This was a 4-week pilot randomized controlled trial evaluating feasibility, usability, and initial efficacy of the Vivibot chatbot. At 4 weeks, control participants were given full access to the content in the experimental condition. An 8-week follow-up survey thus allowed for validation of the main outcome analyses in the control group.

Participants were young adults, aged 18-29 years, consistent with the developmental literature [36] and the field of oncology [4]; English literate; and reported having a cancer diagnosis and completing treatment for cancer within 5 years of starting the study. They also had to have access to Facebook Messenger for the study duration (either through a Facebook or Instagram account). Participants were excluded if they did not meet age or cancer diagnosis or treatment requirements or were unable to access Vivibot through Facebook Messenger. They were not excluded based on the type of cancer diagnosis.

Recruitment was conducted through Facebook advertising (67% of final sample), survivorship organizations (15%), and direct email after a potential participant expressed interest at a conference or event (17%). Enrollment was managed entirely through the Vivibot chatbot. During the recruitment period, once a user opened the chatbot interface for the first time, he/she was automatically asked four screening questions to determine eligibility. Those eligible and interested were sent an identification code and link to the study consent survey.

The Ethical and Independent Review Services [37] board approved all study procedures. Following completion of the baseline assessment, participants were instructed to return to Vivibot and indicate that they completed the baseline survey. The chatbot then randomized users 1:1 to one of two groups: (1) immediate access to the full Vivibot chatbot content (experimental group) or (2) access to daily emotion ratings through Facebook Messenger with delayed full access to the full Vivibot chatbot only after 4 weeks (control group).

Online assessments through Qualtrics software (Salt Lake City, UT) were administered at baseline and weeks 2, 4, and 8. Participants received US $20 Amazon gift cards for completing each survey (a total of US $80 possible compensation). After 4 weeks, the control group participants were given access to all of the Vivibot content. After 8 weeks, participation in the study was complete, but all participants were informed that they could continue to use the chatbot product as much or as little as they desired.

In total, 51 participants completed a baseline assessment and were randomized to a study condition (experimental group: 25; control group: 26; Figure 1). After excluding 6 control participants who were erroneously given access to chatbot content at 2 weeks instead of 4 weeks, the final analytic sample comprised 45 participants. The rate of follow-up survey completion was 73% (33/45) at 2 weeks, 73% (33/45) at 4 weeks, and 58% (26/45) at 8 weeks. We noted differences by group, with higher rates of survey completion in the control group, reaching statistical significance in week 8 (2 weeks: 80% in the control group, 68% in the experimental group, χ²: 0.8, P=.37; 4 weeks: 85% in the control group, 64% in the experimental group, χ²: 2.5, P=.12; 8 weeks: 75% in the control group, 44% in the experimental group, χ²: 4.28, P=.04).

Demographic information for the 45 participants included in the final analysis is presented in Table 1. Participants were mostly female (36/45, 80%), with an average age of 25 (SD 2.9; range: 19-29 years). Participants were on average 2.7 (SD 2.0) years postdiagnosis and 1.6 (SD 1.3) years postcompletion of active cancer treatment.

During the 4 active weeks of the study, the experimental group spent an average of 73.8 (SD 52) min across an average of 12.1 (SD 7.1) engaged sessions chatting with Vivibot. The control group spent an average of 27.13 (SD 15.8) min across an average of 18.1 (SD 8.6) engaged sessions completing the 6-item emotion ratings only.

On average, participants in the experimental group rated their experience of chatting with Vivibot as helpful, with an average rating of 2.03/3 (SD 0.72; range: 0-3). They were also likely to recommend Vivibot to a friend, with an average rating of 6.9/10 (SD 2.6; range: 0-10). When asked why they were likely or unlikely to recommend Vivibot to a friend, participants remarked on the utility and nonjudgmental nature of talking to an automated agent:

Additional themes related to having a shared experience with others who had undergone cancer treatment and being able to just “vent”:

Participants also particularly enjoyed the positive psychology content itself:

Participants who gave lower ratings of Vivibot were less specific in their feedback:

Full quotes from participants can be found in Multimedia Appendix 5.

High engagement and positive ratings of Vivibot suggest that a chatbot provides a useful and acceptable format for young adults to connect with a positive psychology intervention. Qualitative responses supplemented findings from traditional metrics of user engagement such as the number of times Vivibot was accessed and the length of time in sessions. Feedback indicated an overall positive response to the chatbot and guided developers to generate product improvements on specific features and content once the study was completed. Further development of this tool and potentially combining it with other person-to-person psychosocial interventions may enhance engagement even further.

Overall, positive psychology skills, delivered by a chatbot, were perceived as helpful and nonjudgmental by young adults who had undergone cancer treatment. Positive emotion, when stimulated through skills-based interventions, is thought to influence health outcomes, both directly and through the mediating influence of factors including changing health behaviors, improving physiological functioning, and increasing resources that influence health [15,46]. Possible mechanisms through which a positive psychology intervention could be influencing anxiety in young people who have undergone cancer treatment include increasing feelings of support and social control, which are influenced by positive emotion [47] and potentially poor after cancer treatment. Talking to a nonjudgmental “robot” could also have increased participants’ receptivity to learn skills to manage stress, which is another core mediator of the health effects of positive emotion [15]. Additional work should be done to understand the effects of individual positive emotion intervention skills and to link skill building to subsequent long-term health impacts. The experience of a bot-based intervention as nonjudgmental shows promise for future study using this format with young people.

