The global prevalence of dementia is rising, with recent estimates indicating that approximately 50 million people currently live with dementia. It is estimated that the number will increase to 152 million by 2050 [1]. Dementia has a significant impact on personal, family, and economic levels, costing the world about 1.3 trillion dollars [2]. Prevention is essential for dementia management. Among nonpharmaceutical interventions, digital cognitive training has become more popular because prevention is essential for dementia management. Among nonpharmaceutical interventions, digital cognitive training has gained increasing popularity as long-term clinical studies demonstrate that older adults who engage in regular digital cognitive training could delay dementia onset by preserving memory, attention, calculation, language, visuospatial, and executive functions [3-7]. This preservation helps build cognitive reserve that protects against age-related cognitive decline [8,9]. In practice, traditional cognitive training programs are constrained by several limitations. These include (1) limited accessibility for older adults who have mobility issues, (2) heavy reliance on social workers, and (3) substantial financial costs, which are incurred by individuals through personal expenses for participation and by society through health care and social services for people living with dementia [6,10,11].

Many studies suggest that digital cognitive training programs with remote access can overcome these barriers [3-7]. These programs demonstrate greater scalability since they can reach a larger number of users simultaneously without requiring additional resources or staff. For example, while traditional cognitive training requires more therapists or social workers for each new group of users, digital cognitive training programs can be used by many more users with minimal additional cost. This efficiency in scaling makes them more cost-effective, creating a more sustainable and widely accessible option [12-14].

However, current digital training programs have limited accessibility from home mainly because of inadequate consideration of older adult users’ digital literacy [15,16]. Low digital proficiency among older adults restricts their access to digital training programs [17]. Even older users with the ability to use digital training programs at home and with the necessary technological infrastructure often struggle with prolonged engagement [18]. Older adults face many challenges, including physical limitations, difficulty processing information, and digital service interfaces that are not easy to use [19]. These barriers worsen clinical results as the disease progresses, increase medical costs, and impose a greater patient burden [20].

Several solutions address these challenges, including simplified user interfaces that reduce cognitive load, gamification elements designed to promote user engagement, personalized digital literacy support, sending push notifications, using chatbots, and using tools to help people connect with others. These features (1) enhance usability, (2) reduce cognitive demand and screen fatigue, and (3) ultimately improve compliance and clinical outcomes [21]. Among these solutions, chatbots and group-based interventions have shown relatively strong potential for enhancing user engagement and effectiveness in digital cognitive training programs demonstrated by numerous studies [21-23].

Conversational agents, also known as chatbots, have the potential to resolve these barriers [24]. A substantial number of studies have shown that conversational agents used for health care interventions elicit moderate to high levels of participant satisfaction and acceptance, largely due to social support features [25-29]. These agents show promise to advance patient-centered care by supporting users in managing their own health [24,25]. Additionally, group-based interventions enhance program engagement by fostering interaction among members [27]. It is widely accepted that conversational agents provide effective social support when integrated with group interventions [28,29]. Two recent studies have validated the effectiveness of a fully automated mobile conversational agent in providing social support [28,29]. In health care settings, the Shared Medical Appointments model, which integrates group-based interventions, continues to gain wider acceptance as a way to improve patient engagement in treatment programs [30-33].

Under this circumstance, we propose Care & Cure, a chatbot-based service. The Care & Cure service offers 15 different cognitive training programs through Sammy Talk, along with a group chat feature called Our Town, where participants can interact with one another. This group interaction is intended to improve user engagement with Sammy Talk. At the core of this service is Sammy, a software robot that delivers personalized cognitive training and fosters social support through conversational interactions. By integrating these features, Care & Cure aims to meet both cognitive training and social needs of older adults.

The primary objective of this study is to evaluate whether the Care & Cure program enhances cognitive function in older adults. The secondary objective is 2-fold: (1) to assess how social support impacts cognitive function and (2) to determine whether increased engagement with Our Town encourages greater usage of Sammy Talk and interaction with the artificial intelligence (AI) agent. To achieve these objectives, we analyzed the utilization patterns of both Our Town and Sammy Talk to assess their effects on cognitive function and social support. As an additional evaluation beyond the primary focus on cognitive functions, we examined the effect of social support functionality embedded in both Sammy Talk and Our Town on emotional well-being, specifically depressive symptoms, which emerged as a complementary benefit to the cognitive improvements.

