Sage Journals: Discover world-class research

Abstract

Background

Individuals with prehypertension are at risk of developing hypertension, which affects many adults globally. Sustained physical activity (PA) can lower blood pressure, but maintaining long-term behavior change remains difficult. While PA habit formation interventions are promising, they face issues with scalability and accessibility. At the same time, behavior change chatbots have appeared, but their development often lacks systematic methods. Additionally, optimizing large language models (LLMs) to improve chatbot efficiency and reduce costs still needs more research.

Objective

This study introduces HabitBot, an LLM-integrated chatbot designed to foster PA habits in prehypertensive adults. HabitBot leverages LLMs for seamless interactions and integrates multidisciplinary insights, theoretical frameworks, and evidence to enhance the behavior change process.

Methods

HabitBot was developed through a systematic five-phase process: Phase 1, needs assessment via multidisciplinary discussions; Phase 2, literature review to identify relevant behavior change theories; Phase 3, selection of effective behavior change techniques (BCTs); Phase 4, intervention mapping for prototype design; and Phase 5, usability testing and focus group interviews for refinement.

Results

The process led to eight identified user needs and synthesized the Health Action Process Approach with Habit Formation Theory. Twelve effective BCTs were selected. The prototype was developed and refined across six dimensions based on user feedback. Evaluations indicated high usability, with a mean chatbot usability score of 3.84 (SD 0.82).

Conclusion

HabitBot integrates behavior change strategies with advanced LLM technology, representing a novel approach in chronic disease prevention. Future research should assess its long-term impact and generalizability.

Keywords

Prehypertension mHealth health behavior change physical activity chatbot large language models

Introduction

Hypertension remains one of the leading global risk factors for premature death and disability, contributing to an estimated 10.8 million preventable deaths and 235 million disability-adjusted life years annually.¹ Given that hypertension often requires lifelong medication once diagnosed,² there is a pressing need to implement early interventions targeting blood pressure regulation. Prehypertension, defined as blood pressure between 120–139/80–89 mmHg, affects approximately 25–50% of adults worldwide.³ According to the 2024 Chinese Guidelines for the Prevention and Treatment of Hypertension, over 41.3% of Chinese adults exhibit prehypertensive levels, with 8–20% progressing to hypertension annually.⁴ These data underscore the importance of early lifestyle interventions in the preclinical stage of hypertension.⁵

Among lifestyle strategies, increasing physical activity (PA) has emerged as one of the most effective non-pharmacological approaches to lowering blood pressure. Meta-analyses show that regular PA can lower systolic blood pressure by about 4.4 mmHg and diastolic by 4.2 mmHg on average,⁶ effects nearly comparable to certain antihypertensive medications.⁷ However, long-term behavior change is notoriously difficult to achieve, as individuals frequently struggle to initiate PA and maintain it consistently over time.^6,8 Thus, effective PA interventions must address not only behavior initiation but also maintenance over time.

Recent advances in behavioral science emphasize the importance of building automatic, context-driven routines—rather than relying solely on conscious motivation—to sustain long-term behavior change.⁹ Dual-process model proposes that while reflective processes (e.g. planning, intention formation) are crucial for behavior initiation, automatic processes (e.g. cue–behavior associations) are essential for habit formation.¹⁰ Theoretical frameworks such as the Health Action Process Approach (HAPA) and Habit Formation Theory offer structured explanations of how individuals transition from intention to sustained behavior.^11–13 These models provide actionable guidance for designing interventions that foster PA habits in real-world contexts.

With the rise of mobile technologies and artificial intelligence (AI), scalable and adaptive digital health interventions have become increasingly feasible.^14,15 In China, smartphone penetration now exceeds 73%, offering a strong foundation for digital solutions.¹⁶ However, adults with prehypertension still lack consistent behavioral support after health screenings.¹⁷ Most existing health apps prioritize data tracking over sustained lifestyle coaching, leading to a persistent disconnect between early risk detection and actual health behavior change.¹⁸ Conversational agents and chatbots show promise in offering real-time guidance and contextual prompts to support health-related behavior change.^19,20 However, few existing systems integrate evidence-based behavior change models and theories, especially for Chinese users and chronic disease prevention.

Recently, large language models (LLMs) have introduced new possibilities for enhancing the responsiveness and depth of behavior change chatbots.²¹ While several LLM-based health bots have emerged,²² most rely on single-turn interactions, limiting their ability to manage complex, multi-stage behavior change processes.^23–25 Notably, GPTCoach—a chatbot designed to promote PA—demonstrates the potential of LLMs in this field.²⁶ However, it lacks integration with validated behavior change frameworks and focuses primarily on motivation rather than long-term habit formation. To address these gaps, the present study introduces “HabitBot,” an LLM-driven chatbot specifically designed to foster PA habits among adults with prehypertension. HabitBot is built on a user-centered, theory-driven, and evidence-informed design process, aiming to bridge the gap between conversational fluency and structured guidance for behavior change.

Methods

Overview

This study was conducted in Beijing, China, where adults with prehypertension are predominantly middle-aged working individuals with sedentary occupations and low levels of PA.²⁷ The region has high smartphone penetration and widespread use of mobile health applications, providing a favorable environment for deploying an AI-enabled intervention to promote PA.²⁸ A preliminary version of this work was presented as an abstract at Nursing Informatics 2024.²⁹ This study follows the ethical principles and has received approval from the Peking Union Medical College (Ethics Approval Number: PUMCSON-2024–23). All participants provided written informed consent before taking part. Additionally, the resulting product of this study, HabitBot, is currently undergoing a randomized controlled trial to assess its effectiveness in behavior change. This trial has been registered with the Chinese Clinical Trials Registry (Registration Number: ChiCTR2400085073). The development framework was based on Song's research,³⁰ which was later adapted and revised. The chatbot was systematically developed through five steps (Figure 1). In phase 1, the needs of prehypertensive individuals were identified through a multidisciplinary group discussion. In phase 2, according to the dual-process model, behavior change theories fitting the selection criteria were identified through a literature review. In stage 3, a systematic review and meta-regression determined which components are most effective in current PA habit interventions. In stage 4, the content of the chatbot-based intervention was designed using intervention mapping (IM), and the prototype was developed by integrating findings from the previous three stages. In stage 5, the chatbot intervention was iteratively refined based on user feedback gathered through focus group interviews.

Figure 1.

Overview of the chatbot systematic development process.

Phase 1: Needs assessment

A multidisciplinary team of eight key stakeholders was assembled, including two members of the target population (individuals with prehypertension) and six experts in psychology, informatics, cardiovascular disease, and public health at Peking Union Medical College. The panel discussion explored the lifestyle habits, challenges, and motivations unique to individuals with prehypertension. Participants (mean age 36.3 ± 4.2 years; 83% female) were recruited through the research team's professional network. This discussion was audio-recorded and transcribed verbatim by HY. We then performed a thematic analysis of the transcripts to identify key themes.³¹ Two researchers (HY and HM) independently coded the transcripts, compared emerging codes, and refined the themes through iterative discussion. Any disagreements were resolved with the help of a third researcher (MP) to ensure the trustworthiness of the analysis.

