Is happiness just a click away? Lessons learned from an unguided online well-being intervention for the general adult public Authors: Sara Claes, Sophie Vandepitte, Els Clays, Lieven Annemans Published: 2026 ABSTRACT AbstractBackgroundWell-being websites and apps are increasingly available. Given their scalability and accessibility, online interventions are generally considered a promising avenue to promote public subjective well-being. However, the success of online interventions in public mental health promotion depends on widespread acceptance and sustained engagement. This study explores the feasibility, acceptability, and preliminary effectiveness of an unguided online intervention for non-clinical Belgian adults.MethodsIn this quasi-experimental study, 208 participants were assigned to either the intervention group (n = 68) or control group (n = 140) based on their own preferences. The online intervention allowed participants to choose from a range of activities, including mindfulness, breathing exercises, walking, and cycling. Data were collected at baseline and after 8 months. Feasibility was assessed via uptake, attrition, and engagement. The assessment of acceptability was based on the Theoretical Framework of Acceptability. Inverse probability of treatment weighting was performed to balance the treatment arms and weighted linear mixed models were used to evaluate improvements in terms of subjective well-being, peace of mind, and meaningfulness.ResultsUptake of the intervention was low (32.7%) and dropout in the intervention group was substantial (63.2%). A major barrier to uptake was competing commitments. Non-completers had higher baseline negative affect (M = 2.54, SD = 0.56) than completers (M = 2.16, SD = 0.61). Among the completers, 81.3% found the intervention acceptable, yet 44.0% reported virtually no engagement. The intervention group showed statistically significant improvements in meaningfulness compared to the control group (β = 0.55, 95% CI [0,11; 0,99], p < 0.05), but not in other outcomes.ConclusionLow uptake, engagement, and adherence remain major obstacles to the success of unguided online interventions. Key lessons learned from this study highlight the need to address barriers to uptake during the design phase, implement strategies to sustain user motivation and engagement, and acknowledge that digital well-being interventions may not adequately support those who need it most. Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s40359-026-05013-3.Keywords: Well-being, Feasibility, Acceptability, Health Promotion, eHealth Background Happiness is more important than money for the majority of people worldwide [1]. This widespread interest in happiness and well-being is directly reflected in the abundance of self-help books and well-being apps available. In parallel, empirical research on happiness has expanded immensely in recent years, inspired by the positive psychology movement, which advocates for a shift in focus from mental illness to human flourishing in psychological practice and research [2, 3]. In scientific literature, happiness is commonly referred to as subjective well-being (SWB), a multidimensional construct that encompasses how people think and feel about their lives [4]. As such, happiness or high SWB is characterized by high satisfaction with one’s life as a whole (evaluative dimension), frequent positive emotional experiences, and infrequent negative emotional experiences (affective dimension) [4]. High SWB is furthermore related to numerous desirable outcomes, including improved productivity, better health, increased life expectancy, and warmer social relationships [3, 5]. Given these benefits, policymakers are also increasingly interested in promoting the subjective well-being of their citizens [6]. Intentional improvements in SWB are indeed possible [7], and numerous studies have identified strategies that can foster human flourishing, including physical activity [8–10] and mindfulness meditation [9, 11, 12]. In recent years, digital adaptation of well-being interventions has become more widespread. Digital interventions indeed appear to offer a promising avenue for enhancing SWB [13, 14]. Given their great accessibility, scalability, and ability to reduce financial, geographical, and social barriers, such interventions could be particularly useful for universal prevention and promotion for the general public [13, 14]. Despite promising results on the beneficial effects of digital tools for promoting SWB, low adherence significantly jeopardizes their potential benefits [13, 15]. Research in real-world settings shows that less than 4% of users engaged with mental health apps for more than 15 days [16]. These low retention rates raise important questions about how these tools are perceived by their intended users. While digitally delivered well-being interventions can reach large and diverse audiences, it remains unclear whether the general adult population is willing to accept these tools and engage with them. Ultimately, lasting well-being improvements not only depend on an intervention’s efficacy, but also on users’ willingness to embrace these tools. While such willingness alone does not guarantee behavior change, acceptability represents a necessary precondition for intervention engagement and, subsequently, positive outcomes. Acceptability, or perceived appropriateness, refers to users’ cognitive and emotional reactions to an intervention [17]. According to Sekhon et al. [17], key factors determining an intervention’s acceptability include its burden and effectiveness, potential opportunity costs, people’s affective attitudes, self-efficacy, understanding of the intervention, and alignment with their values. To date, research on the acceptability