This was a single-arm, prospective, exploratory study. The protocol and all related materials were approved by the Yonsei University Institutional Review Board (IRB; 7001988‐202401-HR-1397-06) prior to the study’s initiation.

We recruited participants from Suncheon, a midsized city in South Jeolla Province, South Korea, with a population of approximately 280,000. The city’s older adult population reflects South Korea’s aging demographics, with about 20% of residents aged 65 years or older [34]. Suncheon is recognized for facilities that support senior citizens through welfare centers, nursing homes, and silver towns that provide housing and health care services for older adults [35] (see Section 1 in Multimedia Appendix 1 for more details).

Suncheon Nonghyup, a key institution in the city, provides substantial support to the older adult community. Its well-established network facilitated the recruitment process (see Multimedia Appendix 1, Section II for more details). The central office of the farmers’ cooperatives distributed formal recruitment documents to local branches. Through this network, we recruited participants who met the inclusion criteria: adults aged 55 and 75 years and proficient in using a smartphone. All participants provided written informed consent prior to participating in the study.

The inclusion criteria were adults (1) aged 55-75 years who could read and write Korean (for more rationale, see Multimedia Appendix 1, Section III) and (2) owned a smartphone with text typing proficiency. We excluded adults who (1) lacked formal education or were illiterate, (2) had severe visual or auditory impairments, and (3) experienced difficulty using a smartphone.

Prior to recruitment, we evaluated these criteria through a preliminary screening survey. Participants provided demographic information, including age, educational background, and literacy levels. They also demonstrated smartphone proficiency by performing basic tasks such as navigating the interface and sending messages. Additionally, the survey assessed visual and auditory impairments. This screening process identified participants who met all study requirements.

Our intervention delivered cognitive enhancement games specifically designed by neurologists through 2 key components on the KakaoTalk platform: Sammy Talk, a conversational agent providing cognitive games, and Our Town, a group chat feature. We chose KakaoTalk because more than 70% of South Korea’s senior population use it [36]. The Care & Cure program, designed to prevent dementia, integrates both components into this widely used messaging platform.

The Sammy Talk chatbot functions as an autonomous digital facilitator, delivering morning greetings and encouraging cognitive training activities. The chatbot sends automated prompts 3 times a day (8 AM, 1 PM, and 7 PM) in Our Town to promote user engagement with digital cognitive games and ensure task completion. To complement Sammy Talk’s limited autonomous functions, a human facilitator motivates less active members, monitors individual participation, and provides personalized feedback with external rewards. The facilitator tracks participant messages in Our Town and monitors game performance through an administrative website.

For rewards, the group with the highest weekly participation rate receives gift cards worth KRW 5000 (approximately US $3.70, based on the exchange rate at the time of the study) for each participant and a group certificate. Additionally, when all group members complete their required 4 daily games, they receive restaurant gift cards worth KRW 10,000 per participant (approximately US $7.40, based on the exchange rate at the time of the study) for offline meetups. This integrated approach, combining an autonomous chatbot and human facilitation, fosters an environment that actively encourages engagement and consistent participation in cognitive training.

The interactions in Our Town exhibit straightforward structures with moderate turn-taking patterns. A daily interaction usually begins when the chatbot sends an automated morning greeting, which prompts individual responses from group members. These responses emerge sequentially, as members choose either personalized greetings or emoji-based communication. The exchanges primarily focus on 3 elements: mutual well-being inquiries, weather-related small talk, and positive message sharing. The structural simplicity of these interactions, which consist of brief greetings and basic phrases, ensures comfortable participation. Members demonstrate varied communication preferences: some address each participant individually, while others prefer concise responses or emojis. Each interaction typically consists of 5-10 turns, which reflects varying engagement levels. The uncomplicated, conversational nature of Our Town creates an accessible environment that accommodates diverse schedules and active communication styles.

A 2-day educational session took place at the Suncheon Nonghyup headquarters to establish baseline training. During this session, participants received an introduction to both the Sammy Talk program and Our Town group chat, alongside a comprehensive guidebook. The research team provided telephone instructions for those unable to attend in person. Subsequently, all participants received a text message with the program access link. The team installed Sammy Talk and Our Town directly onto the smartphones of session attendees.