Phase 2: Identifying applicable underlying theories

In Phase 2, our objective was to identify intervention theories that would most effectively underpin our efforts to promote and maintain PA using the dual-process model as a selection lens. Specifically, guided by the dual-process distinction between reflective and automatic determinants of behavior, we prioritized theories (or combinations of theories) that collectively address both PA initiation/maintenance through reflective self-regulation (e.g. intention formation, planning, self-efficacy) and long-term maintenance through automaticity and cue-dependent repetition (i.e. habit formation). To achieve a comprehensive understanding of the behavioral change theory landscape, we undertook an extensive literature review. Four databases (PubMed, Embase, PsycINFO, and CINAHL) were searched, with an emphasis on those relevant to PA maintenance or habit formation. The procedural details of our literature search and screening criteria for theories are documented in Supplementary Material 1.

Phase 3: Evidence identification

To consolidate evidence on PA interventions, we conducted a systematic review and meta-analysis of existing studies on PA habit formation intervention.³² To apply the most effective behavior change techniques (BCTs; standardized, observable intervention components, such as goal setting, self-monitoring, and feedback) to the design of the intervention content while avoiding those with adverse effects, we extracted BCTs that were positively associated with intervention effects from our meta-regression. Subsequently, we applied APEASE to ensure selected BCTs were not only effective but also feasible, acceptable, and safe for delivery via a WeChat-based chatbot.³³ We also scrutinized BCTs that showed a statistically significant negative correlation with intervention outcomes. The key findings and most relevant BCTs regarding the effectiveness of various interventions in promoting PA habits were extracted and used in the following intervention design.

Phase 4: Prototype design and development

We applied IM, a structured framework for developing theory- and evidence-based health interventions, to integrate identified needs, theories, and evidence into the chatbot prototype.³⁴ Our initial step involved a thorough mapping of the specific needs of prehypertensive individuals, identifying key barriers and facilitators that prevent regular PA. For each identified need, we developed corresponding intervention strategies. This included choosing the delivery platform (e.g. a mobile app or social media), designing the chatbot's dialogue flow to ensure user engagement and continuity, and creating tailored dialogue content to provide personalized support and encouragement. To meet the user's need for “accessibility and convenience” identified in the needs assessment (detailed in the results section), we chose the WeChat Mini Program as the leading platform for the intervention. The WeChat Mini Program has a broad user base and convenient social features that make it easy to increase user engagement and retention. In addition, leveraging WeChat's existing infrastructure can significantly reduce development and maintenance costs.

Second, we extract influential pathways for PA initiation and habit formation from the selected theories (e.g. self-efficacy and action planning from HAPA¹³; cue–behavior association from habit formation theory.³⁵ Based on these theoretical pathways, we developed the intervention model and its contents and strategies.

Third, guided by findings from our systematic review and meta-analysis,³² we prioritized 12 BCTs with demonstrated efficacy in facilitating PA habit formation. We developed a BCT–script matrix to systematically map each selected BCT to its corresponding chatbot dialogue structure (see Supplementary Material 2).

Fourth, we integrated all identified functionalities and content into a unified system architecture. This integration involved the construction of functional modules, interactive interfaces, and databases, ultimately leading to the implementation within the HabitBot WeChat service account and Wechat mini program.

To support this chatbot-based digital intervention, we utilized the HUAWEI WATCH as our wearable device. This smartwatch was selected not only for its China Food and Drug Administration-certified cuff-based blood pressure measurement accuracy³⁶ but also for its vital role in supporting ongoing self-monitoring and personalized feedback within the intervention. Linking BP data with HabitBot enabled users to monitor short-term BP changes, increase risk awareness, and receive personalized behavioral messages. The wearable also provided vibration-based exercise prompts, implementing key BCTs such as self-monitoring, feedback, and contextual cues. Therefore, the wearable was a crucial part of the pilot system.

The development of the chatbot's conversational flow and operational logic was facilitated using the Botpress Cloud platform (https://botpress.com/). This platform was chosen for its intuitive development environment, minimal coding requirements, and its support for advanced language processing capabilities. Notably, Botpress Cloud incorporates the GPT-4o language model,³⁷ which equips our chatbot with the capability to understand and generate responses to user inputs in natural language, tailoring dialogue based on predefined prompts, thereby enhancing the chatbot's conversational quality and user interaction experience.

To protect privacy and confidentiality, HabitBot was designed with basic data-protection safeguards. Only data required for intervention delivery and evaluation (e.g. PA metrics, BP readings, and task completion logs) were collected. Identifiers were minimized in the research dataset, and access to the study database was restricted to authorized research personnel. The intervention was reported following the TIDieR checklist, with a completed checklist provided in the Supplementary Materials 3. In addition, we reported the AI-enabled development and early evaluation following the DECIDE-AI reporting guideline; a completed DECIDE-AI checklist is provided in the Supplementary Materials 4.

Phase 5: Refining the prototype and evaluation

Four online focus groups were conducted via Tencent Meeting (similar to Zoom) from October to December 2023 to gather user feedback and enhance the prototype. Participants were recruited through Xiaohongshu (a popular social media platform in China) using convenience sampling. An open invitation post was used, and those interested were screened for eligibility. Inclusion criteria included: (1) meeting prehypertension diagnostic criteria, (2) aged 18–60 years, (3) low baseline PA level, (4) owning a smartphone with WeChat installed, and (5) normal cognitive function. Participants who completed the focus groups were subsequently invited to participate voluntarily in the pilot usability testing. Those who agreed were enrolled in individual testing sessions. A total of 36 eligible participants joined the focus groups and usability testing sessions.

Three days before each formal interview, we release the chatbot and wearable device to users for preliminary testing. During this period, users were provided with a list of test tasks encompassing the main features of HabitBot and were encouraged to explore as many of these features as possible within the three days. Designed and moderated by HM, the testing and review sessions aimed to facilitate interaction and ensure a structured yet open dialogue among participants. These interviews followed a set of predefined questions and prompts aimed at eliciting detailed feedback on various aspects of the chatbot. The procedures and interview questions utilized in the focus group interview are detailed in Supplementary Material 5.

At the conclusion of the interviews, all participants were asked to fill out the Chatbot Usability Scale (BUS), which had been translated and adapted by our research team to assess the usability of HabitBot.³⁸ The BUS is a reliable and valid usability scale that provides a comprehensive view of a chatbot product's subjective usability at the end of a study. It comprised 11 items, measuring five dimensions of chatbot usability: perceived accessibility to chatbot functions, perceived quality of chatbot functions, perceived quality of conversation and information provided, perceived privacy and security, and response time. The overall result is calculated by averaging all item scores (on a 5-point Likert scale), with higher scores indicating higher usability. Each participant received a voucher worth RMB 50 upon completing the interviews and questionnaires.

