Concerns about digital well-being, the equilibrium individuals maintain in their relationship with mobile devices [1,2], have become a prominent public health issue as smartphones increasingly mediate daily activities [3,4]. Social media platforms, driven by engagement-maximizing recommendation algorithms [5,6], create environments in which users’ conscious intentions to moderate their digital consumption compete against scientifically designed engagement mechanisms [7], contributing to excessive screen time and associated mental health concerns [8]. As awareness of these dynamics has grown, so has interest in interventions that support users in self-regulating their digital behaviors without removing the underlying technologies from their lives.

In response to these emerging public health challenges, researchers and developers have explored technological solutions to promote digital well-being and self-regulation. These attempts led to the emergence of digital self-control tools (DSCTs) designed to help users monitor and regulate their digital activities [9,10]. These tools embody a paradoxical approach: using technology to regulate technology by using digital interventions that modify users’ decision-making environments [11]. DSCTs operate across various platforms and contexts, from browser extensions that modify YouTube’s user interface by removing recommendation feeds [12] to mobile apps designed to address excessive smartphone usage [10].

Central to DSCTs is the concept of the nudge: an element of choice architecture that predictably alters behavior without restricting options [13]. Increasingly applied to digital environments as “digital nudging” [14], DSCTs employ various nudge strategies, from friction-based approaches that intercept every app launch [15] to threshold-based nudges that activate only when usage exceeds a self-configured limit [9,10]. The latter approach preserves user autonomy by operating passively until triggered by actual behavior, but its effectiveness depends critically on configuration: thresholds set too high may never trigger, rendering the tool inert, whereas settings perceived as too restrictive may provoke abandonment [16]. This makes the user’s willingness to configure and maintain meaningful nudge parameters, rather than the tool’s technical capabilities, a central determinant of DSCT effectiveness.

One Sec, a mobile DSCT, exemplifies the friction-based approach: it intercepts every attempt to open a target application with a deliberative interface combining messaging, temporal friction, and the option to cancel [15]. While effective over a 6-week period [15], the high-friction nature of such interventions generates user resistance [16], highlighting a broader tension in DSCT design between intervention intensity and long-term retention.

Empirical evidence underscores this tension: studies on DSCTs have yielded mixed results, with one study finding that such interventions lacked perceived effectiveness [17]. While a meta-analysis found that control tools may be linked to reduced smartphone use, it did not address whether disconnection alone fosters digital well-being [11]. Current implementations of these tools face substantial limitations, particularly in maintaining sustained user interaction [18], which is a common issue for digital and smartphone-based interventions [19-21]. The prevalent approach of implementing prescheduled interventions often results in a misalignment between users’ initial motivations and their subsequent behavioral patterns, sparking a feeling of helplessness or annoyance in users and ultimately compromising the long-term effectiveness of these tools [16]. The primary challenge lies in developing interventions that not only initiate but also maintain behavioral change while preserving user autonomy throughout the process.

User retention is a well-documented problem across mobile health apps [22], with motivational decline identified as a primary abandonment factor [23]. Within DSCTs, specifically, excessive or unreliable interventions might precipitate tool abandonment. Current literature identifies 2 critical research priorities: implementing effective blocking strategies and encouraging users to maintain properly configured nudge settings that generate meaningful interventions over time. Our study focuses on the latter: users who have disabled their nudge configurations are of particular interest because they represent a subgroup that initially engaged with the DSCT but subsequently became passive, potentially signaling declining readiness to self-regulate.

Understanding what drives users to actively configure and engage with DSCTs requires a theoretical framework for motivation and behavior change. Self-determination theory (SDT) posits that sustained behavior change depends on the satisfaction of 3 basic psychological needs: autonomy, competence, and relatedness [24]. When applied to technology design, SDT suggests that autonomy-supportive interfaces, those that allow users to set their own parameters rather than imposing restrictions, foster more lasting engagement [25,26]. Complementary to SDT, the transtheoretical model frames behavior change as progressing through stages of readiness, from precontemplation to maintenance [27]. In the DSCT context, users who actively configure nudge parameters may be at a different readiness stage than those who have disengaged, and observable in-app behaviors could serve as proxies for these otherwise latent motivational states [28].