The evaluation included 4 areas: cognitive function, emotional well-being, participant engagement, and application usage. We used 2 cognitive assessment scales: the Korean-Mini Mental State Examination—Version 2 (K-MMSE-2) [53] (see Multimedia Appendix 1, Section V for more details). These assessments took place at baseline and within 7 days after the 12-week intervention. The K-MMSE-2 uses a 30-point scale, where scores of 24 or above indicate normal cognition, 18‐23 suggest mild cognitive impairment, and below 17 indicate severe cognitive impairment. For emotional function assessment as a supplementary outcome, we used the Short Form Geriatric Depression Scale—Korean version (SGDS-K) [54] and the Medical Outcomes Study–Social Support Survey (MOS-SSS) [55] (see Multimedia Appendix 1, Section V for more details). The SGDS-K uses a 15-point scale, where scores of 5 or below indicate normal mood, scores of 6‐9 suggest moderate depression, and scores of 10 or above indicate severe depression. The MOS-SSS ranges from 19 to 95 points, with higher scores indicating stronger perceived social support. Changes in depression symptoms and social support served as secondary outcomes, with assessments occurring pre- and postintervention.

In addition to these, we measured participation willingness using the Twente Engagement with eHealth Technologies Scale (TWEETS) [55] (see Multimedia Appendix 1, Section V for more details), which uses a 9-point scale where higher scores indicate greater engagement willingness. The analysis on app engagement relied on log data acquired from the administrative websites of both Sammy Talk and Our Town (see Multimedia Appendix 1, Section V for more details). For Sammy Talk, engagement assessment included play frequency and duration across daily, weekly, and monthly intervals. For Our Town, engagement metrics comprised the number of chat exchanges and user-agent interaction duration across identical time intervals. The overall usage metrics included total Sammy Talk hours and Our Town chat bubble counts throughout the intervention period (Table 1).

We arranged in-person meetings for pre- and postsurveys based on practical and methodological considerations: participant age preferences, assessment administration requirements, facility availability, and geographical accessibility. The target population, aged 55‐75 years, typically preferred face-to-face surveys over online formats. The Suncheon Nonghyup branch offered appropriate facilities, including halls and meeting rooms, for administering the K-MMSE-2, which required direct facilitator-participant interaction. The branch’s proximity to participants’ residences enhanced accessibility. Facilitators conducted presurveys during Care & Cure program introduction sessions, while postsurveys coincided with reward collection visits, ensuring effective data gathering.

We conducted the human subjects experiment for 12 weeks (Figure 2A). During this period, 133 participants (1) engaged in daily message exchanges through Our Town and (2) completed 4 daily sessions of Sammy Talk. Each day at 8 AM, participants received their daily activity instructions. Weekday activities consisted of 4 preselected games, whereas during weekends and holidays, the system delivered dementia prevention information and notifications at 8 AM. We conducted the 12-week intervention in a structured sequence of activities. The flowchart illustrates how customized cognitive training in Sammy Talk and group-based interactions in Our Town progressed in parallel and reinforced each other (Figure 2B). Participants maintained Sammy Talk engagement throughout the 12-week period. All participants completed these activities in their homes (Figure 2C).

We analyzed data from 133 participants using SPSS (version 28.0; IBM SPSS Inc). All participants completed the 12-week intervention, which included cognitive training games on Sammy Talk and the group chat platform Our Town. The study divided participants into 2 groups based on their engagement levels: active users (n=66) and nonactive users (n=67). Total usage included both Sammy Talk (which counted the number of games played) and Our Town (which counted the number of group chat posts). The median value of 65 sessions served as the dividing point: participants who recorded more than 65 sessions were classified as the active group, and those who recorded 65 sessions or fewer were classified as the nonactive group. We applied a “50% of the time” threshold, which ensured an even distribution between the 2 groups. This approach aligns with the methodologies used by Horwitz et al [56], who categorized participants into top and bottom 50% engagement groups in digital mental health interventions. This approach was also supported by Yardley et al [57] and Bucci et al [58], who identified that higher engagement correlated with improved clinical outcomes.