The focus group participants were all in the prehypertension stage, with an average systolic blood pressure of 127.44 (SD 6.04) and an average diastolic blood pressure of 81.42 (SD 6.26). Among them, 58.33% (21/36) were male, and 41.67% (15/36) were female. Participants aged 18–25, 26–30, 31–40, 41–50, and 51–60 years accounted for 16.67% (6/36), 30.56% (11/36), 27.78% (10/36), 19.44% (7/36), and 5.56% (2/36), respectively. Additionally, 66.67% (24/36) of the participants had prior experience interacting with other chatbots.

The interviews were audio-recorded, transcribed (by HM), and verified (by HY). The first coder (HY) analyzed the transcripts using NVivo software and initially extracted themes with indexed codes. Subsequently, HY discussed the preliminary coding and thematic results with the second coder (HM). Then, HM independently coded 50% of the transcripts and provided feedback to refine the initial themes. The themes were further refined and revised through an iterative process to ensure they accurately reflected the core content of the discussions. Any disagreements were resolved through discussion with a third researcher (MP) until consensus was reached. Finally, the team reviewed each suggestion and implemented iterative refinements for those deemed feasible, appropriate, and aligned with the intervention goals.

Result

Phase 1: Needs assessment

Eight potential needs for people with prehypertension were proposed through a panel discussion, including lack of awareness and knowledge, motivational gaps, difficulty in habit formation, personalization needs, accessibility and convenience, accountability and monitoring, social support, and safety concerns. Table 1 includes these themes, further description, and direct quotes from the panel discussion.

Table 1.

PA intervention need for prehypertension individual from the panel discussion.

Theme	Theme description	Example quotes
Lack of Awareness and Knowledge	Many prehypertension individuals demonstrated limited awareness of the specific benefits of regular PA in managing and preventing the progression of hypertension. Most individuals have a superficial understanding of the benefits of PA (e.g. weight loss) and is not comprehensive or in-depth.	“…a lot of people don't know they have prehypertension. And even if they did, it's not clear what to do… doctors don't have time to explain in detail, so we have to look it up online, but the information is so cluttered that it's hard to tell what really works.” (Male, 32 years old, Prehypertension individual) “It's striking, really… a lot of those we talked to don't quite see how vital regular workouts can be for controlling blood pressure issues. Their focus is often just on medication, not realizing the whole picture, which includes staying active.” (Female, 31 years old, Public health expert)
Motivational Gaps	A lack of motivation to initiate or maintain regular exercise routines is often mentioned among prehypertension individuals.	“Some of the intervention message being perceived as too academic and not engaging enough to inspire action…It seems there's a missing element that would spur them into action, keep them committed.” (Female, 42 years old, Nursing Informatics expert) “…how to tap the inner motivation of people through this interactive relationship is what we need to focus on. For example, in psychology, there is a technique called motivation interviewing, which help people recognize existing or potential problems, thereby increasing their motivation to change…” (Female, 36 years old, Psychologist)
Difficulty in Habit Formation	The challenge of developing and sustaining exercise habits was commonly reported, often due to lack of guidance and support.	“What we're seeing is… a real struggle to establish and maintain an exercise routine… it is OK for the first few days, but once it is busy or interrupted by other things, it is easy to give up, and it is more difficult to pick up again.” (Male, 39 years old, Cardiologist) “We're all very busy… and of course I know that exercise is good for my health, but a lot of times it takes a back seat to work and family.” (Male, 32 years old, Prehypertension individual)
Personalization Need	Prehypertension individuals indicated a desire for personalized exercise recommendations that consider their health status, preferences, and lifestyle.	“…they’re looking for something more tailored. They want exercise recommendations that really consider their health, preferences, lifestyle and not just one size fits all.” (Female, 32 years old, Exercise expert) “Yeah, I have some knee problems myself and I can't do a lot of high-intensity sports. It would be better if there was an exercise program specifically for my condition.” (Female, 53 years old, Prehypertension individual)
Accessibility and Convenience	The need for easily accessible and convenient methods to receive guidance and support for PA was emphasized.	“… Making plans can be really draining. Should I start today or wait until tomorrow? I wish someone could just make these decisions for me, so I can just get on with it. Of course, they'd need to know my specific situation first.” (Female, 53 years old, Prehypertension individual) “Requiring users to specifically download a health app may be an additional burden, and often they do not remember to open and use it. Meanwhile, there's a clear call for more straightforward, user-friendly ways to access fitness guidance. Simplicity and ease of use are what they’re after.” (Female, 38 years old, Nursing Informatics expert)
Accountability and Monitoring	A lack of accountability and self-monitoring mechanisms was seen as a barrier to consistent PA.	“If I had a system to track my progress and give me feedback and advice, I'd be more motivated to stick to it.” (Male, 32 years old, Prehypertension individual) “What's missing, and it's quite evident, is a way for individuals to track and be answerable for their own physical activity. Without this, staying on track becomes a major challenge.” (Female, 32 years old, Exercise expert)
Social Support	The absence of social encouragement and support was identified as a hindrance to engaging in regular PA.	“…they would felt quite good when they can interact with somebody. It's important to have companionship and motivation…” (Female, 36 years old, Psychologist)
Safety Concerns	Concerns about exercising without professional guidance, especially considering their prehypertensive status, were noted.	“Safety worries, particularly in exercising without expert guidance, are quite prevalent… they're looking for assurance that they're not putting themselves at risk.” (Female, 32 years old, Exercise expert)

PA: physical activity.

Phase 2: Theory identification

Following the Phase 2 literature review and screening (Supplementary Material 1), we selected HAPA and habit formation theory because, together, they cover reflective and automatic pathways highlighted by dual-process perspectives: HAPA effectively addresses the reflective, motivational and volitional stages of behavior change (e.g. risk perception, outcome expectancies, planning), whereas habit formation theory focuses on the development of automatic, context-triggered behaviors (e.g. repetition, cue–behavior association). This combined theoretical model covers both the initiation of PA and its maintenance through habit formation. A detailed comparison of these two theories, along with a summary of the screening of 11 candidate theories, is available in Supplementary Material 6. Figure 2 illustrates our integrated theoretical model, highlighting eight key constructs (self-efficacy, risk perception, outcome expectations, action planning, coping planning, behavior repetition, cue–behavior association, and affect) that underpin the HabitBot intervention.

Figure 2.

The theoretical model of intervention.

Phase 3: Evidence synthesis

To consolidate evidence on PA habit formation interventions, we leveraged our previously published systematic review and meta-analysis,³² and conducted an APEASE-based feasibility appraisal to translate the evidence into chatbot-deliverable components. Based on our research, we identified 12 BCTs as being instrumental in facilitating PA habit formation according to APEASE criteria (BCTs evaluations were shown in Supplementary Material 7), which included: Problem Solving, Action Planning, Habit Formation, Feedback on Behavior, Information about antecedents, Goal setting (behavior), Goal setting (outcome), Social support (unspecified), Instruction on how to perform behavior, Self-reward, Self-monitoring of behavior, and Behavioral practice / Rehearsal.

Notably, in our analysis, “social reward” showed a negative association with PA habit formation, suggesting that emphasizing external praise might undermine intrinsic motivation.³² This finding offers valuable guidance for the design of interventions.

