Lorien C. Abroms

UNSTRUCTURED Abstract Background: Generative artificial intelligence (GenAI) chatbots are increasingly being developed to deliver personalized, scalable, and interactive support for health behavior change. Although these technologies have considerable potential, there is little practical guidance for researchers and clinicians on how to design GenAI health coach chatbots that can function as health behavior change coaches. Purpose: This tutorial presents a practical framework for developing theory-driven GenAI health coach chatbots using widely available large language models (LLMs) without requiring programming expertise. Drawing on behavioral theory, ethics, digital intervention development, and recent advances in LLMs, the tutorial describes how to develop a chatbot that is grounded in behavior change theory, informed by evidence, and developed with attention to safety, privacy, and equity. The six-step framework presented here spans conceptualization, instruction writing, knowledge base curation, LLM integration, pre-release testing, and post-release monitoring. Tutorial Framework: The tutorial provides a historical overview of the development of health coach chatbots. It then conceptualizes how theory and ethics can inform chatbot development. Finally, it introduces a six-step framework for health coach chatbot development that guides developers through (1) conceptualizing the chatbot, (2) developing chatbot instructions with attention to theory, behavior change techniques, and ethical considerations, (3) identifying materials for the knowledge base, (4) integrating the instructions and knowledge base into an LLM, (5) conducting pre-release validation and safety testing, and (6) releasing the chatbot and implementing post-release monitoring and refinement. Practical recommendations are provided throughout, including approaches to prompt engineering, use of retrieval-augmented generation, theory-based personalization, red-teaming, and evaluation of chatbot fidelity and safety. Conclusions: By integrating behavioral theory and ethics with contemporary GenAI development practices, this tutorial provides researchers, clinicians, and digital health developers with a structured, replicable framework for building and evaluating GenAI health coach chatbots. As AI-assisted behavioral interventions become more widespread, theory-driven design and rigorous validation will be essential for developing chatbots that are effective, safe, equitable, and suitable for research and practice.

Despite strong evidence supporting behavior change interventions, their population-level impact faces a persistent research-to-practice gap. This perspective piece argues this is partly due to barriers within health systems, including a reliance on clinician-led referral pathways. We introduce electronic signposting (eSignposting) for rapid, proactive, and scalable access to behavior change interventions. It identifies eligible individuals through electronic health records and delivers automated digital signposting to behavioral support. We demonstrate implementation of eSignposting within the NHS to increase reach with minimal workload. We consider the mechanisms that influence its adoption/effectiveness, and how ecosystem mapping and implementation blueprints can support wider adoption and integration. We argue that redesigning implementation at the health-system level is needed for population-level reach.

Background: Large language model chatbots such as ChatGPT may be able to provide support to people who smoke cigarettes and are trying to quit. Objective: This pilot study examined the feasibility and acceptability of integrating a specialized ChatGPT-based chatbot, BeFreeBot, into a smoking cessation text messaging intervention, BeFree. Chatbot fidelity was also examined. Methods: Participants who smoked cigarettes in the previous 7 days were recruited from Amazon Mechanical Turk (N=23), enrolled in BeFree, and provided access to BeFreeBot. Surveys were administered at baseline and 4 weeks after enrollment to assess perceptions of BeFreeBot. Computer records of interactions between BeFreeBot and participants were also analyzed to assess participant engagement and adherence of BeFreeBot to its instructions. For the adherence analysis, transcripts were dual coded, and discrepancies were resolved by a third coder. Results: Most participants (16/23, 70%) texted BeFreeBot with questions or concerns at least once. Participants sent 14.5 (SD 23.6) texts to BeFreeBot on average. Most participants were highly satisfied with BeFreeBot (13/18, 72%) and agreed that it was helpful for quitting (11/19, 58%). They also reported that the BeFreeBot responses were clear and easy to understand (16/17, 94%) and that they trusted responses from BeFreeBot (12/17, 71%). Most participants (17/19, 90%) reported trying to quit smoking for 1 day or longer, and 30% (7/23) self-reported no smoking in the previous 7 days. An analysis of transcripts of BeFreeBot responses (n=328) revealed that BeFreeBot functioned as instructed on most measures, with clear language (328/328, 100%), follow-up questions asked of participants (13/16, 81%), and recommendations to seek out professional counseling (13/16, 81%) or consider the use of Food and Drug Administration–approved medications (eg, nicotine replacement therapy; 14/16, 88%). Responses stayed on the topic of smoking cessation counseling (324/328, 98.8%) and did not include information that contradicted the US Preventive Services Task Force guidelines (328/328, 100%). Conclusions: A specialized large language model chatbot integrated into an SMS text messaging program and accessed through SMS text message was found to be feasible and acceptable to smokers.