For comprehensive understanding of the impact of the training regarding the K-MMSE-2 scores, we additionally conducted a sensitivity analysis that categorized participants into lower (n=66) and higher (n=67) K-MMSE-2 groups, using baseline K-MMSE-2 median and SD. Participants with baseline K-MMSE-2 scores ≤27, which represented the median score, were categorized as the lower K-MMSE-2 group, while those who scored >27 were categorized as the higher K-MMSE-2 group. We also applied an alternative classification: participants whose baseline scores fell more than 1 SD below the mean were classified as having lower K-MMSE-2 status.

The analysis consisted of several components. Descriptive statistical analyses provided demographic information. Paired-sample t tests identified significant differences in cognitive function, emotional status, and participation between pre- and postintervention periods. Additional paired t tests compared outcomes between lower- and higher K-MMSE-2 groups and between high- and low engagement groups. Path analysis, through simple regression analyses, examined how group interventions affected Sammy Talk engagement, social support improvement, and cognitive function. Repeated-measures ANOVA assessed the mean and variability of daily Sammy Talk and Our Town usage across 3 months. Descriptive statistics supplemented this assessment for detailed data interpretation.

This study was approved by the Yonsei University IRB (IRB no. 7001988‐202401-HR-1397-06). All participants provided written informed consent prior to participation. The collected data were fully anonymized and securely stored, with access restricted to the research team. Participants received small incentives, such as gift cards worth KRW 5000 (approximately US $3.70) and restaurant vouchers worth KRW 10,000 (approximately US $7.40), based on the exchange rate at the time of the study, as compensation for their participation.

Initially, we recruited 187 participants for the study. Of these, 44 participants withdrew during the intervention period from the primary measure because they did not complete the survey or due to personal circumstances, leaving 143 participants. After the 12-week intervention, 10 more participants withdrew for similar reasons before completing the second measure, resulting in 133 participants available for final statistical analysis. Participants who did not complete the postintervention survey (n=54) were excluded from the final analysis. No statistical adjustments were applied for missing data from participants who withdrew before completion. Therefore, all statistical analyses were conducted using complete cases only. Figure 3 illustrates the flow of participant recruitment, retention, and exclusion throughout the study.

These 133 participants ranged in age from 55 to 75 years (mean 64.76, SD 6.76 years). Based on previous studies reporting medium effect sizes (Hedges g=0.5) and using G*Power (α=.05, power=0.80), we estimated that a minimum sample size of 128 would be required. This estimate is also consistent with the numbers of participants actually analyzed in similar studies involving older adults (100, 103, 145, 132, and 180 participants, respectively) [59-63]. To account for a potential dropout rate of approximately 30%, we aimed to recruit at least 180 participants. Ultimately, 187 were enrolled (Table 2).

The demographic profile showed that out of 133 participants, 69 (51.9%) were female, 119 (89.5%) had been married, and 51 (38.3%) held elementary school diplomas. Among them, out of 133 participants, 91 (68.4%) reported living with a spouse (Table 3). No missing data were observed for the baseline demographic variables, including cognitive function score (K-MMSE-2), depression score (SGDS-K), social support measure (MOS-SSS), or participation measure (TWEETS), among the participants who were included in the final analysis. Participants who had incomplete outcome data (n=54) were excluded from the analysis.

The Care & Cure program demonstrated effectiveness in improving cognitive function across all 133 participants, with particularly strong enhancement among individuals with lower K-MMSE-2 scores. This aligns with the program’s evidence-based development, which incorporated clinical expertise from neurologists at Mokdong Ewha Womans University Hospital.

The Sammy Talk content followed validated clinical frameworks across multiple cognitive domains. The calculation games were derived from the WRAT-3 Arithmetic subset and Digit Symbol Substitution Test [41,42]. Visuospatial tasks incorporated the Mental Rotation Test [43], whereas attention-focused games used crossword puzzles and the Montreal Cognitive Assessment letter—a tapping test [44,45]. Executive function games were adapted from the University of California, San Diego Performance-Based Skills Assessment (UPSA-3) and the Trail Making Test [46,47]. Memory tasks integrated the Korean Naming Test, Chinese idioms, and Rey-Osterrieth Complex Figure Test [48-51]. Language games incorporated the Montreal Cognitive Assessment initial letter fluency test and Scrambled Sentences Test [44,52]. However, the specific cognitive impact of individual game components remains to be clarified in future research.