Phase 4: Integrating need, theory and evidence into the prototype design and development

Figure 3 presents the logic model of the HabitBot intervention, illustrating how user needs, theoretical constructs, and evidence-based BCTs were mapped onto the chatbot's design. By aligning every feature of HabitBot with identified user needs and proven principles of behavior change, the intervention was grounded in a comprehensive framework. We anticipate that this integrative design may improve users’ PA levels (short-term outcome) and, in turn, could contribute to better blood pressure control and quality of life (long-term outcomes). These potential benefits remain to be confirmed in future evaluations.

Figure 3.

Logic model of the HabitBot.

In summary, Phase 1 identified eight user needs related to PA habit formation. Phase 2 yielded an integrated theoretical foundation combining HAPA and habit formation theory to address both conscious decision-making and automatic habit processes. Phase 3 provided empirical guidance, identifying the key BCTs (10 with positive influence and one with negative influence on habit formation) to include in the intervention. These elements were systematically translated into HabitBot's design (Table 2 details the mapping of user needs to theoretical constructs and BCT-based features).

Table 2.

Chatbot platform functionalities based on user needs, theoretical pathways, and evidence.

Source (Theory/need/evidence)	Behavior change technique	Chatbot intervention content	HabitBot modules
Need: Accessibility and Convenience	12.5 Adding Objects to the Environment 7.1 Prompts/Cues	For user convenience, the front end of the Chatbot platform is a WeChat mini-program, eliminating the need for users to register and download an app. Each interaction with the chatbot is pushed and displayed in the user's chat list.	Platform integration & user onboarding
Theory: Outcome Expectations; Need: Motivational Gaps Evidence: Goal Setting (outcome) (+)	1.3 Goal Setting (Outcomes)	The chatbot guides and asks users about their health goals and provides evidence-based predictions of positive outcomes based on these goals. For instance, if a user wants to improve their blood pressure, the chatbot informs them of how much they can expect their blood pressure to decrease through weekly exercise.	Personalized goal-setting & planning
Theory: Action Planning Need: Personalization Need Evidence: Goal Setting (behavior) (+)	1.1 Goal Setting (Behavior)	The chatbot assesses the user's health goals, cardiopulmonary fitness, and past injury and illness history. Based on this information, the chatbot provides the user with personalized physical activity recommendations based on prehypertension management guidelines. The chatbot guides the user to start with a small, achievable goal, such as a five-minute walk instead of a thirty-minute walk.	Personalized goal-setting & planning
Theory: Self-Efficacy Need: Motivational Gaps	6.2 Social Comparison	The chatbot generates a success story similar to the patient's situation based on their basic information and provides participants with these success stories and tips to boost their confidence.	Emotion & motivation management + social support
Theory: Outcome Expectations Risk perception Need: Lack of Awareness and Knowledge	5.1 Providing Information about Health Outcomes	We integrated a knowledge base in the chatbot, containing guidelines and evidence-based recommendations for the management of prehypertension. Users can ask the bot any questions they have, and the bot will generate personalized answers, complete with references. The chatbot emphasizes to users the negative effects of blood pressure on cardiovascular health and overall well-being. It provides specific information on how regular physical activities can improve their health conditions and highlights how these health benefits can positively impact daily life.	Intelligent Q&A & education
Theory: Action Planning Need: Personalization Need Evidence: Action Planning (+)	1.1 Goal Setting (Behavior) 1.4 Action Planning	The chatbot guides the user to develop their own plans. The chatbot collects information through dialogue interaction about the type of exercise, exercise timing, and contextual cues that trigger the exercise. Further, the chatbot forms a personalized habit formation plan for the user.	Habit formation support
Need: Safety concerns Evidence: Action Planning (+) Evidence: Instruction on How to Perform the Behavior (+)	4.1 Instruction on How to Perform the Behavior 3.1 Social Support (Unspecified) 7.1 Prompts/Cues	The chatbot will send the user a safety notification regarding the exercise. This includes setting a target heart rate range based on the user's heart rate, reminding them to warm up and hydrate with electrolytes, ensuring the exercise environment is safe and comfortably temperatured. It also reminds the user to exercise within their limits and to stop immediately and seek help if any discomfort occurs.	Safe exercise tips
Evidence: Behavioral practice/Rehearsal (+)	8.1 Behavioural practice/Rehearsal	After the user sets the task, the chatbot will guide the user through the whole process of task execution in their mind, including the scene before the task starts, the specific process during the task execution, and the scene after the task is completed.	Habit formation support
Need: Accountability and Monitoring; Theory: Self-efficacy Evidence: Feedback on Behavior(+)	2.2 Feedback on Behavior; 2.3 Self-monitoring of Behavior; 1.6 Discrepancy between Current Behavior and Goal; 5.1 Information about Health Consequences	First, the user's monthly plan will be visually displayed in the “My Plan” module, where the monthly calendar will differentiate between “Completed,” “Pending,” and “Not Completed” tasks using different colors, as illustrated in the second image of Figure 4. Secondly, health data, including heart health metrics (e.g. heart rate, atrial fibrillation), exercise data (e.g. steps, calories burned), blood pressure, and blood oxygen levels, which are monitored by wearable devices, will be synchronized with the cloud server on an hourly basis. This data will be visually displayed in the “My Data” module. Users have continuous access to their task completion progress and health metrics.	Personalized goal-setting & planning
Theory: Coping Planning Evidence: Problem-solving (+) Need: Personalization Need	1.2 Problem Solving 1.4 Action Planning 11.2 reduce negative emotions	The chatbot will address the following issues: plan adjustments; forgetting plans; lack of motivation; excessive stress; other targeted solutions for specific exercise barriers encountered; lack of knowledge; requests for human services. Taking plan adjustments and stress management as examples: the chatbot will replan the user’ PA tasks, including the type of exercise, timing, location, duration, and exercise companions. It will then adjust the action plan based on the gathered information. When a user faces the challenge of excessive stress, the chatbot will suggest using a wearable device for breath meditation and provide a 5-min meditation audio resource.	Emotion & motivation management + social support
Need: Difficulty in habit formation Theory: Behavior-Cue Association; Behavior Repetition Evidence: Habit Formation (+); Information about antecedents (+)	8.3 Habit Formation; 7.1 Prompts, Cues 4.2 Information about antecedents	The chatbot explains to the user the importance of setting contextual cues. It prompts the user to recall routines in their life and identify specific cues (based on time or events) that can be linked with a new exercise habit. Based on the collected cues, the chatbot will automatically remind the user to perform the “habit formation task” at the set times through a WeChat mini-program.	Habit formation support
Theory: Behavior-Cue Association Need: Difficulty in habit formation	7.1 Prompts, Cues 12.5 Adding Objects to the Environment	The chatbot encourages the user to add exercise prompts and cues in the environment, such as changing the phone's wallpaper to a gym background or placing workout gear in a prominent place.	Emotion & motivation management + social support
Theory: Affect Theory: Behavior repetition Need: Motivational Gaps Evidence: Feedback on behavior (+) Self-Reward (+) Social reward (−)	10.4 Social Rewards 2.2 Feedback on Behavior	The chatbot reminds the user to reward themselves for progress made. Furthermore, the chatbot discusses with the user how they felt after exercising, eliciting positive experiences associated with the activity, thereby enhancing the participant's affect towards PA. To enhance their intrinsic motivation, the chatbot will provide targeted praise and encouragement after users complete tasks, emphasizing the importance of personal progress and the effort process rather than merely focusing on achieving a specific amount of physical activity or outcomes. This approach makes the process of effort rewarding, rather than the results of the effort being the sole reward.	Habit formation support
Evidence: Social support (+)	3.1 Social Support	The chatbot advises users to find reliable social support, such as finding a fitness buddy or asking family and friends to help themselves.	Emotion & motivation management + social support
Theory: Behavior Repetition Need: Motivational Gaps	5.1 Providing Information about Health Outcomes	The chatbot discusses with the user about the benefits of developing an exercise habit when the user encounters issues needing resolution and reports their unawareness of the benefits of habit formation. For example, it tells the user that forming a habit usually requires extra willpower in the early stages, but once an exercise habit is established, it will no longer require as much willpower and will become a part of your life, helping you stay healthy even in a busy lifestyle. Moreover, in the process of forming an exercise habit, there may be a spillover effect; you might start paying attention to your diet, make more friends to exercise with, and let exercise contribute to a better life.	Intelligent Q&A & education

