Anxiety disorders are prevalent and undertreated among young adults. Digital mental health interventions for anxiety are promising but limited by a narrow range of therapeutic components and low user engagement. To investigate the efficacy of and engagement with Maya, a scalable, self-guided, comprehensive mobile cognitive behavioral therapy (CBT) intervention with embedded engagement features, comparing the effects of 3 incentive conditions. This randomized clinical trial recruited young adults aged 18 to 25 years with anxiety disorders through online advertisements and outpatient psychiatry clinics at Weill Cornell Medicine. Enrollment was between June 16, 2021, and November 11, 2022. Data analysis was performed from December 21, 2022, to June 14, 2024. Participants received a 6-week program of the intervention and were randomized to 1 of 3 different text message-based incentive conditions (gain-framed, loss-framed, or gain-social support). The primary outcome was change in anxious symptoms from baseline to end of treatment, as measured by the Hamilton Anxiety Rating Scale (HAM-A). The Anxiety Sensitivity Index and the Leibowitz Social Anxiety Scale scores were secondary measures. The sample consisted of 59 participants (mean [SD] age, 23.1 [1.9] years; 46 [78%] female; 22 [37%] Asian, 3 [5%] Black, 5 [8%] Hispanic or Latino, 1 [2%] American Indian or Alaska Native, 25 [42%] White, and 6 [10%] >1 race; 32 [54%] college-educated and 12 [20%] graduate or professional school-educated; mean [SD] baseline HAM-A score, 15.0 [6.5]). Anxiety, measured by HAM-A, decreased across conditions from baseline to end of the intervention (mean difference, -5.64; 95% CI, -7.23 to -4.05), and symptomatic improvement was maintained at the week 12 follow-up (baseline to follow-up mean difference, -5.67; 95% CI, -7.29 to -4.04). However, there was no evidence that change in anxiety differed by incentive condition (loss-framed vs gain-social support mean difference, -1.40; 95% CI, -4.72 to 1.93; gain-framed vs gain-social support mean difference, 1.38; 95% CI, -1.19 to 3.96). Secondary anxiety measures (Anxiety Sensitivity Index and Liebowitz Social Anxiety Scale scores) showed a similar pattern of improvement, with no evidence of differences between incentive conditions. Participants completed most of the 12 sessions (mean [SD], 10.8 [2.1]; 95% CI, 10.3-11.4), and User Mobile Application Rating Scale app quality ratings exceeded the published threshold for acceptability at all study visits. There was no evidence that either session completion or app quality ratings differed by incentive condition. In this randomized clinical trial of an app-based intervention for anxiety, the primary hypothesis that improvement in anxiety would be greatest in the condition using gain of points plus social incentives was not supported; however, the results suggest that a CBT application incorporating a full suite of CBT skills and embedded user engagement features was efficacious in improving symptoms in young adults with anxiety disorders. Given these findings, digital interventions represent a promising step toward wider dissemination of high-quality, evidence-based interventions. ClinicalTrials.gov Identifier: NCT05130281.