This study was a 4-arm parallel randomized controlled trial evaluating 2 self-guided mobile mindfulness interventions against an active psychoeducation condition and a WL. Participants were assigned in a 1:1:1:1 ratio to TM, HMP, MBP, or WL. The intervention lasted 21 days, with assessments conducted at baseline (T1), postintervention (T2), and a 3-month follow-up (T3). WL participants were offered access to intervention materials after postintervention, so no untreated WL follow-up data were collected. A sample size estimate using G*Power (version 3.1; Heinrich-Heine University Düsseldorf) indicated that 280 participants were required for the primary postintervention 4-group comparison, assuming a conservative small-to-medium effect size based on prior meta-analytic evidence on mindfulness apps [22], a Cohen f of 0.20, an α level of .05, a power of 0.80, and 4 groups. Given the high dropout rates commonly observed in self-guided digital mindfulness interventions [23], recruitment was increased to preserve sufficient data for the primary comparisons. The trial was approved by the Institutional Review Board (IRB) of the Faculty of Psychology at the affiliated university (IRB number BNU202404240092) and preregistered with the Chinese Clinical Trial Registry (ChiCTR2400093771). All procedures were completed online. The completed CONSORT-eHEALTH (Consolidated Standards of Reporting Trials of Electronic and Mobile Health Applications and Online Telehealth) checklist is provided in Checklist 1.

Participants were adults aged 18 to 65 years who had smartphone access, were able to provide informed consent, and reported mild to moderate symptoms of anxiety or depression, defined as a score of 5 to 14 on the Generalized Anxiety Disorder–7 or Patient Health Questionnaire–9 [24-26]. Exclusion criteria included scores above 14 on either measure, current or planned psychological treatment, psychiatric medication use within the past 3 months, prior formal mindfulness training, a history of psychosis, or active suicidal ideation. These criteria were used to maintain the study’s focus on mild to moderate distress, reduce treatment-related confounding, and direct higher-risk participants toward more appropriate care. Recruitment was conducted through social media in August 2024, and all screening, consent, and baseline procedures were completed online. Individuals who were not eligible were provided with mental health support information, including crisis resources, and were encouraged to seek appropriate care when needed.

The study was approved by the IRB of the Faculty of Psychology at Beijing Normal University (IRB number BNU202404240092), China. All procedures were conducted in accordance with institutional ethical guidelines.

After the baseline assessment, participants were randomized by a computer-based procedure embedded in the study platform. Allocation was concealed from investigators before assignment. Participants were aware of their assigned materials after randomization, but they were not informed of the specific comparative hypotheses. Outcome data were collected online, and analysts were masked to group assignment during the primary analyses. No changes were made to the trial protocol or planned analyses after the study began.

All intervention materials were delivered through a mobile app. TM, HMP, and MBP participants received access to condition-specific materials, whereas WL participants completed assessments only during the 21-day study period and were offered intervention access afterward. The active conditions were comparable in delivery mode and daily task expectations but differed in whether and how guided practice was provided. App logs indexed exposure and access duration, although they could not verify practice fidelity. Additional intervention protocol details are provided in Multimedia Appendix 1.

Participants were asked to log in daily and complete at least 15 minutes of assigned content within each 24-hour window. TM involved one 15-minute guided mindfulness practice per day, HMP involved 3 routine-linked 5-minute guided practices per day, and MBP involved psychoeducational materials without guided practice. No monetary compensation was provided.

Outcome assessments were completed at baseline, postintervention, and 3-month follow-up. Follow-up analyses were limited to TM, HMP, and MBP because WL participants received intervention access after postintervention. No adverse events or unintended effects were reported during the study period.

All analyses were conducted using SPSS Statistics (version 27; IBM Corp) under an intention-to-treat framework. Descriptive statistics were examined at each assessment point. Baseline equivalence across the 4 groups was tested using ANOVA for continuous variables and chi-square tests for categorical variables.

