Table 3.
Summary of chatbot and intervention characteristics
| No. | First author/published year/Country | Intervention duration and frequency 1 | Chatbot only or multiple components intervention 2 | Overall technological features of the intervention 3 | Chatbot characteristics | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Chatbot identity (name, gender) 4 | Media 5 | User input 6 | Chatbot initiation/ User initiation7 | Relational capacity 8 | Persuasion capacity 9 | Safety 10 | Ethics discussion 11 | |||||
| Randomized controlled trials | ||||||||||||
| 1 | Kramer J/a2020/Switzerland [31] | 6 weeks and daily | Chatbot only | The Assistant to Lift your Level of activitY (Ally) app was developed using the MobileCoach platform, on both Android and iOS systems | Name: Ally Gender: Woman | Texts, graphs | Constrained | Daily and weekly messages delivered at a random time between 10 AM and 6 PM/NR | Using personalized greeting, initiating daily conversations, and occasionally sending unrelated messages to keep user interests | Setting personalized goals, using action-planning, coping-planning, and self-monitoring prompts | NR | NR |
| 2 | Kunzler F/b2019/Switzerland [33] | 6 weeks and daily | Chatbot only | Ally app (available on both iOS and Android) is based on the MobileCoach platform. Physical activity was measured by subscribing to CoreMotion Activity Manager on iOS and Google Activity Recognition API | Name: Ally Gender: Woman | Texts, graphs | Constrained | Daily messages delivered 3 times a day (between 8 and 10 AM, 10-6 PM, or 8 PM) and weekly messages delivered at random times/NR | Initiating daily conversations with a greeting | Using persuasive prompts such as goal setting, self-monitoring, goal achievement, and weekly planning | NR | NR |
| 3 | Piao M/2020/ South Korea [32] | 12 weeks and daily | Chatbot only | The Healthy Lifestyle Coaching Chatbot was developed using the Watson Conversational tool (IBM Corp) and was linked to the KakaoTalk Smart Chat application programming interface (API) through the RESTful API. It was deployed through KakaoTalk messenger app |
Name: Chatbot Gender: NR |
Texts, Images | Unconstrained | Daily message delivered at participants’ specified time/On-demand | Sending personalized goal-related messages based on their daily routines, send a compliment message providing positive feedback | Setting personalized goals, providing extrinsic (e.g., financial incentives) and intrinsic (e.g., pleasure, satisfaction) were used. | NR | NR |
| 4 | Carfora V/2019/Italy [34] | 2 weeks and daily | Chatbot only | Chatbot was deployed through Facebook Messenger |
Name: NR Gender: NR |
Texts | Constrained | Daily messages delivered at 7:30 AM/ NR | NR | Informing participants about the health and environmental impact of excessive RPMC; Using persuasive message appeal (i.e., emotional appeal such as regret) | NR | NR |
| Non-randomized studies | ||||||||||||
| 5 | Maher CA/2020/Australia [20] | 12 weeks and weekly | Multicomponent | The Paola chatbot was developed using the IBM Watson Virtual Assistant AI software and was hosted on the cloud-based instant messaging platform Slack. The program also used the MedLipal website and the Garmin Vivofit4 physical activity monitor | Name: Paola Gender: Woman | Texts | Unconstrained | Weekly check-in messages/On-demand | Referring to users by their first name and responding to questions at any time. | Assisting participants to set a personal daily step count goal based on age-based normative values + 2000 steps. This daily step goal was revisited and edited at each weekly check-in. | NR | NR |
| 6 | Fadhil A/2019/NR [35] | 3 weeks and daily | Chatbot only | Chatbot used the coaching portal combined with the dialogue engine which consists of rails state machine, user clustering model and fitbit wearable data. Chatbot was deployed through Telegram Messenger |
Name: CoachAI Gender: NR |
Texts, Images | Constrained | Messages delivered recurrently (e.g., daily, weekly, on weekends, or weekdays) or at a scheduled date and time/ NR | Using greeting and some preliminary evaluation chat at the beginning of the interaction and assessing users’ comfort in discussing personal information with the agent. | Providing motivational messages consisted of positive reinforcement and a feedback on user’s overall adherence for the week to increase adherence | NR | NR |
| 7 | Stephens TN/ 2019/ U.S. [21] | 10–12 weeks NR | Chatbot only | The Tess chatbot was deployed through multiple channels (i.e., SMS text message, Slack, WhatsApp or Facebook Messenger) and also was integrated with Google Home and Amazon Alexa |
Name: Tess Gender: NR |
Texts, Voice | Unconstrained | NR/On-demand | Mimicking empathy and compassion, adjusting the conversational style or modality to address each client’s needs, and responding based on the individual’s reported emotion or concern. | Specific goals and targeted behaviors were entered to the system to offer individualized conversations. Delivered customized integrative support, psychoeducation, and interventions | Data processing and storage are on secure servers that satisfy Health Insurance Portability and Accountability Act (HIPAA) regulations and within the country of residence for all participants given access. | NR |
| 8 | Casas J/2018/Switzerland [36] | Seven days and daily | Chatbot only | The Rupert chatbot was developed using the Chatfuel service, and deployed through Facebook Messenger |
Name: Rupert le nutritionniste Gender: NR |
Texts, graphs, images, videos | Constrained | Daily messages/NR | Showing empathy, being friendly, positive, and not judgmental; speaking in users’ native language (i.e., French) | Asking participants to choose their goals, monitoring food intake, answering questions, and giving recommendations | NR | NR |
| 9 | Kocielnik R/ 2018/ U.S. [37] | 2 weeks and daily | Chatbot only | Twillio API was used to communicate with users via mobile phones through SMS/MMS. Fitbit API was queried for user activity data to generate activity graphs. LUIS API offered automated recognition of free-text user responses |
Name: Reflection Companion Gender: NR |
Texts, Graphs | Unconstrained | Daily messages/NR | Personalized the experience by introducing questions that referenced users’ own behavior change goals | The mini-dialogues were delivered with a graph showing user’s physical activity metrics | NR | NR |
Studies a and b employed the same chatbot named Ally.
NR not reported.
1 Intervention duration is how long the intervention lasted and frequency is how often the programed intervened with the participants
2 Multicomponent means the intervention had multiple intervention components (e.g., in-person and using chatbots); chatbot only means the intervention was sorely delivered by the chatbot
3 Document the technological infrastructure, platform, and features of the intervention
4 Chatbot identity documents identity cues the chatbot is designed with. The cues can include name, gender, age, etc.
5 Media documents the types of media that the chatbot can use to deliver information
6 User inputs document the capacity of which participants can interact with the chatbot. Constrained means users can only select pre-programmed responses in the chat; unconstrained means users can freely type or speak to the chatbot
7 Chatbot/User initiation indicates whether and how often chatbot/user initiated the conversation
8 Relational capacity documents conversation strategies the chatbot can use to establish, maintain, or enhance social relationships with the participants (e.g., greetings)
9 Persuasion capacity documents conversation strategies the chatbot can use to change participant’s behaviors and behavioral determinants (e.g., knowledge, attitudes, norm perceptions, efficacy, etc.)
10 Safety documents strategies the chatbot is designed to ensure safety of the participants
11 Ethics discussion documents any ethical principles or standards the chatbot is designed with. Key ethical considerations include having transparency and user trust, protecting user privacy, and minimizing biases