Participants were Youper subscribers (ie, users who paid for full access to Youper) who downloaded the app between March 4, 2020, and July 10, 2020. This time frame was selected because Youper was relatively stable during this period (ie, no significant updates or changes to the intervention were deployed during this time). Subscribers paid US $44.99 to have unlimited access to Youper’s interventions for 1 year. Users who did not subscribe were only able to access the emotion regulation interventions once as part of a free sample, and therefore, were not included in this analysis.

Of 5943 users who completed at least one symptom measure in the study timeframe, 76.01% (n=4517) agreed for their data to be used for research, leaving a useable sample of 4517 participants. The sample was composed of 81.62% women (n=3687), 14.15% men (n=639), and 3.43% nonbinary individuals (n=155), and the average age of participants was 28.73 years (SD 9.63). Additional participant demographics and clinical characteristics are presented in Table 1. Participants completed symptom assessments at baseline (T0; within 3 days of subscribing to Youper), 2 weeks after baseline (T1), and 4 weeks after baseline (T2). Assessments were available to users every 14 days, and the majority of users completed their assessments within 3 days of them becoming available. Participants received access to anxiety or depression symptom measures based on their responses to screening questions. Symptom measures were administered if they endorsed a history of being diagnosed with clinical anxiety or depression, or if they reported elevated anxiety or depression symptoms on 2-item screening measures. Participants could receive only an anxiety measure, only a depression measure, or both depending on their responses to the screening items. Throughout the course of the measurement period, participants engaged in emotion regulation interventions at their discretion when emotional episodes arose.

Youper is a novel intervention approach that aims to enhance the user’s emotion regulation skills using empirically supported treatments for anxiety and depression. Although the emotion regulation strategies employed in Youper have precedent in existing treatment protocols for anxiety and depression, the adaptation of these interventions to help a user manage emotional distress at the present moment is novel. Youper’s intervention is delivered via a conversational (ie, chat) interface and is entirely automated. Youper primarily uses a decision tree to select its responses to the user input. Each interaction with Youper is called a “conversation.” Conversations follow a prespecified sequence (see Figure 1 for examples): identify current emotion and intensity (0%-100%), select contributing factors from a prespecified list, complete an open text entry about what is causing the current mood, complete emotion regulation skill practice for a negative mood or wellness practice for a positive mood (see Table 2), and identify current emotion and intensity (0%-100%).

The goal of each conversation is to help the user learn adaptive emotion regulation skills. If the user is experiencing a negative emotion, the skill targets the current emotion. If the user is experiencing a positive emotion, the skill encourages upregulation of that emotional state. If the user is in a neutral state, the skills encourage practice of activities that promote emotionally adaptive behaviors and attentional and cognitive control. Youper primarily uses just-in-time interventions (delivered at the moment of need) to help users practice and learn skills for emotion regulation. Youper’s interventions target the 3 categories of treatment mechanisms defined by the common elements framework [29].

The common elements framework provides a review of common elements across cognitive and behavioral therapies inclusive of both traditional (cognitive therapy, behavioral activation) and third-wave (acceptance and commitment therapy, dialectical behavior therapy) approaches. They identify 3 mechanistic targets common to multiple effective therapies, including attention change (improving attentional focus and flexibility), cognitive change (improving ability to change perspective on an event), and context engagement (engaging new internal and external contexts to counteract maladaptive patterns). Youper’s interventions aim to increase emotion regulation skills by targeting these common elements. For example, Youper includes interventions to increase attentional control such as mindfulness, cognitive change such as cognitive restructuring or gratitude journaling, and context engagement via behavioral activation exercises. The common elements framework was used to guide the development of Youper’s interventions due to the extensive empirical support for the efficacy of each of these targets in enhancing emotion regulation and reducing symptoms of emotional disorders [30-36].

Each intervention is described in Table 2. Each skill follows a series of steps modeled after existing treatment manuals or research protocols. Skills practice includes a variety of formats including open-text entry following a prompt, graphical user interfaces, written content delivered via the chat, and audio.

The present study had 3 aims. First, we examined the acceptability of Youper AI therapy by assessing user ratings and retention metrics among subscribers. Second, we tested whether there were significant reductions in anxiety and depression symptoms. Third, we examined the cumulative regulation hypothesis, which predicts that the frequency of within-conversation emotion regulation success would predict symptom reduction.

Findings indicated that users were well retained and provided high ratings of Youper (median 5/5). As hypothesized, users showed significant reductions in symptoms in the first 2 weeks of using Youper with sustained improvements through 4 weeks from initial download. Finally, consistent with the cumulative regulation hypothesis, greater frequency of within-conversation emotion regulation successes significantly predicted greater reductions in anxiety and depression. Although no demographic predictors emerged, users with more self-reported diagnosed psychiatric conditions showed a slight return of depression symptoms between 2 and 4 weeks from first subscribing to Youper.

Because retention poses a significant challenge for entirely unguided treatment programs, our finding that 60.44% (2730/4517) of users continued to engage with the app in the second week and 42.66% (1927/4517) of users continued to engage with the app in the fourth week after initial download is promising. Although there are no clearly established metrics of retention for mobile apps, a recent paper examining retention among different mobile apps showed that Youper had the highest “stickiness” (measured by the ratio of active users to downloads in a given month) compared to any other treatment app for anxiety and depression [57]. Because Youper users experienced symptom improvements on average within the first 2 weeks of app use, with the present retention rate, it is likely that a large portion of users will stick with the app long enough to experience some positive effects. It is also notable that the median satisfaction rating given by users was 5 out of 5. Taken together, these findings indicate that Youper has great potential as a highly acceptable and adequately engaging digital treatment program.