Despite growing research on DSCT effectiveness, a critical gap remains: current literature relies primarily on self-reported measures to assess user engagement and motivation, yet the behavioral factors, observable through configuration logs and usage data, that determine whether users continue to engage with nudges over time remain underexplored. There is limited work investigating whether in-app behavioral indicators can predict intervention receptiveness more effectively than traditional self-reported measures. Identifying such indicators is essential for developing adaptive interventions that not only initiate behavioral change but sustain it through continued user engagement.

This study investigates whether prompting passive DSCT users to reconfigure their nudge settings can restore engagement and whether observable behavioral indicators are more strongly associated with intervention receptiveness than self-reported measures. Through a quasi-experimental protocol, we target users who have disabled nudges (or have set their usage threshold excessively high so that nudges are never triggered) and invite them to reconfigure their nudge settings. We address the following 3 research questions (RQs):

The findings aim to inform the design of more effective digital well-being interventions by identifying observable indicators of user engagement and readiness to change.

We structured our analyses around the 3 research questions.

For RQ1, we computed the daily user-nudge interaction ratio for each group and compared 7-day preintervention (November 16‐22, 2024) and postintervention (November 24‐30, 2024) averages across the acceptance subgroup, rejection subgroup, and control group. The preintervention and postintervention windows were selected to maximize temporal separation from the rolling enrollment period (November 22‐24, 2024) while maintaining 7-day averaging; minor overlap on boundary days was mitigated by aggregation. To assess whether the acceptance subgroup change was attributable to the intervention rather than temporal trends, we applied a difference-in-differences (DiD) analysis at the daily aggregate level, using permutation tests (10,000 iterations) to derive P values. This approach compares each subgroup’s pre-to-post change against the control group’s change.

For RQ2, we analyzed the time-series of the user-nudge interaction ratio across a 5-week window (1 week preintervention and 4 weeks postintervention) to assess the sustainability of re-engagement. We compared DSCT configuration parameters (consecutive usage threshold and cooldown length distributions) and manual app blocking interactions between the acceptance and rejection subgroups at 2 time points (preintervention: November 20, 2024; postintervention: November 29, 2024) using independent-samples t tests with Cohen d as a measure of effect size.

For RQ3, we compared 2 categories of indicators between the acceptance and rejection subgroups: (1) observable behavioral indicators—consecutive usage threshold and cooldown length (independent-samples t tests) and daily goal selection coded as binary leisure versus nonleisure (chi-square test with Cramer V)—and (2) traditional self-reported measures—onboarding survey responses on screen time goals, scrolling regret frequency, and emotional responses (independent-samples t tests) and postintervention survey responses. By comparing the statistical significance of behavioral indicators against self-reported measures, we assessed which category better differentiates users who accept from those who reject the reconfiguration prompt.

All statistical analyses were conducted by exporting Mixpanel data into a Jupyter Notebook environment. The study applied a statistical significance threshold of P<.05. All t tests were 2-tailed. Effect sizes were calculated using Cohen d for continuous variables and Cramer V for categorical comparisons.

This study was considered a service evaluation of a commercially available app, analyzing anonymized and nonidentifiable in-app usage data. As such, it was not subject to the requirements for formal ethics committee approval, consistent with institutional guidelines for nonclinical research involving deidentified data. No personally identifiable information was collected, and participants could not be reidentified by researchers due to the app’s tokenization system, which is compliant with Apple’s privacy guidelines for app monitoring. Participants, or their legal representatives in the case of minors, consented to the use of their anonymized data for research purposes through the app’s terms of service and onboarding process. All data were stored and processed in compliance with applicable privacy regulations, and no compensation was provided to participants.

Analysis of participant demographics (Table 1) confirmed alignment with the broader DSCT user base. The sample was predominantly young, with 51.4% (71/138) of experimental participants and 57% (65/114) of control participants aged 18 to 24 years. Among the 137 (54.4%) participants who reported gender, 86 (62.8%) identified as male. Participants were geographically distributed across 55 countries, with the largest concentrations in North America and Europe. No significant demographic differences were found between the experimental and control groups (all P>.05; Table 1), validating the initial randomization. Demographic distributions were also consistent between the acceptance and rejection subgroups within the experimental condition (age: P=.68; gender: P=.33; geography: P=.19), demonstrating that demographic characteristics did not influence participants’ decisions to accept or reject the prompt.