Based on previous work with positive psychology interventions [46], we hypothesized that engagement with Vivibot would result in improved psychological well-being in the form of reduced depression and anxiety symptoms and more positive and fewer negative emotions. Engagement with Vivibot was associated with a reduction in anxiety symptoms. Although not significant, the effect size for the main anxiety analysis of .41 in this pilot trial was small to moderate by Cohen [48] standards. Further, it was comparable to that found in a meta-analysis of psychosocial support for cancer (effect size=0.42) [49] and a positive psychology intervention for caregivers (effect size=0.33) [50]. After converting to a common scale (Multimedia Appendix 6), the change we saw in the General Anxiety Disorder-7 items equivalent scores from baseline to 4 weeks in the treatment group (reduction of 1.82 points) was greater than that found in trials testing the Woebot chatbot (reduction of 0.7 points over 2 weeks [35]) and the X2AI Artificial Intelligence tool Tess (reduction of 1.4 points over 4 weeks) [51]. Thus, Vivibot is associated with, at a minimum, a comparable reduction in anxiety symptoms as other psychosocial interventions for people with cancer and similar digital health tools available on the market for general populations.

Among those who used Vivibot, depression symptoms were not reduced compared to those who only completed daily emotion ratings. The active nature of the control condition in this trial could have accounted for this. The emotion rating control on its own appeared to carry some benefits for participants (rating emotions was associated with decreased negative emotion and increased positive emotion). Additionally, anecdotally, some users remarked on how completing the emotion ratings was helpful to them when asked why they would be willing to refer Vivibot to a friend (eg, “This is helpful to review your feelings” and “I like the daily check in questions because it reminds me to check in with myself on how I am feeling and think about why my answers might change from day to day;” Multimedia Appendix 5). This is consistent with the literature showing that emotion labeling itself can contribute to beneficial emotion regulation in response to negative events [52]. Of note, the depression symptom reduction, when converted to the Patient Health Questionnaire - 9 items equivalent, in the experimental group in our trial (1.5 points over 4 weeks) was comparable to that reported in the trial of the X2AI tool (0.9 points over 4 weeks) [51]. Since both the experimental group and the control group experienced the same emotion rating exercise, it is possible that the emotion ratings themselves carried the benefit of reductions in negative emotion, increases in positive emotion, and reductions in depression reported here in both experimental and control conditions. Further studies using a control group that did not engage in emotion labeling would be needed to more fully identify the mechanisms behind these effects.

This work extends findings showing promise for technology-enabled solutions to address mental health problems. Internet-based interventions have been shown to work at least as effectively as face-to-face intervention in improving mental health across a range of conditions. In addition to opportunities to scale interventions, online delivery offers self-pacing and the ability to incorporate practices into daily life, thereby transcending geography, time, and space [53]. Our work further suggests the chatbot interface, which is less costly than a human-delivered intervention, is a promising format to deliver psychosocial interventions to young adults. Vivibot included many features found to be associated with positive outcomes in internet-based interventions for depression among adolescents, including surface credibility, esthetics, and content appeal to the specific population; appropriate reduction of content compared to that in traditional interventions; and use of self-monitoring [54]. Acceptance of this positive psychology intervention is therefore not surprisingly consistent with the broad acceptance found for online-based interventions for depression and anxiety among adolescents [55].

Further, this work contributes to the growing body of literature showing promise for interventions delivered via social media in supporting short-term symptom improvement and behavior change among young adults. Most work in this area has focused on health behaviors such as smoking cessation [56] and physical activity among young people with cancer [57]; this study provides early evidence that social media–mediated intervention can be harnessed to reduce mental health symptoms. Additional randomized trials are needed to strengthen conclusions for this particular intervention and for this field in a broad sense.

There are several limitations to this study that weaken the generalizability of these findings and motivate further investigations. First, this was a small pilot study that was not powered to detect significant effects in the psychological outcomes measured. Due to the relatively small population size and heterogeneity of treatment facilities, it is difficult to identify and recruit adolescents and young adults who had completed treatment for cancer for research studies. Our results need to be replicated in a sample large enough to reliably detect a moderate-sized effect between groups. Second, this study was limited to participants willing and able to access Vivibot via Facebook Messenger. Although Facebook Messenger had the advantage of being widely accessible to this population and has almost full saturation among young adults in the United States [58], it is possible that participants did not wish to engage with Facebook Messenger in light of the controversy surrounding the privacy policies of Facebook during the study period. Other platforms, including WhatsApp and smartphone apps, should be considered in future adaptations of Vivibot. A third limitation was that neither the type nor stage of cancer was considered in the analyses. Due to the unique developmental and psychosocial needs of young people who have been treated for cancer, this investigation collapsed across cancer diagnoses to focus on a more targeted age range. It is possible that patients with different types of cancers would react differently to the intervention content of Vivibot; future work with a larger sample size should investigate this aspect further.

The Vivibot chatbot was engaging to users and resulted in anxiety reduction comparable to that found in the literature with digital health interventions targeting anxiety and depression. Given their short duration, non–artificial intelligence-based content-delivery system results, and ease with which it could be delivered in Facebook Messenger or other chat platforms, chatbots are promising for supporting the mental health needs of young people with cancer at a vulnerable time in their lives. The randomized design of this trial extends the promising findings of other positive psychology interventions delivered online [59] and gives confidence that the effect size seen with anxiety is worthy of follow-up. In addition to a larger trial, future research should consider whether the tool is as effective when delivered to other populations or platforms.