Behavior Change Techniques are divided according to the Behavior Change Techniques (BCTs) Taxonomy. (+): indicates that the BCT is positively correlated with the intervention effect. (−): indicates that the BCT is negatively correlated with the intervention effect.

PA: physical activity.

Specially, HabitBot's implementation comprises four main components: (1) a WeChat-based user interface (delivering accessibility and convenience), (2) a chatbot backend powered by both rule-based flows and an LLM for personalized guidance, (3) a cloud server aggregating user data for monitoring and feedback, and (4) a BP-monitoring smartwatch providing safety and accountability through physiological tracking and reminders (Figure 4 illustrates the system architecture).

Figure 4.

System architecture of HabitBot.

By leveraging WeChat's ubiquity, HabitBot minimized barriers to use (no separate app installation required) and turned the routine act of messaging into an intervention delivery mechanism (WeChat notifications served as stable cues for habit formation). Figure 5 shows a screenshot of the “My Coach” chat interface, and Figure 6 outlines the five main dialogue pathways users can engage with. Figure 7 provides an example chat sequence, and Figure 8 demonstrates how the rule-based and LLM-driven components interact during a personalized exercise planning dialog. All the prompts we used were presented in Supplementary Material 8.

Figure 5.

Screenshot from “My coach” of the HabitBot chatbot intervention.

Figure 6.

Overall configuration of the “My coach” module.

Figure 7.

Screenshot of example dialog with “My Coach.”

Figure 8.

Example explanations for integration of rule-based and LLM components in HabitBot. LLM: large language model.

Phase 5 prototype refining and evaluation

A total of 36 users participated in the focus group evaluations of HabitBot. Qualitative analysis of the focus group discussions revealed six major themes regarding user experience and suggestions: (1) User Interface & Interaction (usability and visual appeal of the chatbot interface), (2) Content & Resource Accessibility (clarity and usefulness of information provided, links to resources), (3) Chatbot Features (desired functionalities, such as memory of past conversations), (4) Individualization & Adaptability (personalization of advice and coach persona options), (5) Privacy (comfort with data sharing and security), and (6) Support & Community (the need for social support or group features). We used this feedback to implement several key improvements in the prototype. For example, we simplified medical jargon in the chatbot's messages to improve clarity (Theme 2), added the option for users to choose between different coach “personalities” to increase engagement (Theme 4), and introduced more social support features (like encouraging messages and an option to share achievements with a friend) to address Theme 6. We also emphasized privacy protections in the user onboarding (e.g. informing users that their data is stored securely and not shared without consent) in response to Theme 5. A detailed list of user feedback and corresponding changes is provided in Supplementary Material 9.

The usability evaluation results, as shown in Table 3, indicate that the overall usability score of HabitBot was 3.84 (SD = 0.82), suggesting that HabitBot was generally effective and easy to use for users. Across all items in the BUS, the average scores received positive ratings above 3.0. Among the five dimensions assessed, the scores ranked from highest to lowest as follows: Perceived Accessibility to Chatbot Functions (Mean = 4.07, SD = 0.78), Perceived Quality of Chatbot Functions (Mean = 3.88, SD = 0.78), Perceived Quality of Conversation and Information Provided (Mean = 3.84, SD = 0.80), Time Response (Mean = 3.72, SD = 0.91), and Perceived Privacy and Security (Mean = 3.36, SD = 0.90).

Table 3.

Chatbot usability scale scores of each item (n = 36).

Items-English version	Score, mean (SD)
Perceived accessibility to chatbot functions (item 1–2)	4.07 (0.78)
1. The chatbot function was easily detectable	4.17 (0.81)
2. It was easy to find the chatbot	3.97 (0.74)
Perceived quality of chatbot functions (item 3–5)	3.88 (0.78)
3. Communicating with the chatbot was clear	4.00 (0.72)
4. The chatbot was able to keep track of context	3.53 (0.81)
5. The chatbot's responses were easy to understand	4.11 (0.71)
Perceived quality of conversation and information provided (item 6–9)	3.84 (0.80)
6. I find that the chatbot understands what I want and helps me achieve my goal	3.72 (0.74)
7. The chatbot gives me the appropriate amount of information	3.94 (0.75)
8. The chatbot only gives me the information I need	3.75 (0.91)
9. I feel like the chatbot's responses were accurate	3.94 (0.79)
Perceived privacy and security (item 10)	3.36 (0.90)
10. I believe the chatbot informs me of any possible privacy issues	3.36 (0.90)
Time response (item 11)	3.72 (0.91)
11. My waiting time for a response from the chatbot was short	3.72 (0.91)
Overall usability (item 1–11)	3.84 (0.82)

Discussion

Principal findings

This study introduces HabitBot, a novel LLM-integrated chatbot developed to promote PA habit formation among prehypertensive individuals through a systematic design process. To our knowledge, it is the first study to develop AI chatbot-based PA behavior change intervention for prehypertensive individual integrating needs, theories, and evidence. Notably, it is also the first to embed LLM technology within a comprehensive theoretical and evidence-based framework, thereby addressing highly personalized user needs while maintaining a focus on behavior change principles.

Based on the development process of our system, we identified eight user needs, two behavior change theories, and 12 BCTs that support the formation of PA habits. These findings can be used to help prehypertensive individual develop long-term PA behavior. Moreover, the systematic development process may offer a roadmap for other behavior change interventions seeking to leverage both theory and advanced AI-driven tools.