The 46% (63/138) acceptance rate for nudge reconfiguration among passive users indicates that nearly half possess latent readiness to re-engage with DSCTs when prompted. The DiD analysis confirms that the acceptance subgroup’s increase in user-nudge interaction (DiD=+36.3 percentage points, P<.001) is attributable to the intervention rather than temporal trends, as the rejection subgroup’s decline was not significantly different from the control group’s decline (DiD=−1.4 percentage points, P=.82). The comparable preintervention baselines between subgroups further support the conclusion that the observed change is driven by the intervention itself.

The preintervention behavioral differences between subgroups are particularly noteworthy. Acceptance subgroup participants configured 21.53% shorter consecutive usage thresholds and descriptively longer cooldown periods (+20.56%, though not statistically significant) compared with the rejection subgroup, even before receiving the intervention. The widening of the consecutive usage threshold gap postintervention (Cohen d increasing from −0.47 to −0.67) indicates that the prompt amplified existing behavioral differences rather than creating new ones. The spillover effect observed in manual app blocking behaviors (Cohen d increasing from 0.39 to 0.49) further suggests that successful re-engagement with one DSCT feature may catalyze broader self-regulatory behaviors. However, the gradual decline in user-nudge interaction after approximately 2 weeks, converging with control group levels by 1 month, demonstrates the limitations of single-prompt interventions and aligns with SDT: extrinsic prompts alone cannot sustain behavioral change without corresponding intrinsic motivation [24].

The most striking finding is the disconnect between self-reported measures and behavioral indicators. Despite no significant differences in stated screen time goals, scrolling regret frequency, or emotional responses between the acceptance and rejection subgroups (all P>.1), observable behavioral indicators—shorter consecutive usage thresholds (P=.03) and lower selection of leisure-oriented daily goals (P=.001)—clearly differentiated those who would accept the reconfiguration prompt. Even the postintervention survey, which directly captured participants’ rationales, showed no significant differences (P>.05). This intention-behavior gap [35] underscores the limitations of relying on self-reported data to characterize DSCT engagement and highlights the value of DSCT configuration logs and in-app behavioral choices as proxies for readiness to change.

Our findings challenge conventional approaches to DSCT user engagement. The 46% (63/138) acceptance rate among passive users and subsequent boost in user-nudge interaction challenge the binary classification of users as either engaged or disengaged, suggesting instead that engagement exists on a continuum with opportunities for reactivation even among seemingly passive users. This perspective encourages DSCT developers to implement proactive re-engagement strategies rather than simply accepting attrition as inevitable.

These proactive re-engagement strategies have proven more effective for users with higher readiness to change. Our preintervention data revealed that acceptance subgroup participants configured significantly shorter consecutive usage thresholds (−21.53%, P=.03), with a directionally consistent but nonsignificant difference in cooldown length (+20.56%), indicating that observable DSCT configuration behaviors are significantly associated with intervention receptiveness. DSCTs aiming to offer personalized experiences should leverage such behavioral markers to identify users most likely to benefit from re-engagement prompts, thereby optimizing intervention effectiveness.

However, the temporary nature of increased engagement reveals a critical limitation. While our intervention successfully boosted the acceptance subgroup’s 7-day average user-nudge interaction ratio from 29.7% to 58.5% (peak of 65% on day 1), convergence with the control group after approximately 1 month demonstrates that single-prompt interventions alone cannot sustain behavioral change. Sustainable engagement requires progressive feature designs that transition users from external prompts to self-directed regulation, perhaps through adaptive goal setting, competence-building feedback, and autonomy-supportive interfaces.

Throughout these design considerations, the preservation of user autonomy emerges as a fundamental principle. DSCTs must navigate the inherent tension between providing behavioral support and avoiding paternalistic control. This delicate balance requires thoughtful intervention framing, opt-in mechanisms, customizable parameters, and transparent communication about the rationale behind nudges and restrictions. Only by respecting user agency can DSCTs foster the intrinsic motivation necessary for lasting behavior change, as our divergent group outcomes clearly demonstrate.