BACKGROUND Large language model chatbots such as ChatGPT may be able to provide support to people who smoke cigarettes and are trying to quit. OBJECTIVE This pilot study examined the feasibility and acceptability of integrating a specialized ChatGPT-based chatbot, BeFreeBot, into a smoking cessation text messaging intervention, BeFree. Chatbot fidelity was also examined. METHODS Participants who smoked cigarettes in the previous 7 days were recruited from Amazon Mechanical Turk (N=23), enrolled in BeFree, and provided access to BeFreeBot. Surveys were administered at baseline and 4 weeks after enrollment to assess perceptions of BeFreeBot. Computer records of interactions between BeFreeBot and participants were also analyzed to assess participant engagement and adherence of BeFreeBot to its instructions. For the adherence analysis, transcripts were dual coded, and discrepancies were resolved by a third coder. RESULTS Most participants (16/23, 70%) texted BeFreeBot with questions or concerns at least once. Participants sent 14.5 (SD 23.6) texts to BeFreeBot on average. Most participants were highly satisfied with BeFreeBot (13/18, 72%) and agreed that it was helpful for quitting (11/19, 58%). They also reported that the BeFreeBot responses were clear and easy to understand (16/17, 94%) and that they trusted responses from BeFreeBot (12/17, 71%). Most participants (17/19, 90%) reported trying to quit smoking for 1 day or longer, and 30% (7/23) self-reported no smoking in the previous 7 days. An analysis of transcripts of BeFreeBot responses (n=328) revealed that BeFreeBot functioned as instructed on most measures, with clear language (328/328, 100%), follow-up questions asked of participants (13/16, 81%), and recommendations to seek out professional counseling (13/16, 81%) or consider the use of Food and Drug Administration–approved medications (eg, nicotine replacement therapy; 14/16, 88%). Responses stayed on the topic of smoking cessation counseling (324/328, 98.8%) and did not include information that contradicted the US Preventive Services Task Force guidelines (328/328, 100%). CONCLUSIONS A specialized large language model chatbot integrated into an SMS text messaging program and accessed through SMS text message was found to be feasible and acceptable to smokers.

[This corrects the article DOI: 10.2196/51331.].

Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author).

Background: Large language model (LLM) artificial intelligence chatbots using generative language can offer smoking cessation information and advice. However, little is known about the reliability of the information provided to users. Objective: This study aims to examine whether 3 ChatGPT chatbots—the World Health Organization’s Sarah, BeFreeGPT, and BasicGPT—provide reliable information on how to quit smoking. Methods: A list of quit smoking queries was generated from frequent quit smoking searches on Google related to “how to quit smoking” (n=12). Each query was given to each chatbot, and responses were analyzed for their adherence to an index developed from the US Preventive Services Task Force public health guidelines for quitting smoking and counseling principles. Responses were independently coded by 2 reviewers, and differences were resolved by a third coder. Results: Across chatbots and queries, on average, chatbot responses were rated as being adherent to 57.1% of the items on the adherence index. Sarah’s adherence (72.2%) was significantly higher than BeFreeGPT (50%) and BasicGPT (47.8%; P<.001). The majority of chatbot responses had clear language (97.3%) and included a recommendation to seek out professional counseling (80.3%). About half of the responses included the recommendation to consider using nicotine replacement therapy (52.7%), the recommendation to seek out social support from friends and family (55.6%), and information on how to deal with cravings when quitting smoking (44.4%). The least common was information about considering the use of non–nicotine replacement therapy prescription drugs (14.1%). Finally, some types of misinformation were present in 22% of responses. Specific queries that were most challenging for the chatbots included queries on “how to quit smoking cold turkey,” “...with vapes,” “...with gummies,” “...with a necklace,” and “...with hypnosis.” All chatbots showed resilience to adversarial attacks that were intended to derail the conversation. Conclusions: LLM chatbots varied in their adherence to quit-smoking guidelines and counseling principles. While chatbots reliably provided some types of information, they omitted other types, as well as occasionally provided misinformation, especially for queries about less evidence-based methods of quitting. LLM chatbot instructions can be revised to compensate for these weaknesses.