Missing data were evaluated before outcome analyses. Little test suggested that missingness was consistent with missing completely at random (P=.12) [35,36]. Missing values were therefore estimated using the expectation-maximization algorithm. Given the substantial attrition, supplementary analyses compared completers and noncompleters on baseline clinical and psychological variables; no systematic baseline differences were found.

For postintervention outcomes at T2, between-group differences were examined using analyses of covariance, with baseline scores entered as covariates. This approach was used to compare the 4 randomized groups on the primary outcomes of depressive symptoms and anxiety symptoms, as well as on the secondary and exploratory outcomes. When the omnibus group effect was significant, pairwise comparisons were conducted with Bonferroni adjustment. Effect sizes are reported as partial ηp². To reduce the risk of inflated type I error across the primary postintervention tests, the Holm-Bonferroni procedure was applied.

To examine changes over time, longitudinal analyses were conducted using linear mixed-effects models. These models included fixed effects for group, time, and the group× time interaction, with a random intercept for participants and an unstructured covariance matrix. Because the WL group was not assessed at follow-up, the longitudinal models were estimated for the 3 active intervention conditions only: TM, HMP, and MBP. Post hoc comparisons for the mixed models were adjusted using the Sidak procedure. The mixed-model framework was used to characterize trajectories from baseline to postintervention and follow-up while accommodating repeated observations within individuals.

Engagement and acceptability analyses focused mainly on HMP and TM. Group differences in practice duration and self-reported engagement were tested using 2-tailed independent-samples t tests. Attrition-related comparisons used chi-square tests, t tests, and 2-way analyses of variance, as appropriate. All tests were 2-tailed, with statistical significance set at P<.05.

A total of 876 individuals were screened for eligibility. Of these, 190 were excluded because they did not meet the inclusion criteria (n=109), did not respond after screening (n=73), or were no longer interested in participating (n=8). The remaining 686 participants completed the baseline assessment and were randomized to TM (n=172), HMP (n=172), MBP (n=171), or WL (n=171). All 686 randomized participants were included in the intention-to-treat analyses.

Retention differed across conditions over time. At postintervention, outcome data were available for 64 participants in TM, 74 in HMP, 118 in MBP, and 123 in WL. At the 3-month follow-up, outcome data were available for 35 participants in TM, 40 in HMP, and 49 in MBP. As specified in the study design, no untreated follow-up data were collected for WL after participants were offered access to intervention materials. No adverse events or unintended effects were reported. Figure 2 presents the participant flow.

The 4 randomized groups were broadly comparable at baseline. The mean age ranged from 29.39 (SD 6.28) years to 30.68 (SD 7.97) years across conditions, and most participants had completed college education or higher. No significant between-group differences were observed for age, gender identity, educational attainment, or baseline scores on the primary and secondary outcomes.

Observed means by group and assessment point are shown in Table 1. For the primary outcomes, ANCOVA controlling for baseline scores showed significant postintervention group effects for depressive symptoms (F_3,681=28.67, P<.001, ηp²=0.11) and anxiety symptoms (F_3,681=30.11, P<.001, ηp²=0.12). Both TM and HMP had lower depressive and anxiety symptom scores than MBP and WL. TM also showed lower anxiety than HMP (P=.04), whereas TM and HMP did not differ significantly on depressive symptoms (P=.63). Full Bonferroni-adjusted pairwise comparisons with P values are provided in the Detailed Postintervention Results section of Multimedia Appendix 2.