This study has several limitations that inform the interpretation of our findings. Although the DiD analysis strengthens causal inference by controlling for temporal trends shared across groups, our quasi-experimental design cannot fully rule out self-selection effects, as preexisting behavioral differences between acceptance and rejection subgroups—particularly in nudge configurations—suggest self-selection dynamics [36] that complicate attribution of outcomes solely to the intervention. While these baseline differences provide valuable insights about readiness indicators, they prevent definitive conclusions about intervention causality.

Beyond design constraints, our reliance on behavioral proxies for psychological constructs introduces measurement limitations. Although usage patterns and configuration choices offer objective indicators, they cannot fully capture the complexity of readiness to change. Moreover, onboarding survey data used to assess baseline attitudes may not reflect participants’ psychological states at the time of the experiment, as these surveys were collected at app installation, which for some participants occurred months before study enrollment.

Additionally, generalizability constraints arise from our sample characteristics. Participants were drawn exclusively from a single DSCT (Detox) with a specific design philosophy, predominantly young users (with over half aged 18‐24 y and approximately one-quarter aged <18 y), and geographic concentration in North America and Europe. These factors may limit the applicability of our findings to other DSCTs, age groups, or cultural contexts in which digital self-regulation practices differ. An important point to note is that our sample consisted exclusively of users who voluntarily installed a DSCT to reduce screen time, suggesting higher baseline motivation than the general population. This limits generalizability to individuals who have not actively sought digital self-regulation tools.

Furthermore, we cannot determine whether the decline in user-nudge interaction observed in Figure 3 represents a successful change in behavior that reduces reliance on the tool or disengagement from self-regulation efforts, as we lack direct measures of participants’ real-world digital well-being behaviors outside the DSCT environment.

Our findings illuminate critical avenues for advancing DSCT effectiveness research. The convergence of acceptance and control groups after 1 month raises fundamental questions about intervention durability. This pattern suggests that single-prompt interventions have limited sustainability, pointing toward the need for longitudinal studies investigating optimal intervention cadences. Research should test whether periodic re-engagement at empirically determined intervals can maintain elevated user engagement without inducing prompt fatigue.

Building on our identification of behavioral readiness indicators, predictive modeling represents a promising research direction. While our study identified several behavioral indicators associated with acceptance, comprehensive models incorporating temporal patterns, contextual factors, and usage trajectories could enable dynamic, personalized intervention timing. Machine learning approaches could identify complex behavioral signatures that signal optimal intervention moments, moving beyond static threshold rules to recognize when users are most receptive to change. Future research may integrate comprehensive psychological assessments, ecological momentary assessments of human-device interaction, objective and longitudinal measures of smartphone and app use (eg, usage logs), as well as contextual data on physical and social environments to better understand how these factors jointly influence digital well-being.

Furthermore, comparative effectiveness research across theoretical frameworks is essential for establishing evidence-based DSCT design principles. Our study examined 1 SDT-based approach, but the field would benefit from systematic comparisons of different intervention philosophies. Understanding which approaches work best for different user segments, cultural contexts, and problematic usage patterns could guide the development of more personalized and effective interventions.

These research directions collectively aim to transform DSCTs from static tools to adaptive systems that recognize and respond to users’ evolving motivational states while preserving the autonomy essential for lasting behavioral change.

This study demonstrates that prompting passive DSCT users to reconfigure their nudge settings can effectively restore engagement (RQ1), with effects persisting for approximately 1 month before converging with baseline levels (RQ2). Observable behavioral indicators—including shorter consecutive usage thresholds (P=.03) and lower selection of leisure-oriented daily goals (P=.001)—differed significantly between participants who accepted versus rejected the intervention, whereas self-reported measures showed no such differences (RQ3). This intention-behavior gap underscores the value of in-app behavioral data over self-report measures for understanding user engagement with digital well-being interventions.

Our findings suggest that the future of effective DSCT design lies not in creating more restrictive or feature-rich tools but in developing adaptive systems tailored to the needs of specific population clusters and based on SDT principles that recognize and respond to users’ motivational states while preserving their autonomy, increasing their competence and reinforcing relatedness. As digital well-being becomes an increasingly critical public health concern, understanding the psychological mechanisms underlying successful self-regulation in digital contexts becomes essential for developing interventions that create lasting positive change.