A single-arm, nonrandomized feasibility study with pre-post measurements was conducted to assess a chatbot intervention for informal caregivers. The study focused on feasibility, usability, and acceptability and was not designed to formally test intervention effectiveness [59]. Recruitment activities began in October 2023. Due to technical delays, participant-facing intervention delivery began in December 2023. After completing the baseline survey, participants entered the 12-week intervention on a rolling basis. Intervention delivery and follow-up data collection took place until the end of the study in March 2024.

The study was reported in accordance with the CONSORT-EHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile Health Applications and Online Telehealth; V1.6.1) checklist, adapted for a nonrandomized, single-arm feasibility study [60], and the TIDieR (Template for Intervention Description and Replication) checklist [61]. The completed checklists are provided in Checklist 1 and Checklist 2.

The intervention was developed and coordinated in Bremen, Germany, a socioeconomically and ethnically diverse city of approximately 700,000 inhabitants [62]. The feasibility study was open to informal caregivers residing across Germany; however, recruitment activities were concentrated in Bremen.

Eligible participants were adults (≥18 years) who provided unpaid, nonprofessional care to another adult, had access to a smartphone with WhatsApp installed, had sufficient German or Turkish literacy, and were able to provide informed consent. Individuals caring exclusively for their own minor children were not eligible.

Participants were recruited through local and regional organizations providing caregiving, migration-related, educational, self-help, health-promotion, and employment-related services. Recruitment partners included the Dementia Information and Coordination Center (Demenz Informations- und Koordinationsstelle), the Center for Migrant and Intercultural Studies (Zentrum für Migranten und Interkulturelle Studien), the Bremen Foundation of Nursing Homes and Residences, and the Bremen Adult Education Center, with additional support from the Bremen Chamber of Employees and trade unions.

Study information was disseminated through newsletters, mailing lists, social media, and in-person events. To enhance reach among caregivers of Turkish origin, community-based recruitment was conducted in settings such as mosques, cultural centers, and Turkish-speaking caregiver groups. Study materials were available in German and Turkish. Interested individuals accessed study information through a QR code or contacted the study team directly. At the end of the survey, participants received access information for the WhatsApp chatbot.

A target sample size of 100 participants was defined. This target was chosen pragmatically based on the planned recruitment period and expected recruitment capacity and was considered sufficient for a descriptive assessment of feasibility, usability, and acceptability. No formal power calculation was conducted because the study was designed to assess feasibility, usability, and acceptability [59].

A needs assessment informed by qualitative interviews with 27 informal caregivers from diverse migration and educational backgrounds identified priorities related to caregiving information, psychosocial support, stress management, navigation of the long-term care system, and culturally sensitive services. Detailed methods and findings will be reported separately.

Intervention development was further guided by a co-design approach [63]. Three co-design workshops—2 conducted in German and 1 in Turkish—and additional stakeholder consultations were held between July 2022 and September 2023. Informal caregivers, stakeholders, and researchers jointly refined the proposed chatbot content and delivery formats. A review of relevant literature on digital and psychosocial caregiver interventions complemented the co-design process.

Development was guided by 4 principles: supporting caregivers’ mental well-being, using existing digital infrastructures, incorporating evidence-informed content, and providing culturally adapted content in German and Turkish. Content was developed collaboratively by informal caregivers, stakeholders, and an interdisciplinary research team with expertise in public health, psychology, social sciences, and digital media design. An intersectionality-informed perspective was applied to consider how gender, migration history, language, and socioeconomic circumstances may shape caregiving experiences and access to support [40].

Substantially incomplete records, characterized by both missing age and gender information, were excluded. Records from respondents who did not meet the definition of informal adult caregiving were also excluded. Participant flow and availability of follow-up measures were summarized using the survey dataset. Chatbot initiation, initiation of scheduled message delivery, inactivity or nondelivery, and recorded opt-outs were summarized using operational project logs.

All quantitative analyses were descriptive. Continuous variables were summarized as means, SDs, and ranges, and categorical variables as counts and percentages. No missing values were imputed, and no between-group hypothesis tests were conducted.

SUS scores were summarized using the mean, SD, 95% CI, and range, together with the number and proportion of participants scoring at or above the descriptive benchmark of 68. Exploratory variation in SUS scores was described using subgroup-specific means and 95% CIs for categorical characteristics and Pearson correlation coefficients for continuous characteristics. Because some subgroups were small, the estimates were interpreted descriptively and not as evidence of group differences.

Study-specific acceptability items were analyzed using all available item-level responses and summarized using means, SDs, counts, and percentages as applicable. Item-specific denominators were reported because response availability differed across items.