For secondary and exploratory outcomes, significant postintervention group effects were observed for interpersonal difficulties (F_3,681=19.50, P<.001, ηp²=0.08), adaptive emotion regulation (F_3,681=6.85, P<.001, ηp²=0.03), maladaptive emotion regulation (F_3,681=13.00, P<.001, ηp²=0.05), mindfulness (F_3,681=20.88, P<.001, ηp²=0.08), and affective balance (F_3,681=12.03, P<.001, ηp²=0.05). TM and HMP showed more favorable scores than MBP and WL on mindfulness, affective balance, and interpersonal difficulties. For adaptive emotion regulation, MBP showed lower scores than TM, HMP, and WL. For maladaptive emotion regulation, WL showed higher scores than all 3 active intervention conditions. Full Bonferroni-adjusted pairwise comparisons are provided in Multimedia Appendix 2.

Longitudinal analyses were conducted for TM, HMP, and MBP only because the WL group was not assessed at follow-up. For depressive symptoms, mixed-effects models showed a significant effect of time (F_2,515=117.51, P<.001, ηp²=0.31) and a significant group × time interaction (F_4,515=7.74, P<.001, ηp²=0.06). For anxiety symptoms, the effect of time (F_2,515=97.90, P<.001, ηp²=0.28) and the group × time interaction (F_4,515=13.07, P<.001, ηp²=0.09) were also significant. TM and HMP maintained lower depressive symptom scores than MBP at postintervention and follow-up, with no significant difference between TM and HMP. For anxiety symptoms, both mindfulness conditions showed lower scores than MBP at postintervention and follow-up; TM showed lower anxiety than HMP at postintervention, but this difference was not significant at follow-up.

Secondary and exploratory outcomes showed a more variable maintenance pattern. Significant group × time interactions were observed for adaptive emotion regulation (F_4,515=7.74, P<.001), maladaptive emotion regulation (F_4,515=3.10, P=.02), mindfulness (F_4,515=13.86, P<.001), interpersonal difficulties (F_4,515=17.11, P<.001), and affective balance (F_4,515=8.02, P<.001). Affective balance showed relatively sustained improvement, whereas mindfulness and interpersonal difficulties declined somewhat after postintervention while remaining improved relative to baseline. Emotion regulation outcomes also showed evidence of change over time, with stronger patterns for the mindfulness practice conditions than for MBP on maladaptive regulation. Full mixed-effects model statistics and post hoc comparisons are provided in Multimedia Appendix 2.

Attrition was substantial in both mindfulness practice conditions. At postintervention, outcome data were unavailable for 62.8% (108/172) of participants in TM and 57% (98/172) in HMP, compared with 31% (53/171) in MBP and 28.1% (48/171) in WL. Among participants assigned to the active conditions, 59.3% (102/172) in TM, 65.1% (112/172) in HMP, and 71.3% (122/171) in MBP completed at least 10 intervention days.

HMP did not show a clear engagement advantage over TM. Total practice duration, self-reported engagement, and participant satisfaction did not differ significantly between the 2 mindfulness practice conditions. Supplementary attrition analyses showed no significant differences between completers and noncompleters in age or other baseline clinical and psychological variables, and no significant group × completion status interactions were observed. Detailed attrition analyses, including age comparisons, are provided in Tables S2 and S3 in Multimedia Appendix 2.

This randomized controlled trial examined whether a routine-embedded mobile mindfulness format could reduce psychological distress in adults with mild to moderate symptoms of anxiety or depression and whether this format offered practical advantages over a more traditional app-delivered mindfulness program. Three findings deserve emphasis. First, both mindfulness practice conditions, HMP and TM, showed greater postintervention reductions in depressive symptoms than MBP and WL. Second, both mindfulness conditions also showed lower anxiety than MBP and WL, although TM produced a somewhat stronger short-term effect on anxiety than HMP. Third, the 2 mindfulness conditions showed more favorable outcomes than the comparison conditions on several secondary measures, including mindfulness, cognitive emotion regulation, affective balance, and interpersonal difficulties. These findings are consistent with prior evidence that MBIs can reduce symptoms of anxiety and depression [10,11] and can also influence broader aspects of emotional functioning [6-8].

