Population aging has intensified the need for accessible tools that support cognitive health and everyday functioning among older adults. Cognitive decline, including memory impairment and reduced attentional capacity, affects not only clinical populations but also older adults seeking to maintain independence and quality of life [1,2]. Digital interventions for older adults have become increasingly common, yet many continue to center on specific support functions, such as reminders, monitoring, and structured cognitive exercises, while tailoring them to individual needs and sustaining active, meaningful engagement remain recognized design challenges [3-5]. Emerging advances in AI offer new opportunities to address these challenges; in particular, the integration of machine learning with personal data allows systems to move beyond generic content delivery and toward individualized, context-aware interaction [6,7]. However, evidence remains limited regarding AI-enabled systems that integrate personalized memory interaction, emotional-response analysis, and caregiver-oriented interpretation within a single app [4,6,7], and many existing apps in this domain focus on prediction or automation rather than interaction with personally meaningful content [8]. This leaves a gap for systems that actively engage users while remaining interpretable and easy to use, supporting ongoing cognitive engagement and providing meaningful feedback for both users and caregivers [4,9].

To address the gap, this work presents the development and evaluation of RecallLive, an AI-assisted mobile system designed to support cognitive engagement through structured interaction with personal memories. Within the app, personal photos are clustered into meaningful events using temporal and spatial metadata and are transformed into structured memory videos; during playback, the system analyzes facial expressions in real time through the smartphone camera to construct a continuous emotion timeline. These responses are then aggregated into interpretable summaries presented through a caregiver-facing dashboard, creating a feedback loop between memory exposure and emotional engagement.

Because the potential value of this integrated system depends not only on its technical operation but also on whether older adults perceive it as usable, engaging, and useful, a formative evaluation of its usability, engagement, perceived usefulness, and adoption potential is warranted. The Technology Acceptance Model (TAM) identifies perceived usefulness and ease of use as important determinants of behavioral intention [10], making it an appropriate framework for examining RecallLive’s adoption potential. Engagement was included as a complementary experiential construct, as RecallLive is intended to encourage meaningful interaction with personal memories in addition to the efficient completion of functional tasks.

Accordingly, this study aimed to provide a comprehensive understanding of how older adults experience AI-assisted cognitive engagement systems and whether such systems can support meaningful interaction in everyday contexts. It was guided by the following research question: Is an AI-assisted mobile app integrating metadata-based clustering, real-time facial expression analysis, and language-model summarization perceived as usable, engaging, and useful by older adults? In addressing this question, the study evaluates whether the RecallLive system is perceived as usable, engaging, and valuable when older adults are introduced to its core functions and when they directly interact with the app. We hypothesized that an AI-assisted system integrating visual memory interaction and emotional response analysis would foster high levels of usability, engagement, perceived usefulness, and intention to use among older adults. Based on these objectives, the following hypotheses are proposed:

To evaluate the system across both initial perceptions and direct use, we used a sequential 2-phase mixed methods design: phase 1 provided a large-scale, perception-based evaluation in which older adults viewed a standardized demonstration and completed a structured survey, and phase 2 examined direct interaction through a hands-on usability session followed by semistructured qualitative interviews. Detailed procedures for both phases are described in the Methods section.

A sequential mixed methods design was used to evaluate the RecallLive system. The first phase examined user perceptions following a controlled demonstration, while the second phase investigated direct interaction through hands-on use. This approach aligns with recent formative studies that combine quantitative and qualitative evidence to evaluate user perceptions and experiences [17,18]. Participants were older adult users and possessed varying levels of familiarity with mobile technologies. The evaluation was structured into 2 phases to better integrate large-scale perceptual data with in-depth experiential insights.

Phase 1 participants were required to meet these inclusion criteria: (1) age 65 years or older, (2) current residence in the United States, (3) ability to provide electronic informed consent, and (4) ability to view and hear the audiovisual demonstration and complete the online survey using an internet-connected device. Participants were recruited nationally through Connect (CloudResearch), an online research platform, allowing for broad access to a geographically diverse sample. The survey was voluntary and closed, accessible only to panel members who received the study invitation, and it was not publicly advertised. Each panel account could submit the survey only once, which prevented duplicate responses from the same participant; no cookie or IP address checks were separately applied. Geographic information beyond US residence was not collected; therefore, participants’ states and cities of residence were unavailable. Specific medical diagnoses and hearing, vision, or motor impairments were neither used as exclusion criteria nor collected as participant characteristics, and no cognitive screening or assessment was administered.

