BACKGROUND: Compared with other forms of online mental health interventions, programs delivered through social media apps may require less training and be more acceptable and accessible to various populations. During and after the pandemic, both the number of social media users and the prevalence of social-media-based mental health interventions increased significantly. However, to the best of the authors' knowledge, no meta-analysis so far has focused on rigorous social-media-based mental health interventions for general populations. OBJECTIVE: This preregistered meta-analysis synthesized findings from rigorously designed randomized controlled trials (RCTs) (ie, decent sample size, low attrition, and comparable baseline conditions) to understand whether social-media-based mental health RCTs work as expected in reducing mental health issues. METHODS: We searched for articles through database queries, hand searching, and forward and backward citation tracking, which yielded 11,658 studies. We only included social-media-based RCTs with a decent sample size (n≥30 for each experimental condition at baseline assessment), low differential attrition between treatments and controls (<15%), equivalent baseline conditions (differences between conditions <0.25 SDs), published after 2005, and delivered by nonresearchers. These RCTs must aim at reducing mental health issues, such as depression, anxiety, and stress. We excluded one-item outcome measures. RESULTS: After double-blinded screening, 17 eligible studies (total sample sizes=5624) were included in this meta-analysis. Meta-regression results showed that, on average, these social-media-based interventions were effective (effect size [ES]=0.32, P<.001, NES=61, 95% CI 0.24-0.45, I²=88.10, τ2=0.13) for the general population (range of mean age: 15.27~59.65). In other words, social-media-based interventions were effective at reducing anxiety (ES=0.33, P=.04, n=27), depression (ES=0.31, P<.001, n=31), and stress (ES=0.69, P=.02, n=12). Moderator analysis showed that social-media-based interventions are more effective when the participants are more than 70% female, when the programs are human-guided, social-oriented, and when control groups are care as usual. Furthermore, we conducted a risk of bias analysis, publication bias analysis, and sensitivity analysis, which show low risks of bias and robust findings. The biggest limitation of this review is the small sample size of 17 included studies, which restricts the power of our models. CONCLUSIONS: While technology can be a double-edged sword, this meta-analysis highlighted social media's benefits and future potential in the treatment of mental health symptoms.