Youper users showed a moderate effect size reduction for anxiety (d=0.57) and depression (d=0.46) within 2 weeks of starting app use. The reduction in anxiety symptoms was maintained through the 4-week period (day 0 to day 28: d=0.60). The reduction in depression symptoms was maintained through the 4-week period (day 0 to day 28: d=0.42) although depression increased slightly, but significantly, between weeks 2 and 4. These effect sizes are comparable to those found in RCTs of other commercially available mobile apps tested for a similar duration [9-11,58], suggesting that the AI therapy approach is viable for further testing in a randomized clinical trial. Youper users also had high success at regulating their negative emotions with each conversation. Given the low cost and potential for broad dissemination of Youper, these findings are particularly exciting, as they provide preliminary evidence of Youper’s effectiveness as an emotion regulation tool and a transdiagnostic treatment. It is important to note, however, that the final mean PHQ score of 11.9 still fell in the moderate severity range. Thus, as we begin to understand the mechanisms of the AI therapy approach and gain greater understanding of how to maximize user engagement, we are hopeful that effects on symptom reduction will continue to improve.

Youper’s symptom reduction, retention, and satisfaction ratings are notable because they were demonstrated in a real-world setting. Although highly controlled feasibility pilot trials allow determination of causal inference, these studies may not be generalizable to real-world settings and may fail to address issues of external relevance and dissemination [59]. Our analysis included a very large sample of Youper users who voluntarily downloaded and purchased the Youper program. Unlike in typical research settings, users were not recruited to participate or compensated for their assessments or for providing their feedback during their participation. Observed retention rates and symptom reduction therefore have already been shown in a real-world setting and population.

The finding that cumulative within-session emotion regulation was strongly predictive of symptom reduction provides preliminary evidence for a potential mechanism of the AI therapy just-in-time intervention approach. Youper is theorized to enact its effects by enhancing emotion regulation skills via -in-time interventions. Thus, more effective emotion regulation sessions would indicate progress towards enhanced general emotion regulation skills and ultimately, symptom reduction. Therefore, it is promising that the effectiveness of the emotion regulation practice predicts the longer-term impacts of app use on symptom reduction. Although these results provide initial support for the theorized model underlying Youper’s treatment approach, randomization is critical for rigorously testing within-session emotion regulation as a mediator of symptom reduction.

Interestingly, no demographic predictors of symptom reduction emerged. These findings are largely consistent with the existing literature where demographic features rarely predict symptom reduction [60-67]. These findings are promising, suggesting that digital treatment programs can be broadly disseminated with similar potential benefit across demographic groups. The number of comorbid diagnoses was a significant predictor of response such that users who reported more diagnosed mental health conditions showed a slight return of depression symptoms between 2 and 4 weeks from the first subscription date. These findings are consistent with prior literature showing poorer outcomes with greater comorbidity in depression treatment [68-70]. Users with more diagnosed conditions likely have a more severe clinical presentation, meaning that an entirely self-guided program may be less effective for this group. The finding that concurrent medication and therapy did not significantly impact symptom reduction suggests that the demonstrated effects of Youper on symptoms are unlikely to be explained by concurrent treatment, and that participating in other treatments alongside Youper does not hinder its effects.

Despite many strengths, our study had a few limitations. First, because these data were not collected as part of a research study, we did not have a control group, making it impossible to determine whether symptom reduction was simply due to the passage of time. However, given that effect sizes for symptom reduction that we found are comparable to those found in RCTs of other mobile app programs that showed significant differences between active treatment groups and wait list controls [9-11,58], it is unlikely that these effects can be explained by spontaneous remission. Second, because this was an observational study, we used the symptom data that were available to us, which included only self-report measures. Although we used validated measures, solely relying on self-report does not give a complete picture of the impact of Youper on clinical symptoms and overall functioning that could be more thoroughly assessed via clinical interviews. Third, this study included only 2 brief measures as outcomes: the PHQ and the GAD-7. Although these measures are widely used and show excellent psychometric properties, additional measures of anxiety, depression, and other purported outcome targets, such as quality of life and functioning, could help us better understand Youper’s effectiveness. Fourth, 47.84% (2161/4517) of Youper users were concurrently taking medication or engaging in therapy, meaning that it is possible symptom reduction resulted from participation in these other treatments rather than Youper (although concurrent treatment was not a significant moderator of symptom reduction). Finally, our emotion regulation measure was not designed to assess the magnitude of emotion regulation success, meaning that our metric included only success or failure with each conversation. These limitations should be addressed in future studies that include a control group, that assess symptoms using clinician-administered measures, that include a broader array of self-report measures, and that use more precise measures of emotion regulation success.

This study provides preliminary evidence for Youper’s acceptability in a real-world setting that is unfettered by the constraints of highly controlled clinical trials. It also provides evidence of Youper’s effectiveness as an entirely unguided intervention for anxiety and depression. Finally, we demonstrated that Youper’s effects on symptom reduction may be explained by repeated within-session emotion regulation successes, providing preliminary support for the process by which a just-in-time intervention can be effective for the treatment of emotional disorders. Our results highlight the potential impact of Youper as a low-cost, light-touch, transdiagnostic intervention for anxiety and depression that can be broadly disseminated to improve mental health for millions of people around the world.