This systematic review aligns with the meta-aggregation principles outlined by Lockwood et al [12] (adapted for qualitative evidence synthesis), which we have applied to the domain of health education and behavior change interventions delivered via conversation-based digital tools, including chatbots. This approach was chosen to explore in-depth user and provider experiences through a specific focus on qualitative data, directly addressing the RQs on perceptions, theoretical frameworks, and adoption facilitators and barriers. We adhered to the guidelines of the Joanna Briggs Institute (JBI) [13] for critical appraisal and reported our methods in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flowchart and standard systematic review protocols. The review was registered with the Open Science Framework to ensure methodological transparency.

We searched PubMed, the Cochrane Library, and ScienceDirect from January 1, 2018, to October 1, 2023, for English- or German-language, peer-reviewed qualitative or mixed methods studies. A medical librarian assisted in developing the search strategy to ensure comprehensive coverage. Our search syntax included terms such as “chatbot,” “conversational agent,” “digital health interventions,” “qualitative,” and “mixed methods.” These terms were applied to titles, abstracts, and keywords to identify relevant studies examining conversation-based digital tools for health education or behavior change. The detailed search syntax is presented in Multimedia Appendix 1.

Two reviewers (SF and KEC) independently screened all titles and abstracts (Cohen κ=0.82), retrieving full texts when either deemed an article potentially relevant. Discrepancies were resolved by discussion, obviating the need for a third reviewer. Mixed methods studies were included if they provided relevant qualitative data, which were prioritized over quantitative findings during synthesis to ensure a comprehensive understanding of user and provider experiences. The final inclusion criteria limited articles to those presenting qualitative data on conversation-based digital tools in health education or behavior change, with participants or providers offering experiential insights. References were tracked and managed via EndNote (Clarivate) and manual cross-checking. An overview of the complete selection process is shown in Multimedia Appendix 2, along with the PRISMA flow diagram.

Studies were selected on the basis of the detailed inclusion and exclusion criteria outlined in Table 1. In brief, we included qualitative or mixed methods studies published in English or German between 2018 and 2023 that investigated user or provider experiences with conversation-based digital tools designed for health education or behavior change. These tools encompass various designations, including “chatbots,” “conversational agents,” and “digital health assistants,” all of which rely fundamentally on dialog-based engagement. This inclusive approach enables a comprehensive examination of automated conversational systems within health intervention contexts. The time frame (2018‐2023) was selected to capture recent developments in chatbot technology, while the language restriction was based on the research team’s proficiency. Studies were excluded if they lacked conversational functionality, did not provide qualitative data, or focused on domains unrelated to health education and behavior change. These criteria were systematically applied to ensure consistent study selection and reduce potential bias.

A standardized form was used to extract data on study design, participant demographics, chatbot features, outcomes, and key qualitative findings, ensuring consistency across studies. Two reviewers initially extracted data from 5 articles to establish a consistent approach before dividing the remaining articles; ongoing discussions resolved any ambiguities. The JBI Critical Appraisal Checklist for Qualitative Research was used to assess the methodological rigor and risk of bias (Multimedia Appendix 3). Rather than excluding studies failing to meet certain criteria, our appraisal informed the weighting of findings during synthesis.

We conducted reflexive thematic analysis [14], beginning with immersive reading and coding of text fragments and progressing through 6 phases: (1) familiarization, (2) initial coding, (3) theme development, (4) review, (5) definition and naming, and (6) reporting. Subsequent iterative rounds of coding and discussion refined the analysis, and the themes were validated against the context of each article. To ensure rigor, investigator triangulation was used, with both reviewers achieving high interrater reliability (Cohen κ=0.82 for the screening process), along with regular analytic meetings and reflective journaling to mitigate researcher bias. The identified themes were organized in alignment with our 3 RQs, and representative participant quotes were integrated to illustrate key findings and enhance the credibility of the analysis.

From the 1754 records identified, 27 articles from 10 countries satisfied our criteria, encompassing 169 participants in exclusively qualitative studies and 10,802 in mixed methods studies, of whom 657 contributed qualitative data (Figure 1). The studies varied in focus, encompassing chatbot design, technological innovations, behavior change, mental health, and user experience research. Participant ages ranged from adolescents to older adults, highlighting the broad adaptability of chatbots across diverse health contexts. The participants included patients, health care providers, and general users, with studies primarily using qualitative interviews or mixed methods designs. For a detailed summary of the study characteristics, see Table 2.

RQ1 explores the general experiences and perceptions of users and providers regarding chatbots in health education and behavior change, which is distinct from the specific adoption factors addressed in RQ2.

RQ2 investigates specific factors influencing the adoption of healthcare chatbots, distinguishing them from the broader experiences covered in RQ1.

RQ3 examines how theoretical frameworks of health behavior change and how technology acceptance guides chatbot design and implementation.

This section synthesizes insights from RQ1-RQ3 to illustrate how themes of empathy, trust, and theoretical grounding converge in the experiences of users and providers. Chatbots were seen as evolving interfaces capable of triaging minor questions before escalation, as highlighted by participants in Nadarzynski et al [30], who described chatbots effectively directing users either to immediate care or reassurance. They were also viewed as expansions of social networks, offering psychosocial support for isolated populations. Barnett et al [16] emphasized that chatbots might attract users needing extreme privacy and confidentiality, potentially engaging more individuals in treatment. By transparently integrating artificial intelligence (AI)–driven personalization with behavior-change theories, these digital tools not only strengthened emotional connections but also addressed key adoption barriers, creating a unified and trustworthy health-education experience.