Pre-post changes in health outcomes were calculated separately for each measure among participants with valid baseline and follow-up scores for that outcome. Baseline and follow-up means with corresponding standard deviations and within-participant mean changes with corresponding 95% CIs were reported. CIs were used to describe the precision of the estimates rather than to determine statistical significance. Quantitative analyses were conducted in Stata version 19.5 (StataCorp LLC). For methodological quality assurance, the authors used ChatGPT (GPT-5.6 Thinking; OpenAI) to audit, revise, and debug the Stata analysis code. Only privacy-sanitized code, the analytical specification, and aggregate execution logs were provided; the raw survey dataset was not uploaded. The authors reviewed and executed the final code locally in Stata 19.5 and retained full responsibility for the analyses. The prompts and audit documentation are provided in Multimedia Appendix 3.

Open-ended feedback was examined using a systematic inductive descriptive approach. One researcher fluent in German and Turkish reviewed all substantive responses in their original language, identified relevant statements, and developed preliminary category labels and definitions. Turkish-language responses were translated into German for secondary review. To enhance analytic trustworthiness, a second researcher reviewed all German-language responses, the German translations of Turkish-language responses, the category assignments, category definitions, and category summaries; the final classifications and descriptions were agreed upon through discussion. Researcher annotations and interpretations were excluded from the descriptive categorization. Paraphrased telephone field notes contributed to the categorization but were not used as illustrative quotations. The material was not independently double-coded. Further methodological details are provided in Multimedia Appendix 4.

The study was approved by the Ethics Committee of the Bremen Medical Association in July 2023 (application number 866). All participants received written study information and provided informed consent prior to participation. Participation was voluntary, and participants could withdraw at any time without consequences.

Survey data were collected through LimeSurvey, stored on secure institutional servers, and linked across assessment points using pseudonymized participant identifiers. Survey responses and health outcome data were not collected or stored through WhatsApp or Chatfuel. Participants were informed that Chatfuel and WhatsApp were used for intervention delivery, and data minimization principles were applied in accordance with the General Data Protection Regulation. Participation was free of charge, and no financial compensation or incentives were provided.

A total of 140 individuals entered the survey dataset. After exclusion of 30 (21%) substantially incomplete records, with age and gender information missing, and 8 (6%) additional records that did not meet the definition of informal adult caregiving, the baseline analysis sample comprised 102 participants.

According to project logs, 70 initiated the chatbot intervention after completing the baseline survey. Scheduled message delivery started for 67 participants; inactivity or nondelivery was documented for the remaining 3 participants. One participant opted out after message delivery had begun.

Of the 102 participants included in the baseline analysis sample, 49 provided at least one follow-up measure, including usability, acceptability, open-ended feedback, or a health outcome. Complete SUS data were available for 46 participants. Outcome-specific paired analyses included 34 to 36 participants with valid baseline and follow-up scores for the respective health outcomes. Participant flow is summarized in Figure 1.

Table 3 presents the baseline characteristics of the baseline analysis sample and the subset of participants who provided at least one follow-up measure. Percentages in Table 3 were calculated using the column totals and rounded to whole numbers; consequently, they may not sum to 100. Missing values and the available sample were reported where applicable. Participants were predominantly women and were born in Germany and most cared for a parent or a spouse or partner. High caregiver burden and elevated depressive symptoms were common at baseline.

Paired sample sizes varied by outcome (Table 5). Mean perceived stress, depressive symptoms, and loneliness were lower at follow-up, whereas mean caregiver burden was higher. All 95% CIs for the within-participant mean changes included zero, indicating inconclusive data regarding the direction of the changes. Given the feasibility design and limited sample sizes, the findings were interpreted descriptively and not as evidence of intervention effectiveness.

This study assessed the feasibility, usability, and acceptability of PflegeBot, a bilingual, rule-based WhatsApp chatbot for informal caregivers in Germany. Regarding feasibility, the baseline recruitment target was reached, although the number of participants decreased across chatbot initiation, message delivery, and follow-up, indicating challenges in intervention uptake and follow-up participation. Among participants providing follow-up data, usability was high, and acceptability was rated favorably, although satisfaction with the daily messages was lower than the other acceptability ratings. Reported use of the FAQ function was limited.

Pre-post analyses indicated lower mean scores for perceived stress, depressive symptoms, and loneliness at follow-up and a higher mean caregiver-burden score. However, all CIs for the within-participant changes included zero. Given the single-arm design, loss to follow-up, limited paired samples, and heterogeneity in caregiving circumstances, these findings should not be interpreted as evidence of effectiveness.

Open-ended feedback indicated that intervention fit depended on the correspondence between delivery, content, and participants’ caregiving circumstances.

The favorable usability and acceptability ratings are broadly consistent with previous evaluations of digital interventions for informal caregivers [45,49]. At the same time, limited follow-up participation and the low reported use of the FAQ function point to challenges in sustaining participation and making interactive features sufficiently visible [71].