Since the early studies exploring the use of SMS text messaging for health intervention, text messaging has played a pivotal role in the advancement of mobile health. As an intervention modality, text messaging has provided vital learnings for the design and delivery of interventions, particularly in low-resource settings. Despite the advances in technology over the last 25 years, text messaging is still being used in largely the same way to deliver health information, behavior change interventions, and support. The strong, consistent evidence for the benefits of this type of intervention has made text messaging a routine part of health interventions around the world. Key to its success is its simplicity, alongside the benefit of being arguably the most accessible form of consumer digital health intervention. Text message interventions are well suited for public health interventions due to their low cost, vast reach, frequent use, high read rates, and ability to be tailored and personalized. Furthermore, the nature of text messaging interventions makes them ideal for the delivery of multilingual, culturally tailored interventions, which is important in the context of increasing cultural diversity in many countries internationally. Indeed, studies assessing text message–based health interventions have shown them to be effective across sociodemographic and ethnic groups and have led to their adoption into national-level health promotion programs. With a growing focus on artificial intelligence, robotics, sensors, and other advances in digital health, there is an opportunity to integrate these technologies into text messaging programs. Simultaneously, it is essential that equity remains at the forefront for digital health researchers, developers, and implementers. Ensuring digital health solutions address inequities in health experienced across the world while taking action to maximize digital inclusion will ensure the true potential of digital health is realized. Text messaging has the potential to continue to play a pivotal role in the delivery of equitable digital health tools to communities around the world for many years to come. Further new technologies can build on the humble text message, leveraging its success to advance the field of digital health. This Viewpoint presents a retrospective of text messaging in health, drawing on the example of text message–based interventions for smoking cessation, and presents evidence for the continued relevance of this mobile health modality in 2025 and beyond.

Background: Smartphone apps are a convenient, low-cost approach to delivering smoking cessation support to large numbers of individuals. Yet, the apps are susceptible to low rates of user engagement and retention. Objective: This study aims to test the effects of a new game module (called Inner Dragon) integrated into Smoke Free (23 Limited), a leading smoking cessation app with established efficacy. The primary outcomes measured user engagement with the app. Methods: A 2-arm, parallel-group, randomized controlled trial was conducted in the United States with an 8-week follow-up. Adult individuals who smoked ≥1 cigarettes daily and planned to quit smoking within 7 days were recruited and randomized (N=500), with equal allocation. Both groups received free access to the original Smoke Free app with “core” features of its smoking cessation program (eg, a diary and craving log). The treated group received additional access to the integrated Inner Dragon game that incorporated several game mechanics designed to increase user engagement. User engagement outcomes were the number of unique app sessions, average minutes per session, days with a session, and program adherence. Self-reported and verified smoking abstinence and app satisfaction were also assessed. The main analysis estimated the intention-to-treat effect of access to Inner Dragon on each outcome. Further analyses assessed effect modification by participant characteristics and the association of intensity of game use with program adherence and abstinence. Results: Overall, user engagement was greater for treated versus control participants: they had 5.3 more sessions of Smoke Free (mean 29.6, SD 36.5 sessions vs mean 24.3, SD 37.9 sessions; P=.06), 0.8 more minutes per session (mean 6.9, SD 5.4 min vs mean 6.1, SD 5.2 min; P=.047), and 3.4 more days with a session (mean 14.3, SD 15.3 days vs mean 11.9, SD 14.3 days; P=.03). Program adherence, based on the number of times core features of the original Smoke Free app were used, was higher for treated versus control participants (mean 29.4, SD 41.3 times vs mean 22.6, SD 35.6 times; P=.03). Self-reported 7-day and 30-day point-prevalence abstinence and verified 7-day point-prevalence abstinence at 8 weeks did not significantly differ by study group. The mean repeated 1-day prevalence of quitting was higher among the treated group versus the control group (mean 17.3%, SD 25.6 vs mean 12.4%, SD 21.3; P=.01). App satisfaction and the motivation to (stay) quit did not differ by study group. Higher intensity of game use was associated with increased program adherence and self-reported abstinence. Conclusions: Findings suggest that the Inner Dragon game increased user engagement and program adherence. Additional refinements to the game design may clarify whether the game increases abstinence rates. Overall, it is feasible to deploy games and gamification to enhance user engagement in existing smoking cessation interventions. Trial Registration: ClinicalTrials.gov NCT05227027; https://clinicaltrials.gov/study/NCT05227027