The goal of this trial is to learn if chatbot-based instant messaging works to help smoking cessation in general adult smokers. It will also learn about the experience, attitude, and perception of using an LLM-based chatbot. The main questions it aims to answer are: 1. Will an engagement-focused LLM-based chatbot smoking cessation intervention have a non-inferior validated abstinence rate than the control group? 2. Will an LLM-based chatbot smoking cessation intervention have a non-inferior self-reported abstinence rate, smoking reduction rate, and smoking cessation services use rate than the control group? Researchers will compare an LLM-based chatbot smoking-cessation intervention to a human-led instant messaging support group (brief advice based on AWARD and personalised active referral) to determine whether chatbot-based instant messaging support promotes smoking cessation. Participants in the intervention group will receive: 1. AWARD advice 2. Personalised active referral 3. 12 weeks of chatbot-based instant messaging support (via WhatsApp)
Although smoking prevalence in Hong Kong has declined to 9.1% in 2023, achieving the government's target of 7.8% by 2025 remains a major public health challenge. Unassisted "cold turkey" quitting has a long-term success rate of less than 5%, whereas evidence-based behavioural and pharmacological interventions can raise success rates to approximately 20% or higher. However, existing cessation services in Hong Kong face a critical utilisation gap: only 17.5% of smokers have engaged with professional services, and merely 23% have used nicotine replacement therapy. This underutilisation suggests that traditional human-resource-intensive models may lack accessibility, scalability, and local appeal. Generative AI, particularly large language models, offers a transformative solution by delivering consistent, scalable, and personalised support. In the 2025 "Quit to Win" round, investigators integrated an LLM-based chatbot via WhatsApp and received positive qualitative feedback. Yet quantitative analysis revealed a sharp decline in engagement, with weekly participation dropping from 32% in week 1 to 14% by week 12, indicating that conversational ability alone does not guarantee sustained user commitment. To address this implementation gap, investigators have developed an engagement-focused GenAI companion that incorporates structured onboarding, context-aware personalisation, multimodal (text/audio) input, empathetic support, habit-aligned reminders, localised humour, and gamified features such as success stories and knowledge quizzes. Therefore, the current study aims to test, via a two-arm non-inferiority randomised controlled trial, the effectiveness of a comprehensive intervention combining brief cessation advice (AWARD), personalised active referral, and this engagement-enhanced GenAI chatbot support compared with human-led instant messaging counselling among current smokers who join the Quit to Win Contest across all 18 districts of Hong Kong.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
HEALTH_SERVICES_RESEARCH
Masking
SINGLE
Enrollment
998
A brief (30-60 seconds) face-to-face or remote smoking cessation advice delivered using the validated AWARD model: Ask about smoking history; Warn about high health risks (accompanied by a health warning leaflet); Advise quitting as soon as possible and setting a quit date (to qualify for contest prizes); Refer to smoking cessation services using a referral card; Do it again - repeat the intervention at each follow-up, encouraging re-quitting after relapse or relapse prevention after success.
A two-sided, colour-printed A4 leaflet covering: (1) absolute risk of death from smoking; (2) full list of diseases caused by active and second-hand smoking; (3) ten pictorial warnings of health consequences on one page for maximum impact; (4) benefits of smoking cessation; and (5) simple encouraging messages to quit.
A three-folded card containing brief information and highlights of existing smoking cessation services in Hong Kong, contact methods, motivational messages, and strong supporting slogans.
A generic booklet provided covering: benefits of quitting, smoking-related diseases, methods to quit, how to handle withdrawal symptoms, a quitting declaration, and other practical tips.
Smokers will be introduced to various SC services in Hong Kong (via the referral card) and motivated to use them. Well-trained SC ambassadors will assist smokers in choosing their favourite or most convenient type of service. Research staff will assist participants in booking or re-booking the SC services at the 1- and 2-month follow-ups (after very brief questionnaire surveys). Participants' contact information will be forwarded to SC service providers within 7 days, and providers are expected to contact participants within 1-2 weeks. Research staff will also monitor participants' use of SC services at each follow-up (1-, 2-, 3-, and 6-month) and, at the 1- and 2-month follow-ups, assist participants in booking or rebooking appointments if necessary. Investigators shall liaise with existing service providers and seek their assistance in promptly supporting our smokers.
Participants in the intervention group will receive 12 weeks of instant messaging support delivered by an LLM-based chatbot (GPT-4o or newer) on WhatsApp, supporting text and audio input. Using prompt engineering, agent techniques, and Retrieval-Augmented Generation, the chatbot delivers theory-based 5As/5Rs-structured interventions alongside freeform, on-demand support, with engagement features including personalisation, proactive check-ins, and interactive Quick Commands.
Participants in the control group will receive 12 weeks of instant messaging support delivered by a trained human counsellor via WhatsApp. Using the same theoretical frameworks as the chatbot intervention, the counsellor will provide real-time behavioural and psychosocial support grounded in the 5As/5Rs models, Motivational Interviewing (MI), and evidence-based Behaviour Change Techniques (BCTs). The support will be personalised according to each participant's sociodemographic characteristics, smoking patterns, quit intentions, and plans.
WhatsApp messages on follow-up survey reminders.
Hong Kong Council on Smoking and Health (COSH)
Hong Kong, Hong Kong, Hong Kong
Biochemically validated abstinence
Defined as exhaled CO level \<4ppm and saliva cotinine level ≤30 ng/ml
Time frame: 6-month follow-up
Biochemically validated abstinence
Defined as exhaled CO level \<4ppm and saliva cotinine level ≤30 ng/ml
Time frame: 3-month follow-up
Self-reported 7-day point prevalence abstinence
Smokers who did not smoke even a puff in the 7 days preceding the follow-up
Time frame: 3- and 6-month follow-ups
Self-reported reduction
Defined by at least 50% reduction in baseline daily number of cigarettes
Time frame: 1-, 2-, 3- and 6-month follow-ups
Self-reported use of smoking cessation service
Use of smoking cessation service at 1-, 2-, 3- and 6-month follow-ups.
Time frame: 1-, 2-, 3- and 6-month follow-ups
Prolonged abstinence
Abstinence from smoking for 3 consecutive months at 3-month follow-up, or for 6 consecutive months at 6-month follow-up
Time frame: 3-month and 6-month follow-ups
Quit attempt
Abstinence for at least 24 hours
Time frame: 1-, 2-, 3-, and 6-month follow-ups
Post-cessation weight change
Self-reported change in body weight (in kilograms) from baseline to follow-up
Time frame: 6-month follow-up
Self-reported mental health conditions
Patient Health Questionnaire-4 (PHQ-4): The PHQ-4 is a 4-item ultra-brief screening tool for anxiety and depression that combines the GAD-2 and PHQ-2 subscales. Each item is scored 0-3, with a total score of 0-12; subscale scores of 3 or higher indicate positive screening and warrant further clinical assessment.
Time frame: Baseline and 6-month follow-up
Self-reported smoking-related health conditions
Answer "Yes" to experiencing any smoking-related health condition during smoking cessation or reduction
Time frame: Baseline and 6-month follow-up
Chatbot user experience
Chatbot Usability Scale, or the 11-item Bot Usability Scale (BUS), is a validated questionnaire that evaluates chatbot usability across five dimensions (accessibility, function quality, conversation/information quality, privacy/security, and response time) using a 5-point Likert scale. The total score (11-55) is the sum of all items; a higher total score indicates better overall usability and greater user satisfaction. Higher scores on individual dimensions similarly reflect superior performance in those areas.
Time frame: 3-month follow-up
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.