Table 3 shows the distribution of studies in relation to their temporal profiles, which refer to the time-dependent characteristics of a chatbot and can be further classified according to the time horizon, frequency of interaction, duration of the exchange, and consecutiveness of interactions [21]. Our results indicated that a substantial proportion of COVID-19 chatbot interactions (26/56, 46%) is categorized under the medium term, as the interaction of these chatbots can go beyond a single conversation and may last for days and weeks. With regard to frequency of interactions, more than half of the COVID-19 chatbots (31/56, 55%) interacted with their users multiple times instead of a single occurrence [21]. Regarding interaction duration, most COVID-19 chatbots had medium-length conversations with the user (22/56, 39%).

Regarding the consecutiveness of interactions, nearly half of the chatbots (25/56, 45%) could remember information from previous user interactions, whereas the remaining chatbots (8/56, 14%) were found to perform the opposite function toward past interactions with the user.

In Table 4, we characterize the appearance of COVID-19 chatbots (ie, the chatbot’s overall look and feel) into 3 design dimensions: role, primary communication style, and avatar representation [21]. Regarding chatbot roles, our results showed that most conversational agents for COVID-19 were considered experts (33/56, 59%). Expert chatbots have specific skills or expertise in a specific field and provide answers or reactions to the user’s query, keywords, or actions [21]. In contrast, facilitator chatbots assist users in reaching a certain objective or executing a particular task [21]. In our analysis, 27% (15/56) were facilitator COVID-19 chatbots. Finally, peer chatbots aim to be a “sparring partner” for the user, and they value socioemotional behavior as part of their design. Of the 56 chatbots, only 8 (14%) COVID-19 chatbots served as peers.

Regarding the communication style of COVID-19 chatbots, we found that 63% (35/56) of them were task oriented. Task-oriented chatbots prioritize task efficiency and achievement of goals at the most minimal cost and time possible [21]. In contrast, socially oriented chatbots prioritize social engagement and personalization of interactions with their users [21]. These comprised only 34% (19/56) of COVID-19 chatbots.

Chatbots can also have a visual or character representation (embodied avatars), or they may not have one at all (disembodied avatars) [21]. We found that more than half of the COVID-19 chatbots (32/56, 57%) did not have any visual or character representation. Only 34% (19/56) of the COVID-19 chatbots had embodied avatars.

The intelligence perspective is characterized by the functionalities of a chatbot’s inner working mechanisms, which build on an intelligence framework and quotient, personality adaptability, socioemotional behavior, and service integration [21]. As depicted in Table 5, we found that a large number of COVID-19 chatbots (26/56, 46%) were artificially intelligent. Artificially intelligent chatbots use their NLP capabilities to learn from each conversation with the user [21]. Meanwhile, a high proportion of COVID-19 chatbots (25/56, 45%) could respond to users according to set rules and by understanding an individual’s textual input.

With regard to personality adaptability, more than half of the COVID-19 chatbots (31/56, 55%) have a principal self-adaptability, meaning they operate on a structured flow. In contrast, a significant number of COVID-19 chatbots (24/56, 43%) were adaptive because these chatbots adjust their responses depending on the user input [21].

Moreover, 57% (32/56) of the COVID-19 chatbots did not demonstrate socioemotional behavior toward their clients. Among the existing COVID-19 chatbots, only 36% (20/56) could express empathic reactions to the users.

With regard to service integration, there was an equal distribution of the number of COVID-19 chatbots that were limited to only processing external data and those that could process both external data and broadcast resources (21/56, 38%).

The domain encompasses the design dimensions of the users’ communication with the COVID-19 chatbot. As depicted in Table 6, of the 56 COVID-19 chatbots, 38 (68%) were text based, where the end user and the bot interacted and communicated through text or internet messaging. The interactive modality was the most predominant (39/56, 70%), where end users can enter free texts to communicate with a chatbot [21].