To ensure consistency in system exposure, all participants were presented with a structured demonstration video that introduced the key functionalities of RecallLive, including metadata-based photo clustering, memory video generation, real-time facial emotion analysis, and AI-generated summaries for caregiver interpretation. Following the demonstration, participants completed a structured questionnaire comprising 16 evaluation items and 7 sociodemographic items presented over 2 screens, with sociodemographic items on one page and all 16 evaluation items on a second page. The evaluation items assessed multiple dimensions of user perception, namely ease of use, engagement, design clarity, perceived usefulness, and intention to use or recommend the system, and the sociodemographic items collected age, gender, ethnicity, marital status, education, employment status, and total household income. The order of the 16 evaluation items was randomized for each participant, and no adaptive questioning was used. All items required a response before submission, and participants could navigate back to review and change their answers before submitting. The measures were selected to reflect both functional usability and experiential engagement, recognizing that the system is designed not only to be usable but also to foster meaningful interaction with personal memories. This phase provided a large-scale assessment of how the system is perceived prior to direct interaction, offering insight into initial acceptance and perceived value.

All 16 items across the 5 dimensions were rated on a 5-point Likert scale (1=strongly disagree; 5=strongly agree) and were adapted from prior studies: 4 items assessed ease of use and 3 assessed perceived usefulness, with 1 intention-to-use item adapted from Davis [10]; 4 items assessed engagement based on O’Brien et al [19]; 3 items assessed design and accessibility based on Gomez-Hernandez et al [9]; and 1 recommendation item was adapted from Stoyanov et al [20]. Composite scores were calculated by averaging the items within each dimension, with higher scores indicating more favorable perceptions and 3 representing a neutral response. The intention-to-use item (“I would be willing to use this app regularly”) and the recommendation item (“I would recommend this app to other older adults or families”) were averaged into a single 2-item intention or recommendation score, reflecting a broader adoption-oriented response.

Phase 2 participants were required to meet these inclusion criteria: (1) completion of phase 1, (2) ownership of a compatible Android smartphone, (3) ability to install and use the APK (Android Package Kit)-based app, and (4) willingness to complete a hands-on usability session and a semistructured interview. Following phase 1, an invitation describing these criteria was sent to all phase 1 respondents, and volunteer sampling within these predefined eligibility criteria was used [21]. All respondents who volunteered and met the criteria were enrolled. A target of approximately 10 participants was informed by empirical usability research indicating that 10 users can identify at least 80% of usability problems [22]. The recruitment, scheduling, hands-on sessions, and interviews were completed within 1 week after participants completed phase 1. No additional medical, cognitive status, or sensory exclusion criteria were applied, and no cognitive assessment was administered in this phase.

Although phase 1 relied on standardized video-based exposure, phase 2 allowed participants to engage with their own photo collections, thereby introducing personally meaningful content into the interaction process. During the approximately 30-minute session, participants interacted with the 4 interaction features: they generated memory videos through the system’s metadata-based photo clustering, viewed the resulting videos while the front-facing camera performed real-time facial expression analysis, completed the interactive memory quiz, and reviewed the engagement summary. This enabled the system to construct emotion timelines and produce engagement summaries based on actual user responses. Afterward, a semistructured interview lasting an average of 30 minutes was conducted via videoconference to capture detailed reflections on usability, navigation, emotional engagement, clarity of feedback, and overall experience.

The interview guide (Multimedia Appendix 1) was developed to align with the 5 dimensions measured in the phase 1 survey, with additional questions comparing hands-on use with the phase 1 demonstration and eliciting open-ended feedback. The guide was iteratively developed by the research team and reviewed by 2 external scholars with expertise in qualitative research, whose feedback informed revisions before the guide was finalized. Interviews were audio-recorded with participants’ consent; after transcript verification, the audio files were permanently deleted. These interviews provided insight into how users interpret and respond to the system when interacting with it directly [23].