These findings suggest that repeated mindfulness practice, rather than psychoeducational exposure alone, was associated with broader improvements in emotional functioning and distress-related outcomes in this sample. HMP was associated not only with reductions in depressive and anxiety symptoms but also with improvements in affective balance and interpersonal functioning. Because affective balance reflects the relative predominance of positive over negative emotional states [32,33], improvements in this domain may indicate shifts in the overall emotional tone of everyday life. Reductions in interpersonal difficulties further suggest potential benefits in relational functioning, consistent with prior work linking mindfulness to interpersonal processes [7,8]. Together, these results indicate that the value of HMP may extend beyond symptom reduction to broader aspects of daily experience.

The findings also align with the study’s expectation that HMP and TM would show broadly comparable clinical benefits, although with one important qualification. HMP and TM did not differ significantly in depressive symptoms at postintervention, and both conditions showed more favorable outcomes than MBP and WL across several outcomes. For anxiety, TM showed lower scores than HMP immediately after the intervention, but this difference was not observed at follow-up. This finding may reflect the different practice structures used in the 2 conditions. TM involved one continuous 15-minute practice, whereas HMP divided the same total daily practice time into three shorter routine-linked practices. Prior work comparing formal and informal mindfulness formats similarly suggests that both formats can be beneficial, but that informal or daily-life-integrated practice is not necessarily superior to formal practice [37]. A continuous practice period may provide more opportunity for sustained attention and settling during a single episode, which may be especially relevant for short-term anxiety symptoms. Given that the between-condition difference was limited to postintervention, this finding should be interpreted cautiously. The present findings therefore suggest that HMP is a viable routine-embedded format, while also indicating that changing the structure of practice does not necessarily produce stronger immediate anxiety effects than a session-based format.

The longitudinal findings provide some support for the durability of these effects, although the pattern was not uniform across outcomes. Improvements in depressive symptoms were largely maintained at follow-up, and both mindfulness conditions continued to show more favorable trajectories than MBP. Anxiety remained lower than baseline, although the temporal pattern differed somewhat between conditions. For secondary outcomes, maintenance was more variable, with affective balance showing relatively sustained improvement, whereas mindfulness and interpersonal difficulties showed some decline over time.

The exploratory expectation that HMP would show more favorable engagement-related outcomes than TM was not supported. Although HMP was designed to reduce friction in practice initiation and improve practical fit with daily routines, HMP did not differ significantly from TM in practice duration, attrition, or participant satisfaction. This suggests that restructuring practice around daily routines may help preserve clinical benefit but may not be sufficient on its own to overcome the broader engagement challenges common in self-guided digital interventions. Engagement may also depend on motivational and accountability supports, including feedback, reminders, brief guidance, or peer- and group-based practice [38,39]. A more interactive or blended version of HMP could therefore be examined in future work as a way to strengthen adherence while preserving the accessibility and routine-embedded design of the intervention.

The present findings are broadly consistent with prior work suggesting that digitally delivered mindfulness interventions can reduce anxiety and depressive symptoms [9-11] and can also influence broader aspects of emotional functioning [5,33]. What this study adds is a more direct comparison between 2 app-delivered mindfulness formats that differed not in general therapeutic orientation but in how practice was structured in daily life.

Much of the digital mindfulness literature has focused on whether app-based mindfulness is beneficial relative to control conditions [11,15]. This study addressed a somewhat different question: whether changing the delivery structure of practice, from one longer formal session to multiple brief routine-linked practices, might preserve benefit while better fitting the realities of everyday use. In that respect, the findings are encouraging but qualified. HMP showed broadly comparable benefits to TM across most outcomes, although TM showed a short-term advantage for anxiety at postintervention. This suggests that mindfulness practice need not be restricted to a single formal daily session to remain useful, while also indicating that routine-linked practice should not be assumed to be superior to session-based practice. Brief, context-linked exercises may still support symptom improvement and broader emotional functioning. This is consistent with work emphasizing the integration of mindfulness into ordinary activities rather than restricting practice to formal sessions alone [16,40]. In this sense, HMP may be understood not as a simplified version of mindfulness, but as a differently organized practice format designed to fit more naturally into everyday life.