Regarding user assistance design, more than half of the COVID-19 chatbots (31/59, 54%) were reactive, meaning they only reacted to user input. Concerning personalization, most COVID-19 chatbots (36/56, 64%) were adaptive, meaning they could change their meaning and adjust their responses based on user needs and circumstances [21]. Moreover, most COVID-19 chatbots did not receive any additional human support (44/56, 79%) or have any gaming elements (42/56, 75%).

Context refers to the motivation of users to interact with the chatbot, and it is reflected in the application domain, the user motivation to engage with the chatbot, and the collaboration goal of the chatbot. As depicted in Table 7, the vast majority of COVID-19 chatbots were in the health care domain (46/56, 82%). Consequently, we found 20% (11/56) of the COVID-19 chatbots in the education domain, and there was a smaller percentage of chatbots designed activities of daily living (6/56, 11%). Among all the COVID-19 chatbots, 11% (6/56) fell under multiple application domains.

We also found that most COVID-19 chatbots (35/56, 63%) were purposely designed to provide information to the user. Our results also showed that there was an almost equal distribution of COVID-19 chatbots designed for utility (18/56, 32%) and coaching purposes (15/56, 27%). However, some COVID-19 chatbot researchers designed their conversational agents for productivity (6/56, 11%) and entertainment (4/56, 7%) purposes. It is also noteworthy that several COVID-19 chatbots (16/56, 29%) were designed for multiple purposes.

Finally, all the COVID-19 chatbots were goal oriented, which means that the chatbots helped users achieve a predefined goal [78].

A recent scoping review pooled the existing conversational agents combating the negative effects of COVID-19 and classified them based on their applications [6]. In this study, we updated the number of existing COVID-19 chatbots and supplemented the current literature with the design trends of existing COVID-19 conversational agents. We adopted and modified the chatbot design taxonomy of Nißen et al [21] and classified each COVID-19 chatbot based on 5 design perspectives.

From the temporal profile perspective, nearly half of the COVID-19 chatbots (26/56, 46%) interacted with the user for several weeks (medium term). This implies that most COVID-19 chatbots did not establish a deep, long-lasting relationship and companionship with the user, which is observed in chatbots with a long-term time horizon [21]. COOPERA (COVID-19: Operation for Personalized Empowerment to Render Smart Prevention and Care Seeking), for example, provides individualized self-care support, real-time follow-up of the users’ health condition, and the epidemiological situation in the users’ area for up to several weeks [31].

Moreover, COVID-19 chatbots tended to interact with their users for >1 time, which was seen in 55% (31/56) of these chatbots. For instance, Indonesian researchers deployed ARIGAselfCareNursingBot, a teleassessment chatbot, for 5 months in a wound care clinic [56]. Furthermore, 45% (25/56) of COVID-19 chatbots could remember information from previous user interactions. COOPERA, as discussed previously, could monitor users’ health conditions daily based on their previous answers about their physical and health conditions [31].

However, it is notable that a significant proportion of COVID-19 chatbot papers did not explicitly report the time horizon (17/56, 30%), frequency (17/56, 30%), duration (20/56, 36%), and consecutiveness of interactions (23/56, 41%) of their chatbot. This may show that there is a need for researchers and computer programmers to have a deeper dive into chatbot design specifications in their respective manuscripts to help future research in the identification of the most cost-efficient design for a pandemic-responsive chatbot.

From the appearance perspective, we found that most COVID-19 chatbots (33/56, 59%) were considered experts. For example, Chloe provides Canadians with up-to-date information regarding COVID-19 by answering users’ inquiries about the pandemic [22]. In contrast, only 27% (15/56) of the COVID-19 chatbots assumed a facilitator role. An example of this is Atena, a chatbot that delivers digital mental health intervention and suggests effective coping strategies to improve the mental well-being of university students [36]. A screenshot showing a sample of conversation from the Atena chatbot is shown in Figure 2 [36].