Quantitative data collected in phase 1 were analyzed using descriptive statistics to summarize overall trends. Internal consistency was assessed using Cronbach α [24]. One-sample, 2-tailed t tests compared the ease-of-use and engagement composite means with the neutral midpoint of 3.0 to evaluate H1a and H1b. Simple ordinary least squares (OLS) linear regression was used to examine the association between perceived usefulness and the 2-item intention or recommendation score. All quantitative analyses were performed using Python (v.3.11; Python Software Foundation). Qualitative data from phase 2 were transcribed verbatim and analyzed using thematic analysis, following established procedures that emphasize iterative coding and theme development from qualitative data [25]. Transcripts were read repeatedly for familiarization and coded, and codes were grouped into candidate themes organized around the interview domains of usability, navigation, emotional engagement, clarity of feedback, and overall experience. Themes were then reviewed and refined against the full dataset to ensure they accurately represented participant responses. The integration of these 2 phases enabled a comprehensive evaluation of the RecallLive system by combining broader perception-based findings with detailed experiential insights derived from real-world interaction, with phase 2 themes compared with the phase 1 quantitative results during interpretation.

This study was reviewed and approved before data collection by the regional Scientific Review Committee of the International Science and Engineering Fair and the school-based Institutional Review Board of Hamilton Southeastern High School, Indiana, USA (approval date: February 2026) and was classified as minimal risk. No approval number was assigned; under this review process, approval was documented through signed and dated review forms rather than by an approval number. All participants were adults aged 65 years or older, and participants in both phase 1 and phase 2 completed an electronic informed consent after receiving information about the study purpose, procedures, potential risks, privacy protections, and their rights as research participants. Participation was voluntary, and participants could withdraw at any time without penalty. Phase 1 data collection was conducted on March 20, 2026, and phase 2 interviews were conducted from March 20 to March 26, 2026. No minors, medical procedures, diagnostic assessments, treatment, or medical records were involved. Survey responses were collected without names or directly identifying information. All research data were analyzed in deidentified form. During phase 2, facial expression analysis was performed on-device; raw facial images or videos from the front-facing camera were not stored or transmitted, and only aggregated engagement metrics were retained for analysis. Phase 2 interview recordings were permanently deleted after transcript verification. Participants received US $1.50 for participation in phase 1 and US $50 for participation in phase 2.

The results suggest that the RecallLive system was perceived as both usable and meaningful by older adults in this initial evaluation. Participants reported positive evaluations across usability, engagement, and usefulness, and regression analysis showed that perceived usefulness was strongly associated with stated intention to use or recommend the app. This association is consistent with prior research linking perceived usefulness with behavioral intention [10,26]. However, recent evidence suggests that older adults vary substantially in their readiness to adopt digital health technologies. One identified subgroup perceived these technologies as useful but reported low self-efficacy and low perceived ease of use, indicating that favorable perceptions of usefulness may coexist with a need for operational support [27].

The qualitative findings suggest that usefulness alone may not fully explain the initial user experience. Participants described emotional reactions while viewing memory videos, indicating that engagement appeared to be related to personal relevance and reflection. Participant accounts indicated that seeing their own photos in sequence helped them recall past events and maintain attention longer than expected. This highlights an important point for system design, in that apps for older adults should not focus only on efficiency or simplicity, as emotional relevance may strengthen immediate engagement and willingness to use the system. Prior studies have also noted that clear structure and reduced cognitive effort are important, but meaningful content, guided feedback, and personally relevant interaction may further support engagement among older adults [9,28,29].

This role of personal relevance can be interpreted through research on autobiographical memory. Autobiographical memories serve self-related and social functions by connecting past experiences to a continuing sense of identity and to relationships with others [30]. Research on digital storytelling similarly suggests that personal media can support memory, affect, self-expression, and interpersonal connection, with many interventions involving active or co-created engagement with the material [31]. RecallLive extends this principle by automatically organizing personal photographs into memory videos and linking them with a quiz and engagement summary. Participants valued the reduced effort involved, while the sequence extended the experience beyond passive viewing by prompting responses to the presented memories. Thus, personalization in RecallLive derives not only from the use of personal photographs but also from their organization into a sequence of recall, response, and feedback.