At the same time, the findings do not indicate that HMP conferred a clear advantage over TM, either in clinical outcomes or engagement. The contribution of the study is, therefore, less about demonstrating superiority and more about showing that a routine-embedded format can remain viable without a loss of efficacy across most outcomes.

Another useful comparison concerns the psychoeducation-only condition. MBP showed a weaker and less consistent pattern of effects than the 2 mindfulness practice conditions. This pattern is consistent with the view that psychoeducational exposure alone may be less likely to produce broader changes in emotional functioning when it is not accompanied by repeated experiential practice [41,42]. In this trial, MBP did not match the mindfulness conditions on the primary outcomes and was generally less favorable on the process-related outcomes as well. This does not imply that psychoeducation is unhelpful, but suggests that conceptual knowledge alone is unlikely to account for the observed improvements.

Several limitations should be acknowledged. First, attrition was substantial, particularly in the 2 mindfulness practice conditions. Although the intention-to-treat analyses and the baseline comparisons between completers and noncompleters reduce concern that the findings were driven entirely by selective retention, the level of dropout still places an important constraint on interpretation. In particular, the engagement promise of HMP remains unresolved because the routine-embedded format did not clearly reduce attrition relative to TM.

Second, engagement was assessed using app-recorded usage and brief self-report indicators, neither of which can fully capture the quality or fidelity of mindfulness practice. App logs indexed exposure to the intervention materials, but they could not establish whether participants completed every exercise as intended or whether some routine-based practice occurred outside the tracked platform. This is especially relevant for HMP because its structure may make off-platform practice more likely. As a result, the present engagement indicators should be interpreted as approximations rather than precise measures of actual mindfulness practice.

Third, the WL group was not followed beyond postintervention because participants in that group were offered access to the intervention materials after T2. This limits the strength of conclusions about longer-term comparative effects. The follow-up analyses are still informative for describing the trajectories of the 3 active conditions, but they do not allow firm statements about whether the observed maintenance effects would remain stronger than no intervention over the same period.

Fourth, the sample consisted of adults with mild to moderate self-reported distress recruited online, rather than participants selected on the basis of diagnostic interviews or treatment-seeking clinical status. The illness duration, time since diagnosis, and prior treatment history were not assessed, precluding examination of whether attrition differed by the chronicity of mental health difficulties. The findings, therefore, speak most directly to low-intensity digital intervention use in individuals with elevated distress, not to more severe or complex clinical presentations. This limits generalizability to higher-risk populations and to settings in which self-guided app use is not the primary mode of delivery. Future studies should collect more detailed clinical history, as prior research suggests that treatment history and clinical severity may be relevant to dropout [43,44].

Finally, although HMP was conceptually informed by habit formation and contextual cueing [18-20], this study did not directly measure habit strength, automaticity, or cue-response consistency. For that reason, the study can show that a routine-embedded format was feasible and beneficial, but it cannot demonstrate that habit formation itself was the mechanism responsible for the observed effects. Future work would benefit from including more direct assessments of routine linkage, cue-dependent initiation, and sustained practice over time.

In summary, this trial provides evidence that a routine-embedded, self-guided mobile mindfulness intervention may help reduce mild to moderate psychological distress. HMP and TM both outperformed psychoeducation alone and WL on depressive and anxiety symptoms and showed broader benefits across several measures of emotional functioning. However, HMP did not show better adherence, greater satisfaction, or stronger short-term efficacy than TM. The main implication is, therefore, not that routine embedding is superior to traditional app-delivered mindfulness, but that it may represent a viable alternative way of delivering mindfulness practice in digital mental health settings. This may be useful for future intervention design, particularly if combined with stronger strategies to support retention and continued engagement.