Our research also found that most COVID-19 chatbots (35/56, 63%) were task oriented in terms of primary communication style. To illustrate, Symptoma assesses the end user’s risk of COVID-19 based on the symptoms the person inputted or answered in the textbox, without considering the user’s emotional needs [34]. This is in contrast with socially oriented chatbots, which represent only 34% (19/56) of the literature examined. An example is Bella, which demonstrates empathy and emotional expressions based on the user’s inputs to personalize every interaction [28].

Finally, our analysis showed that more than half of the existing COVID-19 chatbots (32/56, 57%) were disembodied, meaning they did not have any character representation. Chatbots with avatars could strengthen their social presence, which may reinforce a good social relationship with the users [79]. This finding may suggest that the immediate priority of designers at the height of the COVID-19 pandemic is to fulfill their functional efficiency while deprioritizing social engagement that may lengthen development cycles.

From the intelligence perspective, a significant number of COVID-19 chatbots (26/56, 46%) were artificially intelligent. One example is Drever, an educational assistant developed in Japan using an AI engine that acts as a teacher capable of answering questions from students [62]. Similarly, Real Conversation is an AI-based chatbot in India that provides COVID-19 information while assisting with customer queries [64].

Meanwhile, a high number of COVID-19 chatbots (25/56, 45%) could respond to users according to a set of rules and by understanding the textual input of the user. For example, CoronaGo is a health care chatbot embedded in a website that uses NLP to understand user intention but also has a rule-based system architecture to track the COVID-19 situation in India efficiently [66]. This means that most COVID-19 chatbots maximize their capabilities in understanding the users’ intentions in a timely manner, which is critical for a chatbot designed for a pandemic.

Most COVID-19 chatbots (31/56, 55%) operate on a structured flow, for example, Chatbot-Mental Wellness, a chatbot developed in India that provides guided activities in simple exercises, meditation, and positive affirmation [46], although nearly half (24/56, 43%) of the chatbots can adjust their responses depending on user input. An example is the JuggleChat, an educational assistant chatbot that addresses students’ specific needs dependent on their input [47].

Our analysis also showed that most COVID-19 chatbots (32/56, 57%) did not present socioemotional behavior to the users, whereas only 36% (20/56) of COVID-19 chatbots could express empathic reactions to the consumers. This suggests that expectations for chatbots to exhibit social behaviors that are customary in human-to-human communication should always be considered as there is a growing demand for humanlike conversational engagement styles [80]. Some empathic chatbots include Chasey, a character-based conversation agent for COVID-19, who responds to users with phrases such as “Don’t worry my friend, everything will be fine.” and “Get well soon, pal!” [51].

Finally, COVID-19 chatbots that are limited to only processing external data and those that can process both external data and media resources were of equal distribution (21/56, 38%). Such chatbots include Mano, an interactive chatbot that provides up-to-date COVID-19 information from reliable sources including, but not limited to, the WHO and predicts pandemic effects in the Mano River Union region [76], and Aapka Chikitsak, a medical bot that provides its users with quality information from the National Health Portal and engaging media as a means to deliver telehealth services in India during the pandemic [70].

From the interaction perspective, most COVID-19 chatbots (38/56, 68%) reach out and communicate with the user through text or internet messaging and even allow the user to respond via free text inputs. This is essential as limiting interaction options may indicate a disregard for the users’ inputs [81]. One example is Elena+, a smartphone-based and conversational agent that delivers pandemic lifestyle care interventions that communicate with the user through textual input and is currently being expanded into Spanish and Tamil languages to reach larger audiences [35]. Another example is Wakamola, a chatbot that recognizes unstructured messages and is used to encourage confined individuals to modify their lifestyle practices [37].

In addition, more than half of the COVID-19 chatbots (30/56, 54%) reacted only to user input rather than steering the conversation. This may imply that chatbot designers should aim to design constructive, goal-oriented interactions [82]. Similar to a proof-of-concept study implemented at the University of Trento in Italy, Atena was able to react to user inputs only [36], unlike the lone proactive Alia that can steer conversation [71].

