Cigarette smoking continues to be the leading preventable cause of disease and death globally and in the United States [1]. Cigarette smoking kills more than 480,000 Americans each year and results in more than US $240 billion in health care spending and nearly US $372 billion in lost productivity [2]. Thus, promoting smoking behavior change with the long-term goal of achieving successful cessation is critical to reducing the public health impact caused by tobacco use. Within the United States, the past few decades have seen a groundswell of evidence-based interventions to quit smoking. These have included a wide range of strategies such as mobile health (mHealth) apps [3,4], availability and increased access to effective pharmacotherapy for cessation [5], SMS text messaging support [6,7], evidence-based behavioral counseling [8], and comprehensive state-based programs, such as tobacco quitlines [9,10]. However, despite their widespread availability, less than one-third of people who smoke cigarettes actively avail themselves to cessation strategies [11]. For example, tobacco quitlines—free evidence-based smoking cessation programs providing telephone-based behavioral counseling combined with cessation pharmacotherapy available in all 50 US states—are used by less than 3% of people who smoke [12]. One potential reason for the low uptake of existing services for cessation may be that strategies to promote cessation differ according to an individual’s readiness and motivation to quit [13]. Although smokers actively committed to quitting respond positively and engage with cessation services, ambivalent smokers (ie, those not actively seeking treatment) may benefit from approaches that promote modest changes or “behavioral nudges” toward smoking behavior change (but not cessation) [14,15]. Most cessation research is focused on those expressing an intention to quit, and research on promoting behavior change in ambivalent smokers (not actively thinking about quitting) is scant.

A conversational agent or “chatbot” is an inexpensive and scalable approach to deliver personalized health information in a human-like manner. Common in the consumer-service sector, chatbots are increasingly being deployed in the health sector to provide education, support, service navigation, and interventions for a range of health issues, from COVID-19 to asthma [16,17]. Human-delivered interventions for smoking behavior change are effective but are resource-intensive, lack scalability, and can be expensive to implement. Chatbots have the intuitive appeal of being low-cost automated interventions that can widely reach ambivalent smokers. Moreover, chatbots can provide “virtual accompaniment” at the frequency and duration preferred by the smoker until the motivation to quit arises, at which time they can be swiftly connected to an existing evidence-based cessation intervention. A few studies on chatbots for smoking behavior change target participants actively intending to quit smoking [18-20] and, while important, only address the tip of the iceberg in terms of reaching the smoking population. Since motivation to quit smoking is dynamic and occurs along a continuum [21], strategies to engage ambivalent smokers may be most helpful if they are tailored to accommodate smokers’ changing preferences and are responsive to rapid changes in motivation (eg, connecting smokers to quitlines when ready to quit), while focusing on evidence-based strategies that have been effective for smoking reduction leading to cessation (eg, quit tips, self-monitoring, and tracking savings). Since interventions to assist smokers ambivalent about quitting are in their infancy, little is known about key elements of a chatbot-based smoking behavior change intervention tailored for this target population.

To overcome these gaps in the field, the goal of our study is to test a proof-of-concept of a novel chatbot to support and engage smokers who are not actively engaged in a quit attempt (ambivalent smokers). We propose accomplishing our aims in 2 phases. In phase 1, we will identify the key attributes that will guide the development of the chatbot prototype. We will survey ambivalent smokers to determine their preferences and priorities for a smoking cessation support chatbot. Using conjoint analysis, we will systematically elicit participant preferences for the proposed chatbot prototype (see details below). Grounded in macroeconomic principles [22], conjoint analysis is a quasi-experimental approach that decomposes a product or service into its key attributes, then poses the attributes to patients to understand patient-determined values for each attribute [23,24]. These preferences will be programmed into the chatbot prototype. In phase 2, we will invite participants to use the prototype over a 2-week period to assess the feasibility and acceptability of the developed chatbot prototype.

The study was funded in June 2022, and regulatory processes and approval were completed by August 2022. Data collection for the study is currently ongoing, and we anticipate publication of expected outcomes by December of 2023. In phase 1, the methodology described above will result in the identification of key attributes that participants report a preference for in a smoking cessation support chatbot. These data will be used to program a chatbot prototype. Using SmartBot360 [39], a “point-and-click” chatbot development software, a prototype smoking cessation support chatbot will be programmed that contains the top preferences and features as expressed by the participants in phase 1. In addition to the attributes, information on existing evidence-based smoking cessation resources will also be programed to gauge participant preferences for existing evidence-based cessation resources. These will include, but not be limited to, links to sign up for free SMS text messaging services (eg, smokefree TXT) or mHealth quit apps (eg, QuitGuide), connecting to smoking cessation experts (eg, National Cancer Institute’s LiveHelp or a state quitline), and information on pharmacotherapy.

At the end of phase 2, we will obtain participant data on chatbot acceptability and feasibility.

The goal of our study is to develop and pilot-test for feasibility and acceptability of a novel chatbot prototype to support and engage smokers who are not actively engaged in a quit attempt (ambivalent smokers). The study will test a proof-of-concept of the chatbot, and the anticipated main result of our project is a pilot-tested conversational agent ready to be tested for efficacy in a larger study. To our knowledge, this is the first study to develop a conversational agent (chatbot) for ambivalent smokers using a novel methodology.

Interventions to engage individuals ambivalent about quitting smoking are an unexplored area in smoking cessation research [11]. This study will begin an innovative line of research on strategies to optimize and personalize smoking behavior change services for individuals ambivalent about quitting smoking. Of specific mention is the use of conjoint analysis as an innovative strategy to assess and incorporate user preferences in the development of an mHealth intervention. Methodologies to systematically assess and incorporate individual preferences in the delivery of health care services are in their infancy [40], and little is known about the preferences of ambivalent smokers for chatbots as they navigate the motivation to quit continuum. While conjoint analysis has been increasingly used in the health sector, including in the smoking cessation field [41,42], to our knowledge, no mHealth interventions for smoking behavior change have used this methodology for design and implementation. Our investigative team has successfully used conjoint analysis for other health issues, including determining consumer preferences for oral and topical HIV chemoprophylaxis [43,44]. If acceptable in this pilot study, our chatbot could be a low-cost and low-resource intervention that fills a much-needed gap in the field.

The study has a few limitations. First, the study design does not lend itself to testing the effects of the chatbot on smoking behavior change outcomes. However, given the dearth of research in this area, the development of the chatbot and acceptability and feasibility testing are the first steps. At the end of the study, we will have a developed chatbot that will be ready for efficacy testing in a larger randomized controlled trial. Next, our participant recruitment methodology uses a web-based survey panel, which may reduce generalizability. While such survey panels use purposive sampling to gain a representative sample of participants [45], we anticipate using a combination of social media, community-based recruitment efforts, and web-based recruitment panels for larger efficacy trials, which will increase the generalizability of our results.

Our intervention that uses a chatbot to promote smoking behavior change in a sample of ambivalent smokers fills a distinct gap in the field. Our model is a scalable and pragmatic model that can be adopted by cessation services (quitlines) and is designed for easy accessibility to the broader population of individuals who are not actively contemplating quitting smoking. It could additionally serve as a tool for health care providers as they recommend quitting smoking resources to their patients. In the long run, our study design and methodology can be used to develop chatbots that can increase access to other evidence-based interventions within the context of substance use disorders (eg, cannabis and opioids) or compulsive behaviors (eg, eating disorders and gambling).