training caregivers in digital literacy, and developing tiered tool systems (eg, basic 1-click operation for novices) to enhance accessibility for low technology-proficiency groups [69]. Ethically, advancements like edge-AI–powered wearables will enhance privacy by enabling localized data processing [44]. Crucially, transparent ethical frameworks must prioritize data privacy protection, implement user-controlled mechanisms, and use cross-disciplinary ethics review boards to assess risks like algorithmic bias and emotional dependency, ensuring technological innovation upholds dignity and rights [70,71].

This study, based on a systematic analysis of 28 articles [30-57] focused on GenAI applications in older adults’ health, presents several limitations that may affect the generalizability and external validity of the conclusions. First, the included studies exhibit insufficient sample representativeness. The research populations predominantly consisted of older adults with normal cognitive function, while studies targeting high-risk groups, such as those with cognitive impairments, multiple chronic disease comorbidities, or advanced age, were scarce. Furthermore, marginalized populations were critically underrepresented, meaning the conclusions primarily reflect the needs and experiences of “technology-adaptive” subgroups rather than the full diversity of the global older adult population. Second, the intentional exclusion of non–application-focused studies, such as reviews, nonacademic publications, and research emphasizing only user attitudes, perceptions, or intentions toward GenAI, limits the comprehensiveness of the review. Although this exclusion maintained a focus on practical implementation, it potentially reduced insights into the broader contexts that significantly shape GenAI adoption and effectiveness. These limitations necessitate cautious interpretation of the findings. Future research must prioritize expanding sample diversity to include high-risk and marginalized populations, enhancing long-term follow-up designs, optimizing real-world application testing, and deepening ethical and mechanistic explorations to advance the safe, effective, and equitable application of GenAI in older adults’ health.

This systematic scoping review analyzed 28 studies sourced from 11 electronic databases, summarizing the efficacy, current application status, and future potential of multimodal GenAI in promoting older adults’ health. The findings confirm that GenAI offers benefits across the care spectrum, particularly in maintaining cognitive function, supporting mental health, and enhancing comprehensive health management. Its efficacy is fundamentally driven by its core strengths—multimodal interaction capabilities and the ability to provide highly personalized content. However, most interventions are short-term, lacking robust long-term efficacy data and detailed mechanistic exploration. Studies overwhelmingly focus on low-risk, cognitively intact older adults, leaving a critical gap in coverage for high-risk and marginalized populations. Furthermore, persistent challenges related to age-friendly interaction design and unresolved ethical risks continue to hinder equitable adoption. Future research must prioritize inclusivity by expanding investigations into high-risk and underserved populations, optimizing multimodal interaction designs and age-friendly interfaces, and establishing long-term follow-up studies to validate long-term effects.