Building Capacity and Promoting Smoking Cessation in the Community Via "Quit to Win" Contest 2025: Real-time Smoking Cessation Instant Messaging Support Using a Large Language Model (LLM)-Based Chatbot

Although smoking prevalence is decreasing in Hong Kong (1982: 23.3%; 2023: 9.1%), it accounts for over 7,000 deaths per year and a large amount of medical cost, long-term care and productivity loss of US$ 688 million (0.6% Hong Kong GDP). Quitting is difficult because nicotine is highly addictive. Long-term habitual tobacco smoking could foster a series of physical and psychological dependence on nicotine, and thus induce cravings and nicotine withdrawal symptoms when remaining abstinent. Tradition "one-intervention-for-all" approach cannot work optimally for overall smoking population because of the individual differences in the background characteristic and variations in response to the intervention. Intervention approaches that account for personalization and variation should be explored.

Preliminary studies suggest that AI-based chatbots can deliver structured counseling to support tobacco cessation through personalised, empathetic, and authentic conversations, thus enhancing the effectiveness of smoking cessation interventions. When integrated into social media, AI chatbots can provide timely, targeted responses and connect users with a resources on widely used platforms. A 2023 meta-analysis involving 58,796 participants further highlighted the promise of chatbots for tobacco cessation (RR=1.29, 95%CI 1.13-1.46).

In late 2022, the release of ChatGPT revolutionized AI and large language models by offering unprecedented reasoning and conversational capabilities, which enables the development of more sophisticated, human-like chatbots. Although ChatGPT (and the GenAI in general) was trained as a general-purpose virtual assistant, its tasks-specific performance can be significantly enhanced through prompt engineering. Leveraging this approach, we developed an LLM-based chatbot to autonomously deliver smoking cessation support via WhatsApp according to established protocols. Our previous rule-based chatbot effectively prevented smoking relapse, and a pilot trial with an LLM-based chatbot for youth smokers (n=154) showed feasibility, achieving an 80.2% retention rate. These findings support the implementation of GenAI chatbots as promising tools for brief smoking cessation interventions in adult smokers.

Therefore, our study aims to test the effectiveness of a comprehensive intervention using brief cessation advice, personalized active referral, and chatbot-based instant messaging support compared with the control group on current smokers who join the Quit to Win Contest.