Building on this personalized interaction, the RecallLive system contributes to the emerging field of cognitive apps for older adults by combining 3 components into one continuous process: memory organization, emotion detection, and summary generation. The integration of these components enables the system to summarize response patterns occurring during memory interaction rather than merely presenting personalized content. Related computer vision research has examined facial expressivity alongside validated measures of apathy in people with neurocognitive disorders [32]. RecallLive extends this line of work by applying facial-response analysis to personalized memory interaction, with the aim of identifying contextual patterns of engagement that may inform caregiver understanding and subsequent support. Interpretation should nevertheless remain cautious because facial movements do not always correspond directly to internal emotional states; the system’s outputs are therefore best understood as context-dependent indicators of engagement [33,34]. By focusing on patterns across time rather than single observations, the system may reduce the influence of isolated classifications and provide more stable contextual summaries.

The caregiver-facing summary extends this use of response patterns by creating a potential bridge between personal reminiscence and subsequent caregiver support. Technology-supported reminiscence may be particularly valuable when personal media provide a focal point for conversation and help others better understand the experiences of the person receiving support [35,36]. From this perspective, the summary may direct attention to memory content associated with stronger or more sustained engagement. Caregivers could use such information to prompt follow-up conversation or shared reminiscence.

The practical implementation of these functions also requires attention to privacy and ease of initial use. Privacy was therefore embedded in the system design, with all facial analyses performed on-device and no raw images stored or transmitted. This approach may reduce concerns related to surveillance and data misuse, which are commonly raised in discussions of facial recognition technologies [37]. Participants generally found the system easy to use, but some reported brief uncertainty during the first interaction. This suggests that even simple systems may benefit from short onboarding guidance, especially for users who are less familiar with mobile apps. Previous usability research supports the importance of clear instructions, simple layouts, low-burden navigation, and accessible content delivery for older adults [38-40]. Overall, the findings indicate that perceived usefulness was associated with stated intention to use, while positive engagement and usability evaluations characterized participants’ initial interaction with the system. These findings warrant further evaluation under longer-term and real-world conditions.

This study has several limitations. First, although phase 1 included a large sample, participants evaluated the system through a demonstration video. This may not fully reflect real interaction. Phase 2 partially addressed this by including hands-on use, but the smaller sample size limits generalization. The phase 2 sample comprised 10 of the 202 phase 1 participants. Eligibility required ownership of a compatible Android smartphone, the ability to install the APK-based app, and willingness to complete a hands-on session and interview within a 1-week window. These requirements likely restricted the pool of potential volunteers. In addition, since phase 1 participants were recruited online and phase 2 participants were drawn from this sample, participation required at least a minimum level of digital literacy and internet access. This may have introduced selection bias and may limit the generalizability of the findings to older adults with lower digital literacy or limited internet access. Second, the evaluation was conducted over a short period and did not capture long-term engagement. Future research should examine sustained interaction over time, particularly in relation to cognitive outcomes. Given the increasing prevalence of cognitive decline [1,2], longitudinal studies are necessary to assess real-world impact. Third, the emotion recognition component has known limitations; namely, facial expression data provide partial insight and should not be treated as a complete representation of emotional state [33]. Future systems may improve accuracy by incorporating additional signals such as voice or interaction patterns. Finally, future research should explore broader app contexts, including integration with caregiver systems and clinical environments. Advances in data-to-text generation may further improve interpretability and usability of system outputs [41,42].

Overall, the results showed high levels of engagement and perceived usefulness, with the latter strongly associated with stated intention to use the system. By combining structured memory interaction with on-device emotional analysis, RecallLive appears to offer a usable and meaningful approach to supporting cognitive engagement among older adults, with potential for real-world application. Further evaluation through longer-term studies and clinically integrated settings is warranted.