Participants with lower baseline K-MMSE-2 scores (hereafter referred to as the lower group) showed larger effect sizes, as they had greater potential for measurable improvement. By contrast, those with higher baseline scores (hereafter referred to as the higher group) showed little change, having already approached their cognitive performance ceiling. This difference reflects a well-known ceiling effect among cognitively normal participants. Previous studies examining short-term interventions have consistently reported that people with cognitively normal conditions achieve only limited gains, whereas those with cognitive impairment show greater improvement [64-66]. Moreover, participants in the lower group faced a higher risk of decline and therefore benefited more from the clinically validated content designed by neurologists to target multiple cognitive domains, including calculation, attention, language, memory, visuospatial ability, and executive function. This finding is consistent with previous research using validated clinical tasks, which confirmed that participants in the lower group achieved greater improvements than those in the higher group [64,65].

Higher usage frequency of Care & Cure corresponded with enhanced cognitive function. Our Town appeared to moderate this effect by increasing Sammy Talk engagement. Participants who engaged more often in Our Town engagement dedicated more time to Sammy Talk and achieved superior cognitive function scores. The group chat format likely enhanced engagement through shared performance goals, achievement reports, and peer encouragement. These findings align with previous research on group chat–based cognitive enhancement programs [31,32].

Nonetheless, Our Town did not directly enhance cognitive function; rather, its impact emerged through social support mechanisms. The integration of Our Town and Sammy Talk (a software robot) in the Care & Cure program enhanced social support, which in turn improved cognitive function. While Our Town did not bring about direct cognitive benefits, it served as an indirect motivational means by sending consistent reminder messages to participants. Participants exchanged greetings and shared daily experiences in Our Town while learning practical knowledge and conversation-related vocabulary through Sammy Talk. This encouraged more frequent communication both online and offline. These repeated interactions in turn created a social support network through daily greetings, experience sharing, and conversational practice. The AI chatbot provided additional support by offering affective encouragement and fostering positive exchanges, leading participants to report higher social support scores. This finding aligns well with previous studies demonstrating that AI chatbots deliver social support and reduce depressive symptoms among older adults [67,68]. Social support and emotional stability function as preventive factors that preserve cognitive function by reducing loneliness and chronic stress, both recognized contributors to memory decline and executive dysfunction [67-71]. Conversations and social interactions demand memory recall, language use, and problem-solving, repeatedly stimulating cognitive processes and strengthening cognitive reserve [72-74].

On a secondary issue, the Care & Cure program effectively reduced emotional distress, with participants experiencing emotional changes through enhanced social support. Program engagement, measured through Our Town chat interactions, showed increasing participant responses to the chatbot-based software agent, reaching 3.8 times the initial level by month 3 (Figure 6). This increased engagement suggested more frequent conversations with both the software agent and other participants, potentially facilitating positive social exchanges and providing emotional, affectionate, and informational support. However, chat interaction frequency alone provides limited insight into intervention adherence. Future research should examine more precise adherence measurements. The study suggests that increased social support influenced participants’ depressive symptoms [75-78].

Nearly 3 out of 10 participants (28.8%) dropped out of the study, with education emerging as the primary differentiating factor. Education, rather than age or sex, accounted for this difference, as people who completed the intervention had significantly more years of education than those who dropped out (t₁₃₁=2.43, P=.016). These results show that education level, which reflects accumulated social and cognitive resources, was associated with dropout, whereas age and sex were not. Participants with limited education and digital literacy often struggled to use the digital cognitive training program. Lower educational attainment, which often limits digital literacy, reduced participants’ ability to use messenger applications and type messages, preventing them from joining group chats or completing training tasks [18,20,21]. Those who lacked primary school education and could not perform basic arithmetic operations were unable to complete calculations tasks and eventually dropped out. Participants with very low education levels may therefore be more likely to drop out of digital cognitive training programs [18,20,21]. Although not statistically significant, males were likely to drop out more frequently than females, suggesting that future studies should consider strategies to improve male retention (see Multimedia Appendix 1, Section VI for more details).