Figure 2 provides a visual representation of the descriptive and analytical themes identified in our review, illustrating how user experiences (eg, empathy, trust) intersect with theoretical frameworks and adoption factors.

The methodological rigor of the 27 studies varied. Many aligned studies aim with qualitative methods, but some lack transparency in reporting researcher biases or ethical considerations. Using the JBI Critical Appraisal Checklist, 13 studies were at low risk of bias, 12 were at unclear risk, and 2 were at high risk, primarily due to limited ethical reporting or reflexivity. Figure 3 provides a summary of the risk of bias assessment for the included studies.

Our synthesis reveals robust enthusiasm for chatbots as tools for health education and behavior change. In addressing RQ1 (User and Provider Experiences), we found that users value empathy and emotional support, yet they express concerns about trust and privacy. For RQ2 (Facilitators and Barriers to Adoption), 24/7 accessibility, ease of use, and personalization emerged as key facilitators, whereas trust deficits, technical glitches, and cultural misalignment were identified as major barriers. With respect to RQ3 (Role of Theoretical Frameworks), the underutilization of models such as the HBM or TAM in chatbot design has emerged as a critical gap, limiting the potential for enhanced motivational and behavioral outcomes.

Earlier meta-analyses, such as those by Laranjo et al [2], reported that chatbots can improve health outcomes, but our qualitative review complements this by emphasizing user-centered dimensions such as trust, empathy, and cultural fit—factors often overlooked in quantitative studies. For example, Nadarzynski et al [30] identified privacy as a barrier to adoption, a concern supported by our findings and detailed in the results section. Our review also identified specific barriers, such as technical errors [19], privacy gaps [16,30], and cultural mismatches [7], which previous studies have not extensively addressed. This qualitative depth highlights that chatbot success depends not only on technical functionality but also on resonating with users personally and culturally—an insight crucial for developers and policymakers aiming to foster trust and adoption across diverse populations.

Our findings offer guidance for developers to embed behavior change theories and culturally adaptive content into chatbot design. Developers should prioritize incorporating prompts that address perceived barriers (eg, tailored reminders to overcome reluctance) and foster personalized feedback loops on the basis of theoretical constructs such as the HBM or information-motivation-behavioral skills model [22]. Developers should also embed culturally relevant content and empathetic responses to build trust and engagement. This systematic integration of theories is essential for optimizing chatbot effectiveness, addressing a gap in current designs.

Healthcare providers can integrate chatbots as triage tools to complement face-to-face care and reduce routine workloads. Providers should advocate for chatbots that support, rather than replace, human interaction. Training healthcare staff in how to use chatbot systems effectively can improve both patient outcomes and provider acceptance. Our qualitative insights suggest that providers are more likely to support chatbot integration when they perceive these tools as enhancing, not threatening, their professional roles.

Policymakers should promote transparent privacy standards and fund theory-based chatbot initiatives. Regulating data security and ensuring transparent privacy policies are essential to fostering user trust. Policymakers should support the integration of theoretical frameworks into digital health tools to ensure that they are evidence-based and effective. Funding initiatives promoting culturally sensitive chatbot designs could further support adoption. The emphasis on cultural relevance and ethical considerations in our findings points to a broader need for policies that prioritize user trust and inclusivity in digital health innovations.

A primary strength is the focus on qualitative evidence, offering granular insights into the emotional, cultural, and trust-based dimensions of chatbot use. The broad timescale (2018‐2023) captures recent innovations, including AI-based chatbots. However, the 2018‐2023 window may exclude older studies or those in languages beyond English/German, limiting the diversity of perspectives. Additionally, the rapid evolution of AI-driven chatbots, particularly large language models (LLMs), means that our results may not fully capture their latest affordances. The inclusion of broader conversational technologies [e.g., Biro et al [18], ; Chang et al [20], ; Svendsen et al [35], ] to offer holistic insights may reduce the specificity of findings for strictly defined chatbots. Selection bias may have influenced the findings, as many studies recruited tech-savvy participants, potentially overrepresenting positive experiences. The heterogeneity of study designs and limited focus on specific populations (eg, older adults) also constrain generalizability.

Future research should examine how user perceptions (trust, empathy) evolve over time (RQ1), assess cultural influences on adoption barriers/facilitators (RQ2), and pilot systematic integration of theories into chatbot workflows (RQ3). For example, longitudinal studies should explore adherence and user engagement across time, whereas cross-cultural research could illuminate how regional dialects and health beliefs affect adoption. Methodologically, future work should strengthen transparency through preregistration and the use of qualitative reporting frameworks (eg, Consolidated Criteria for Reporting Qualitative Research), enhance ethical safeguards, adopt inclusive recruitment strategies, ensure human support during chatbot interactions, and establish clear data usage and privacy protocols. Collaborative efforts among developers, clinicians, behavioral scientists, and ethicists are essential for refining chatbot designs, ensuring data security, and embedding effective behavior