Participants differed in how well standardized daily delivery fitted their available time and current information needs. Some valued regular messages as reminders, whereas others reported information overload, insufficient time, or limited relevance to their immediate caregiving situation. Similar variation has been reported for other digital caregiver interventions [39,72]. Messenger-based interventions may therefore benefit from adjustable delivery and modular pathways that preserve the prompting function of regular messages while allowing participants to select content relevant to their circumstances [41,46,49,73].

Exploratory variation in SUS scores was based on small subgroups and imprecise estimates and therefore does not support conclusions about differences by education, gender, or caregiving characteristics. Nevertheless, evidence suggests that education, digital literacy, and familiarity with digital services may influence perceived usability [71]. These factors should remain a consideration for future interventions.

Mean caregiver burden increased slightly, whereas the other exploratory outcome means decreased; a similar pattern was reported by Ghani et al [48]. These estimates were imprecise and may reflect changes in caregiving circumstances or measurement variability or an increased awareness and recognition of caregivers’ own burden fostered by participation in the intervention. However, these mechanisms were not assessed, and given the study’s single-arm, nonrandomized design, no beneficial or adverse intervention effect can be inferred.

PflegeBot was developed through a participatory process intended to incorporate caregivers’ experiences and language-related needs into its content and delivery—an approach that has already proven promising in other DPHIs [63,71,73]. The study did not establish equitable accessibility or acceptability across gender, intervention language, or social groups. Nevertheless, the findings indicate that language, digital experience, education, available time, and caregiving circumstances should be considered jointly rather than as isolated determinants [18,29,74].

Strengths of this study include the co-designed bilingual intervention, delivery through an established messaging platform, and evaluation under real-world conditions. Combining quantitative measures with open-ended follow-up feedback provided complementary descriptive information on acceptability, contextual fit, and intervention refinement, addressing gaps frequently identified in evaluations of digital caregiver interventions [41,49].

This study has several limitations. First, the single-arm feasibility design without a control group precludes conclusions about effectiveness. Second, the follow-up and paired-analysis samples were substantially smaller than the baseline sample. Because participants entered the intervention on a rolling basis, not all baseline participants had sufficient time to complete the full 12-week intervention and follow-up assessment before the study ended in March 2024. The reduction in sample size between baseline and follow-up therefore cannot be attributed solely to attrition, although selective follow-up participation cannot be excluded. In addition, more than one-fifth of the survey records were substantially incomplete and excluded from the baseline analysis. These records could not be adequately characterized or compared with the included sample, creating a further potential source of selection bias. Third, objective message- and content-level usage metrics were unavailable. Exposure to individual intervention components, content completion, and sustained engagement could therefore not be assessed. Fourth, the sample predominantly consisted of women, and the number of participants using the Turkish-language intervention was small, preventing reliable subgroup or intersectional analysis. Fifth, participants were recruited through diverse channels and self-selected into the study, while eligibility required access to a smartphone with WhatsApp. The sample may therefore have had greater digital familiarity, motivation, or readiness to use digital support than the broader population of informal caregivers. Sixth, the standardized health measures were administered using study-specific translations rather than previously validated German- and Turkish-language versions, and their semantic and psychometric equivalence was not formally evaluated. Finally, although the study was open to informal caregivers throughout Germany, recruitment activities were concentrated in Bremen, which may limit transferability to other settings.

The findings indicate that messenger-based chatbot interventions may complement existing caregiver support services when delivery and content can be adapted to caregivers’ changing circumstances and time constraints. Future development should therefore prioritize user control over intervention delivery and content selection and reduce barriers to navigation and interaction.

Future studies should use adequately powered, controlled designs and integrate objective engagement metrics alongside self-report measures. Mixed methods approaches remain important for contextualizing quantitative outcomes and for understanding engagement and nonengagement [47,71]. Evaluations should also examine differential reach, usability, follow-up participation, and outcomes across relevant social and caregiving contexts. In studies using rolling enrollment, the opportunity to complete the intervention and follow-up should be considered when defining follow-up denominators and reporting attrition.

Sustainable implementation of chatbot-based caregiver interventions will require clear organizational responsibility for content review, technical maintenance, user support, and data protection, particularly in light of ongoing challenges associated with implementing reimbursable digital care applications in Germany [75]. Dependence on commercial messaging platforms should also be considered an implementation risk requiring ongoing technical and organizational adaptation.

This feasibility study indicates that PflegeBot could be delivered to informal caregivers through WhatsApp, although challenges were observed across intervention uptake and follow-up participation. Among participants providing follow-up data, usability was high and acceptability was generally favorable. Further development should focus on adaptable delivery, tailoring, and accessible interaction. Controlled studies are needed to assess effectiveness, sustained engagement, and differential reach across social and caregiving contexts.