Despite the significant improvement in participants’ cognitive capacity through the use of our proposal in general, the research has some limitations to address. First, the single-arm design without a control group reduces confidence in attributing changes to the intervention. Without a comparison group, it would be very difficult to establish that observed improvements resulted solely from the intervention, as changes might be able to reflect natural progression, practice effects from repeated testing, or placebo effects. In addition, it would be challenging to determine whether improvements are clinically meaningful compared with standard care or alternative interventions. Second, participant interaction with the software robot in Our Town (the group chat component) was lower than anticipated, with its primary effect manifesting through increased usage of Sammy Talk. Third, the sample was drawn from a single midsize Korean city with relatively homogeneous demographic characteristics, limiting the generalizability of findings to other regions or populations. Fourth, the program’s reliance on KakaoTalk as the delivery platform might restrict applicability beyond the Korean context. These methodological considerations highlight areas for future investigation, including randomized controlled trials with appropriate control groups, multisite studies across diverse populations, and head-to-head comparisons with standard assessment tools to establish clinical superiority.

The Care & Cure program demonstrated effectiveness for cognitive enhancement, with more frequent participation corresponding to greater improvement. The program showed particular efficacy among cognitively impaired participants. Additional benefits included reduced depressive symptoms, enhanced social support, and increased participation willingness.

Our hypothesis proposed that increased Our Town usage would enhance Sammy Talk engagement, subsequently improving cognitive function. The data confirmed that higher Our Town usage directly increased Sammy Talk participation. However, increased Sammy Talk usage alone did not yield statistically significant cognitive improvements. When examining social support as a mediator between Sammy Talk usage and cognitive enhancement, results indicated that social support directly influenced cognitive improvement. This relationship requires further investigation to understand underlying mechanisms.

While the top 50% of program participants showed greater K-MMSE-2 improvements, the causal relationship between Sammy Talk usage and cognitive enhancement lacked statistical significance. Future research should examine the relationship between Our Town and Sammy Talk usage patterns while addressing the methodological issues raised in this study. Controlled trials with comparison groups are necessary to establish causal effects. For example, a potential future study could compare an intervention group using both Sammy Talk and Our Town, with a control group attending a traditional face-to-face dementia prevention program. Researchers should also test strategies to increase participation in group chat functions, which serve as sources of social support [79-81]. Gamification approaches, such as small financial rewards, goal setting, or competition elements, may enhance motivation to engage [82,83]. Additionally, context-based reminders drawing on Ecological Momentary Interventions or Ecological Momentary Assessments could encourage more consistent use of both Sammy Talk and Our Town [84-86].

Next, future studies had better recruit participants from varied regions and sociodemographic groups to extend findings beyond a single Korean midsized city. For instance, researchers could target English-speaking populations from middle-class and low-class households or urban groups with higher digital literacy and educational attainment [87,88]. Moreover, researchers should conduct longitudinal follow-up studies that track participants for at least 1 year to examine whether cognitive benefits persist and to determine whether digital training combined with social support prevents long-term cognitive decline [89,90]. For example, participants could use the intervention for 1 year or more, with outcomes measured at regular intervals to evaluate intervention effectiveness over time. Finally, since this program relied on KakaoTalk as its delivery platform, which limits generalizability beyond Korea, future studies should better adapt the same model to other messaging services that provide chatbot Application Programming Interfaces, such as WeChat, WhatsApp, or Facebook Messenger, to test how the intervention performs across different cultural and technological contexts [91,92].

Finally, future works should also take into account education as a covariate, which requires stratified randomization and sensitivity analyses to ensure the robustness of findings. Researchers should also design tasks that assess cognition without relying on arithmetic, so that individuals without formal schooling can participate fully in the longitudinal intervention. Finally, incorporating competitive gamification elements, which may appeal particularly to males who often demonstrate stronger achievement motivation than females, could enhance digital program retention by using leaderboards or quizzes that provide modest rewards and recognize performance [93-95]. By pursuing these directions, future research would be able to clarify how digital cognitive training and social interaction interplay with each other, and identify practical strategies that strengthen engagement and broaden applicability across populations.