Interpretation of key findings

First, in alignment with the previous findings, the results from phase 1 highlighted several barriers and facilitators within long-term PA interventions. At the individual level, boosting motivation, raising awareness, self-monitoring, and habit formation are all essential. At the implementation level, it is essential to consider the personalization, accessibility, social support, and safety of the interventions. Additionally, our inclusion of multidisciplinary stakeholders in Phase 1 diverges from traditional user-centric assessments by incorporating the observational insights of professionals from diverse fields, thus providing a more nuanced understanding of user needs.^30,39

Second, our theoretical intervention model, which integrates HAPA and Habit Formation Theory, emphasizes the need to address both reflective and reflexive pathways. In line with this, recent behavioral interventions have begun adopting dual-process strategies,^40,41 recognizing that maintaining health behaviors requires both initial motivation and habit formation.⁴² Moreover, many scholars now consider “habit discontinuity” moments—such as life-stage transitions or new diagnoses—as pivotal opportunities to disrupt established unhealthy routines and foster healthier alternatives.⁴³ By combining HAPA with Habit Formation Theory, HabitBot seeks to capitalize on the contextual upheaval brought about by a prehypertension diagnosis, thereby increasing the likelihood of instilling new, sustainable PA habits.

Third, akin to prior findings,⁴⁴ our research builds upon the foundational work of our team to identify 12 BCTs highly relevant to the formation of PA habits, such as problem solving, action planning, and feedback on behavior. Notably, among these 12 BCTs, we observed a potential negative correlation between social reward and PA habit interventions, which supports the results of a previous study by Cherubini et al.⁴⁵ Because this evidence suggested a potential risk of undermining intrinsic motivation, we revised the reward messaging after the task was completed. Instead of generic praise for good performance, we shifted to commending the effort itself and incorporated guidance for participants to recognize the intrinsic positive feelings following exercise. This comprehensive consideration of BCTs aims to maximize the utility of empirical evidence in promoting effective habit formation.

Fourth, we systematically translated needs, theories, and evidence into the intervention framework and interaction content, representing a more comprehensive and systematic approach to chatbot development than previously observed.²⁰ To meet the demands for convenience and accessibility, the intervention was developed on a WeChat mini program, in line with considerations similar to those of Chen et al.¹⁴ This delivery way relieves the need for users to undergo the burdens of downloading, registration, and logging in. Moreover, as a daily chat application, WeChat could function as a stable cue, allowing user to retain habit-dependent cues even in changing environments and thus better avoid habit discontinuity.⁴⁶

Fifth, in the refinement phase of Phase 5, we undertook direct user testing of the product alongside focus group interviews, which provided profound insights beyond our anticipatory scope. Notably, the qualitative feedback acquired, though not exhaustive to the point of data saturation, offered indispensable perspectives for the AI chatbots’ continuous refinement. Participants highlighted critical aspects such as the user experience, the enjoyment derived from interactions, and the intrinsic need for social connectivity. These aspects resonate with Cheng et al.'s observations, which emphasize the importance of user experience in technology adoption.⁴⁷ Further, our findings align with Zhang et al.'s emphasis on the significance of building relational capacity in chatbots as a crucial factor for sustaining user engagement.⁴⁸ This underscores the evolving role of digital tools in enhancing user engagement and satisfaction, particularly in therapeutic interventions. Notably, the notion that the benefits to one's health may predominantly influence the initial adoption stage of behavior, while the maintenance of healthy behavior hinges on the individual's enjoyment and engagement, adds a nuanced layer to our understanding of user interaction with AI chatbots.⁴⁹ Recent advancements in AI and natural language processing have significantly lowered the barriers to developing chatbot interventions. However, as highlighted by previous works,^50,51 the challenge remains in creating chatbots that are genuinely user-centered and theoretically informed.

Contributions of the large language model-integrated hybrid approach

Lastly, HabitBot integrates a hybrid approach that combines the strengths of rule-based and LLM-based chatbots to balance the advantages of both types.⁵² Many existing applications solely rely on one type of system, either rule-based or generative.^50,53 HabitBot employs a strategy that not only ensures the effectiveness of interventions but also enhances the flexibility of interactions. This dual focus aligns with the findings of Maeng et al., emphasizing the importance of integrating both structured and dynamic elements in digital interventions to achieve optimal outcomes.⁵⁴ By integrating rule-based strategies, the application ensures that both the content and the process of the intervention are grounded in a robust theoretical and empirical habit formation framework.⁵² For instance, the chatbot aids users in locating nearby resources and offers personalized exercise advice before assisting them in creating an exercise plan.

Additionally, its intent-based features, powered by advancements in LLMs, enable the chatbot to understand user intents through natural language processing techniques. For example, when a user says in natural language, “I have to work overtime tonight and cannot complete the task,” HabitBot can not only recognize that he has not completed today's PA task, but also understand the implicit intent behind him. So HabitBot further asks the user if problem solving is needed. Meanwhile, the text generation capabilities of LLMs overcome the limitations of template-based chatbots by providing targeted, empathetic, and high-quality responses, significantly enhancing user experience and contributing to the progression of digital health interventions.⁵⁵ This is especially important in situations where the difficulties faced by each user during their daily activities are highly individualized (e.g. his busy schedule at work, family, and life or physical state limits his opportunities for exercise). Furthermore, using the LLMs, we can build a knowledge base with integrated health management guidelines at a low cost, giving users access to scientific and personalized professional responses through natural language queries.⁵⁶ Through these enhancements, HabitBot demonstrates the potential of blending rule-based and AI-driven approaches in enriching digital health solutions.

Limitations and future work

This study has several limitations. First, the intervention design process involved some subjective judgment. For example, the selection of theories in Phase 2 was influenced by the research team's expertise and interpretation of the literature, potentially biasing the framework by omitting other relevant theories. In future research, methods such as the Delphi consensus approach could be used to incorporate a broader range of expert opinions and reduce individual bias. Second, the Phase 1 needs assessment was based on a small sample of 8 stakeholders. This limited sample size may not have captured all potential user needs, and the insights might not be fully comprehensive. Conducting additional focus groups or interviews with more participants, including end-users, would likely provide a deeper understanding of prehypertensive individuals’ needs and help achieve data saturation. Third, the study participants—both experts involved in development and users in evaluation—were all from central urban areas of Beijing, had relatively high educational levels, and were young-to-middle-aged working professionals. This homogeneity limits the applicability of our findings, especially to rural populations, older adults, or those with lower digital literacy. Future research should test HabitBot in more diverse groups, especially in remote or underserved areas where accessible digital interventions could be particularly beneficial. Fourth, although basic privacy and security measures were integrated into HabitBot (such as secure data storage and limited access to personal information), participants’ ratings for perceived privacy/security were the lowest among the usability aspects. This shows that there is room to improve both the implementation of privacy features and how we communicate these protections to users. Enhancing data-protection measures (such as two-factor authentication and clearer privacy notices) and openly addressing privacy concerns will be essential as we refine the design.

Finally, a large-scale evaluation of HabitBot's effectiveness is essential. We have already started a randomized controlled trial (as mentioned earlier) to assess the chatbot's impact on PA behavior and clinical outcomes such as blood pressure, waist circumference, and body mass index. The results from this trial will offer more conclusive evidence of HabitBot's effectiveness in improving health metrics and will help identify any necessary adjustments before broader implementation. In the future, we also plan to explore long-term user engagement with HabitBot and whether the PA habits formed are sustained over time without ongoing chatbot support.