of digital mental health interventions has been predominantly centered on clinical populations [18]. While these studies suggest that such digital interventions are considered acceptable, the generalizability of these findings to the general, non-clinical population remains unclear. Clinical populations may differ from the general population in terms of symptom severity, motivation, and intervention engagement [19, 20]. Within clinical populations, severe symptoms could increase the interest in and uptake of DMHIs, as a result of a high need for support [21]. However, symptom severity also serves as an important barrier to actual engagement [21]. While this symptom-related barrier to engagement is less pronounced in non-clinical populations, they may not experience an immediate need for support. Hence, in non-clinical populations, acceptance and uptake may depend more heavily on intrinsic motivation and perceived relevance of the intervention. Such differences warrant attention since characteristics of the population are considered to be related to engagement with DMHIs [22]. Some evidence from non-clinical settings shows ambivalent attitudes among the general public towards e-mental health, suggesting that uptake and acceptance of digital initiatives are indeed not self-evident [23–26]. More favorable attitudes are associated with prior experience, greater knowledge, expected benefits, and perceived convenience of e-mental health [23, 26]. Endorsement of an expert in the field or a health professional could also encourage acceptance [26]. Despite having positive attitudes toward digital interventions, some individuals remain unwilling to use them due to additional concerns or limited motivation to engage with technology [26]. Sustained engagement with DMHIs, particularly in unguided formats, depends largely on users’ self-regulated learning skills or their ability to regulate their thoughts, emotions, and behaviors in ways that strengthen intervention engagement and support positive outcomes [27]. When these skills are limited, unguided DMHIs may feel overwhelming, which can, in turn, hinder engagement. Although characteristics of the individual are key to understanding their usage behavior, acceptance of and engagement with digital tools are increasingly perceived as a continuous interplay among multiple factors [28]. These include individual characteristics (e.g. readiness for change, experienced health needs, etc.), environmental conditions (e.g. family support, internet access, etc.), and features of the intervention platform (e.g. tailored messages, content, etc.). As these factors evolve over time, usage behavior may fluctuate accordingly [28]. In sum, acceptance and use of e-mental health are a complex phenomenon. However, prior acceptability research has primarily focused on digital alternatives for therapy [18]. As such, questions remain about the attitudes toward digital initiatives for promoting well-being and preventing mental health problems in the general adult population. Understanding their uptake and acceptance of preventative and promotional initiatives is particularly important since such interventions are typically designed for individuals with no clinical diagnosis. Therefore, the current study aims to address this gap by investigating the feasibility and acceptability of an unguided online intervention for SWB promotion among a non-clinical sample of Belgian adults. Since prior research on digital well-being interventions focused on short-term outcomes [13], a secondary aim of this study was to preliminarily explore intervention effects in the long run. In today’s world, where well-being apps are widely available to countless people worldwide and can be downloaded in just a few clicks, this study will thus offer timely insights on their potential as a public mental health strategy. Based on our findings, we will offer actionable recommendations for developing more acceptable and effective interventions for promoting well-being on a broader scale. Methods Procedure, participants, and intervention A quasi-experimental study was designed to assess the acceptability and preliminary effectiveness of an unguided web-based intervention on SWB, in terms of life evaluation, and positive and negative affect, in adults. The study was approved by the Ethical Committee of the Ghent University Hospital (BC-10848) but was not preregistered. A quasi-experimental design was chosen for practical and methodological reasons, as randomizing participants against their preferences may result in disengagement or ‘resentful demoralization’ [29]. As a result, these participants could experience lower compliance, while those allocated to their preferred condition may report better outcomes irrespective of the intervention’s efficacy [30, 31]. In addition, this design allowed us to examine self-selection in the intervention, providing valuable insights into the characteristics of individuals who voluntarily engage in such interventions in real-world settings. Finally, an unguided intervention was chosen to reflect the type of digital mental health resources that are most commonly available to individuals seeking to improve their well-being. Moreover, the intervention website directed participants towards pre-existing, commercial digital tools. By doing so, this approach enhanced the ecological validity of the study. The study was conducted in a pre-post design with 2 measurement points. The intervention period between pre- and post-measurement lasted 8 months. A longer intervention period was chosen to address a key gap in the literature as existing research predominantly reported on short-term effects of digital interventions on well-being [13]. Participants were recruited during July and