Moreover, most COVID-19 chatbots (36/56, 64%) had adaptive personalization. This shows that pandemic designers concentrated on the chatbots’ capacity to adapt to users’ choices, needs, and circumstances, which is consistent with an earlier study [81]. Cory COVID-Bot, for instance, which was developed to engage with difficult-to-reach populations directly, aims to resolve public ambiguity and uncertainty regarding community health guidelines. With its sophisticated features, an algorithm dynamically determines which questions to ask a user and when to solve them, encompassing recommendations on COVID-19 safety and behavior change interventions [38].

Furthermore, the vast majority of COVID-19 chatbots did not have any additional human support nor gamification (44/56, 79% and 42/56, 75%, respectively). These design elements for chatbot interaction show that COVID-19 chatbots are designed in such a way that most COVID-19 chatbots straightforwardly interact with the user and identify their intentions as soon as possible. Previous research suggests that conversational agents should possess a humanlike quality or assistive nature, particularly within telemedicine platforms [82]. In addition, incorporating games can help evoke personality traits and enhance thoroughness [83]. Notably, a proposed conversational agent from a university hospital in the United Kingdom targets these 2 design characteristics [73]. The in-app features will be gamified as a motivational strategy for older people in self-isolation to alleviate problems with mental health. Moreover, it is planned that this chatbot will have a social support function to facilitate cognitive behavioral therapy sessions.

From the context perspective, almost all COVID-19 chatbots (46/56, 82%) fell under the health care application domain. We expected this because COVID-19 is a public health concern. Some of these chatbots were used for personal risk assessment [10], monitoring of COVID-19 exposure [22], and dissemination of health information regarding the pandemic [26].

Aside from health care, 20% (11/56) of COVID-19 chatbots were used in the education domain. For instance, Student Interactive Assistant Android Application with Chatbot is an academic chatbot developed during the COVID-19 pandemic that assists students interactively concerning their academic requirements [30]. Only a minority of COVID-19 chatbots (6/56, 11%) were used for other purposes such as assisting with daily life. In Morocco, for instance, a chatbot is being developed with the intention of aiding students daily with their mental health [52].

Our results also showed that most COVID-19 chatbots (35/56, 63%) were designed to provide information to the users. For example, a chatbot developed in France informs and educates people about COVID-19 vaccines and reduces vaccine hesitancy [32]. There were also COVID-19 chatbots designed for utility and coaching purposes, but these types comprised only 32% (18/56) and 27% (15/56) of COVID-19 chatbots, respectively. For example, a chatbot developed by the University of California was used as a COVID-19 screening tool for health employees entering their hospital before their clinical shift [40]. Regarding coaching, Managing Stress at University instructs Indian University students on stress management during the COVID-19 pandemic by providing 3 different patterns of tailored explanations (ie, belief-based, goal-based, and belief- and goal-based explanations) on students’ intentions to change the recommended behaviors [27].

Although most chatbots (40/56, 71%) served only a single purpose, there were several COVID-19 chatbots (16/56, 29%) that were multipurpose in nature. For instance, COOPERA, as discussed previously, can be used to monitor COVID-19 symptoms, provide self-care instructions, and suggest preventive actions to avoid infections [31].

The use of COVID-19 chatbots across numerous application domains and contexts demonstrates the pandemic’s impact on other elements of life. Nonhealth industries could investigate the use of a conversational agent to efficiently solve inquiries and issues without the need for human contact. This is critical for mitigating the rapid transmission of a global pandemic. Nevertheless, our results show that the bulk of the COVID-19 chatbots were developed to disseminate updated information regarding the COVID-19 pandemic and to track virus-related symptoms and other health-related concerns. As a result, automation of knowledge extraction from reputable sources may be required for information dissemination chatbots to be deployed efficiently and effectively as knowledge sources [78].