Conclusion

Ensuring that the development process of behavior change chatbots is rigorous and scientifically grounded, while effectively integrating LLM technology to enhance the fluidity of conversations within a structured framework, is a critical issue in global public health. This study describes the rigorous development of HabitBot, a chatbot-based intervention designed to help prehypertensive users form PA habits. In this development process, through rigorous needs assessment, theoretical screening and combined with empirical evidence, we provide useful references for future healthcare chatbot development. As AI and digital health technologies advance, solutions like HabitBot are poised to become fundamental components of public health strategies, providing scalable, personalized, and effective approaches for preventing and managing chronic diseases.

Supplemental Material

sj-docx-1-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-docx-1-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-docx-2-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-docx-2-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-pdf-3-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-pdf-3-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-pdf-4-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-pdf-4-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-doc-5-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-doc-5-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-docx-6-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-docx-6-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-xlsx-7-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-xlsx-7-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-docx-8-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-docx-8-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Supplemental Material

sj-docx-9-dhj-10.1177_20552076261421367 - Supplemental material for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension

Supplemental material, sj-docx-9-dhj-10.1177_20552076261421367 for Development and evaluation of an large language model-integrated chatbot intervention for physical activity habit formation in adults with prehypertension by Haoming Ma, Hongzhen Cui, Hongyu Yu, Runyuan Pei, Sijia Li, Aoqi Wang, Xingyi Tang, Guangnan Liu and Meihua Piao in DIGITAL HEALTH

Footnotes

Acknowledgements

We thank the multidisciplinary team of experts who participated in this study, including Professor Xin Zhang from Peking Union Medical College, Associate Professor Zhenling Ma, Dr Xinghe Huang, Professor Yu Sheng, Manager Guannan Zhu from Taikang Life Insurance Company, and Engineer Yuanhang Zhang from Shijiazhuang Cloud Technology Co., LTD. We also thank every patient who participated in this study.

Abbreviations

ORCID iDs

Haoming Ma

Meihua Piao

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the School of Nursing, Peking Union Medical College (ethical approval number: PUMCSON-2024-23). All participants provided informed consent.

Consent for publication

The authors consent to publication.

Authors’ contributions

All authors have read and approved the final work. Every author significantly contributed to the study. HM contributed to the study through conceptualization, data curation, formal analysis, investigation, methodology, project administration, resources, software, supervision, and writing—original draft, as well as writing—review & editing. HC contributed to the software development and provided supervision throughout the study. HY contributed to the conceptualization, investigation, and methodology of the study. RP contributed to data curation and validation of the study. SL contributed to the validation and visualization aspects of the study. AW contributed to data curation and formal analysis of the study. XT contributed to investigation and formal analysis of the study. GL contributed to the validation and methodology aspects of the study. MP spearheaded the project, was responsible for funding acquisition, supervision, and project administration.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: This manuscript has not been published or presented elsewhere in full, but a preliminary abstract of this work was presented at Nursing Informatics 2024.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Peking Union Medical College 2023 Medical Education Scholar Program, Non-Profit Central Research Institute Fund of Chinese Academy of Medical Sciences, (grant number 2023zlgc0711, 2023-RC320-01).

Availability of data and material

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Supplemental material

Supplemental material for this article is available online.

References

Organization

WH.

Global report on hypertension: the race against a silent killer. World Health Organization; 2023.

Zhou

Perel

Mensah

, et al. Global epidemiology, health burden and effective interventions for elevated blood pressure and hypertension. Nat Rev Cardiol 2021; 18: 785–802.

Egan

Stevens-Fabry

. Prehypertension—prevalence, health risks, and management strategies. Nat Rev Cardiol 2015; 12: 289–300.

Ji-Guang

. Chinese Guidelines for the prevention and treatment of hypertension (2024 revision). J Geriatric Cardiol: JGC 2025; 22: 1.

Krishnamoorthy

Nagarajan

Murali

. Effectiveness of multiple combined lifestyle interventions in reducing blood pressure among patients with prehypertension and hypertension: a network meta-analysis. J Public Health 2022; 45: e319–ee31.

Williamson

Foster

Reid

, et al. Will exercise advice be sufficient for treatment of young adults with prehypertension and hypertension? A systematic review and meta-analysis. Hypertension 2016; 68: 78–87.

Naci

Salcher-Konrad

Dias

, et al. How does exercise treatment compare with antihypertensive medications? A network meta-analysis of 391 randomised controlled trials assessing exercise and medication effects on systolic blood pressure. Br J Sports Med 2019; 53: 859–869.

Olson

Vaux-Bjerke

Quam

, et al. Physical activity guidelines for Americans. NADAR! SWIMMING MAGAZINE-Periódico científico em esportes e fitness aquático-natação, pólo aquático, nado sincronizado, saltos ornamentais, travessias aquáticas 2023; 3: 166.

Murray

Brennan

French

, et al. Effectiveness of physical activity interventions in achieving behaviour change maintenance in young and middle aged adults: a systematic review and meta-analysis. Soc Sci Med 2017; 192: 125–133.

10.

Chaiken

Trope

. Dual-process theories in social psychology. London/England: Guilford Press, 1999.

11.

Verplanken

Wood

. Interventions to break and create consumer habits. J Public Policy Marketing 2006; 25: 90–103.

12.

Mendoza-Vasconez

Badii

Becerra

, et al. Forming habits, overcoming obstacles, and setting realistic goals: a qualitative study of physical activity maintenance among Latinas. Int J Behav Med 2022; 29: 334–345.

13.

Schwarzer

. Health action process approach (HAPA) as a theoretical framework to understand behavior change. Actualidades en Psicología 2016; 30: 119–130.

14.

Chen

Shao

Zhang

, et al. Development of an individualized WeChat mini program-based intervention to increase adherence to dietary recommendations applying the behaviour change wheel among individuals with metabolic syndrome. Ann Med 2023; 55: 2267587.

15.

Alshagrawi

Abidi

. Efficacy of an mHealth behavior change intervention for promoting physical activity in the workplace: randomized controlled trial. J Med Internet Res 2023; 25: e44108.

16.

Statista. Number of smartphone users in China from 2018 to 2022 with a forecast until 2027. 2023. https://www.statista.com/statistics/467160/forecast-of-smartphone-users-in-china/ .

17.

Chen

Sun

Cen

, et al. Characteristics and requirements of hypertensive patients willing to use digital health tools in the Chinese community: a multicentre cross-sectional survey. BMC Public Health 2020; 20: 1333.

18.

Clifford

Tunis

Ariyo

, et al. Trends and gaps in digital precision hypertension management: scoping review. J Med Internet Res 2025; 27: e59841.

19.

Griffin

Khairat

Bailey

, et al. A chatbot for hypertension self-management support: user-centered design, development, and usability testing. JAMIA Open 2023; 6: ooad073.

20.

Aggarwal

Tam

, et al. Artificial intelligence–based chatbots for promoting health behavioral changes: systematic review. J Med Internet Res 2023; 25: e40789.

21.

Caldarini

Jaf

McGarry

. A literature survey of recent advances in chatbots. Information 2022; 13: 41.

22.

Yang

Tan

, et al. Large language models in health care: development, applications, and challenges. Health Care Sci 2023; 2: 255–263.

23.

Luo

Aguilera

Lyles

, et al. Promoting physical activity through conversational agents: mixed methods systematic review. J Med Internet Res 2021; 23: e25486.

24.

Martinengo

Jabir

Goh

WWT

, et al. Conversational agents in health care: scoping review of their behavior change techniques and underpinning theory. J Med Internet Res 2022; 24: e39243.

25.

Kramer

Künzler

Mishra

, et al. Which components of a smartphone walking app help users to reach personalized step goals? Results from an optimization trial. Ann Behav Med 2020; 54: 1–11.

26.

Jörke

Sapkota

Warkenthien

, et al. GPTCoach: Towards LLM-based physical activity coaching. In: Proceedings of the 2025 CHI conference on human factors in computing systems, Yokohama, Japan, 25 April–1 May 2025, pp.1–46. New York, NY: Association for Computing Machinery.

27.

Zhang

, et al. Prehypertension and clustering of cardiovascular risk factors among adults in suburban Beijing, China. J Epidemiol 2011; 21: 440–446.

28.

Fortunati

Manganelli

Law

, et al. The mobile phone use in Mainland China: some insights from an exploratory study in Beijing. Telemat Inform 2010; 27: 404–417.

29.

Pei

, et al. Systematic development of an AI chatbot for physical activity habit formation in prehypertension individual integrating needs, theories, and evidence. Stud Health Technol Inform 2024; 315: 581–582.

30.

Song

Bliokas

, et al. A clinician-led, experience-based co-design approach for developing mHealth services to support the patient self-management of chronic conditions: development study and design case. JMIR Mhealth Uhealth 2021; 9: e20650.

31.

Braun

Clarke

. Using thematic analysis in psychology. Qual Res Psychol 2006; 3: 77–101.

32.

Wang

Pei

, et al. Effects of habit formation interventions on physical activity habit strength: meta-analysis and meta-regression. Int J Behav Nutrition Phys Activity 2023; 20: 09.

33.

Michie

Atkins

West

. The behaviour change wheel. A guide to designing interventions. Vol. 1003. 1st ed. Great Britain: Silverback Publishing, 2014. p.1010.

34.

Bartholomew

Parcel

Kok

. Intervention mapping: a process for developing theory and evidence-based health education programs. Health Educ Behav 1998; 25: 545–563.

35.

Gardner

Lally

. Modelling habit formation and its determinants. Psychol Habit: Theory, Mech, Change, Contexts 2018: 207–229.

36.

Wang

Xian

Guo

, et al. A novel blood pressure monitoring technique by smart HUAWEI WATCH: a validation study according to the ANSI/AAMI/ISO 81060-2: 2018 guidelines. Front Cardiovasc Med 2022; 9: 923655.

37.

Anand

Nussbaum

Duderstadt

, et al. Gpt4all: Training an assistant-style chatbot with large scale data distillation from gpt-3.5-turbo. GitHub 2023.

38.

Borsci

Malizia

Schmettow

, et al. The chatbot usability scale: the design and pilot of a usability scale for interaction with AI-based conversational agents. Pers Ubiquitous Comput 2022; 26: 95–119.

39.

Hagger

Cameron

Hamilton

, et al. (eds) The handbook of behavior change. United Kingdom: Cambridge University Press, 2020, pp.267–442.

40.

Vogelsang

Hinrichs

Fleig

, et al. Study protocol for the description and evaluation of the “Habit Coach” – a longitudinal multicenter mHealth intervention for healthy habit formation in health care professionals. BMC Public Health 2022; 22: 1672.

41.

Kókai

Ó Ceallaigh

Wijtzes

, et al. Moving from intention to behaviour: a randomised controlled trial protocol for an app-based physical activity intervention (i2be). BMJ Open 2022; 12: e053711.

42.

Rhodes

. Translating physical activity intentions into behavior: reflective, regulatory, and reflexive processes. Exerc Sport Sci Rev 2024; 52: 13–22.

43.

Verplanken

Roy

. Empowering interventions to promote sustainable lifestyles: testing the habit discontinuity hypothesis in a field experiment. J Environ Psychol 2016; 45: 127–134.

44.

Domin

Uslu

Schulz

, et al. A theory-informed, personalized mHealth intervention for adolescents (Mobile app for physical activity): development and pilot study. JMIR Form Res 2022; 6: e35118.

45.

Cherubini

Villalobos-Zuñiga

Boldi

, et al. The unexpected downside of paying or sending messages to people to make them walk: comparing tangible rewards and motivational messages to improve physical activity. ACM Trans Comp-Human Interact (TOCHI) 2020; 27: 1–44.

46.

Haggar

Whitmarsh

Skippon

. Habit discontinuity and student travel mode choice. Transp Res Part F: Traffic Psychol Behav 2019; 64: 1–13.

47.

Cheng

Jiang

. How do AI-driven chatbots impact user experience? Examining gratifications, perceived privacy risk, satisfaction, loyalty, and continued use. J Broadcast Electron Media 2020; 64: 592–614.

48.

Zhang

Lange

, et al. Artificial intelligence chatbot behavior change model for designing artificial intelligence chatbots to promote physical activity and a healthy diet. J Med Internet Res 2020; 22: e22845.

49.

Biddle

SJH

. Barriers to physical activity: time to change? A preventive medicine golden jubilee editorial. Prev Med 2022; 163: 107193.

50.

Griffin

Khairat

Bailey

, et al. A chatbot for hypertension self-management support: user-centered design, development, and usability testing. JAMIA Open 2023; 6: ooad073.

51.

Pricilla

Lestari

Dharma

D (eds)

Designing interaction for chatbot-based conversational commerce with user-centered design. In:

2018 5th International Conference on Advanced Informatics: Concept Theory and Applications (ICAICTA). Thailand: IEEE, 2018, pp.244–249.

52.

Luo

Lau

, et al. A critical review of state-of-the-art chatbot designs and applications. Wiley Interdisc Rev Data Mining Knowledge Disc 2022; 12: e1434.

53.

Hauser-Ulrich

Künzli

Meier-Peterhans

, et al. A smartphone-based health care chatbot to promote self-management of chronic pain (SELMA): pilot randomized controlled trial. JMIR Mhealth Uhealth 2020; 8: e15806.

54.

Maeng

Lee

. Designing a chatbot for survivors of sexual violence: exploratory study for hybrid approach combining rule-based chatbot and ML-based chatbot. In: Proceedings of the Asian CHI symposium 2021. Yokohama, Japan: Association for Computing Machinery, 2021, pp.160–166.

55.

Ayers

Poliak

Dredze

, et al. Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social Media forum. JAMA Intern Med 2023; 183: 589–596.

56.

Chung

Park

. Chatbot-based heathcare service with a knowledge base for cloud computing. Cluster Comput 2019; 22: 1925–1937.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.03 MB

0.01 MB

0.02 MB

0.37 MB

0.02 MB

0.12 MB

2.03 MB

0.01 MB