August 2023 via a social media campaign calling for people interested in working on their well-being for a period of 8 months. Interested participants could register on the research website www.happiness23.be (currently not accessible), which provided detailed information on the study. To be eligible for the study, participants needed to be at least 18 years old, live in Belgium, and understand Dutch or French fluently. After registration and providing informed consent, participants could access the online baseline survey. After completing the survey at baseline, all participants received comprehensive information on the intervention and what would be expected of them in both study arms. Based on this information, participants decided whether or not they would participate in the intervention. As such, allocation to either the intervention or control group was based on participants’ own preferences. Next, those who opted for the intervention group were asked which specific activity they wanted to engage in during the study period. Participants could choose between 4 activities: (1) mindfulness meditation, (2) breathing exercises, (3) walking, and (4) cycling. During the intervention period, participants were allowed to change their chosen activity. These specific activities were selected due to their accessibility, making them easy for participants to adopt irrespective of their health or place of residence. Moreover, prior research has shown that these activities could potentially improve SWB [8–12, 32]. Also, these activities can potentially enhance meaningfulness [33] and peace of mind [34], which appear to be important predictors of SWB [35]. Participants were free to determine the frequency and duration of their engagement in the activity and could change their selected activity at any point during the intervention period. To facilitate the implementation of their activity of choice in their daily life, participants received additional practical information, including a description of the activity and required materials, alongside links to pre-existing online tools, such as commercial apps, websites, and videos. Psychoeducation was not offered to the participants as part of this online intervention. The web-based intervention was not grounded in a specific therapeutic model nor did it follow a modular or sequential structure. Instead, participants were allowed to independently explore and engage with the available resources suggested for the different activities. The intervention website thus served as catalogue of pre-existing resources that the participants could use to engage in the activities. An overview of the resources made available to the participants is presented in Appendix 1. Screenshots of the study websites (in Dutch) are provided in Appendix 2 to illustrate its appearance. The selection of these tools was guided by four criteria: they had to be affordable, relevant to the activity, available in Dutch, French or English, and easy to use. Access to the study website as well as most of the proposed digital tools were free of charge, and no compensation was offered to the participants. To clarify any questions concerning the website or the intervention activities, participants could contact the research team by email. Email reminders were automatically sent every month, reminding participants to engage in their chosen activities. The reminders were unrelated to how often participants visited the website or the specific activity they had chosen. No additional guidance or individualized support was offered throughout the intervention period. Participants in the control group did not receive access to the intervention materials and were instructed to maintain their usual activities throughout the study period. Measures Data on SWB (in terms of life evaluation, positive affect, and negative affect), meaningfulness and peace of mind were obtained at baseline and post-intervention. The baseline questionnaire covered sociodemographic information as well as key determinants of SWB as identified in a study among Belgian adults [35]. In the intervention group, the frequency of engagement in the activity as well as their subjective experiences of the intervention were assessed post-intervention. In the control group, reasons for non-uptake were assessed in the post-questionnaire. Intervention feasibility Engagement In this study, engagement was defined as the frequency of practicing one’s activity of choice throughout the intervention period rather than engagement with the intervention website itself. This conceptualization was chosen because the intervention website directed participants toward pre-existing resources. Hence, activity engagement was a more relevant indicator than platform use. Engagement was assessed using a single item that asked how often they engaged in their activity of choice during the intervention period. Responses were recorded on a 5-point Likert scale ranging from 1 ([almost] never) to 5 ([almost] daily). Attrition The proportion of participants who failed to complete the post-test or follow-up measurement was used to evaluate attrition while accounting for those who indicated at post-assessment that they had discontinued the intervention. Reasons for non-uptake Participants in the control group were asked to indicate their reason(s) for not participating in one of the proposed activities. Participants could select one or more of the following options: “I am already happy”, “I don’t have time for it”, “I am already doing the proposed activities”, “I am already doing other activities”, “Lack of alignment with other priorities or goals”, “The activities did not appeal to me”, “I don’t think I would be able to do such an activity”, “Due to financial reasons”, or “Other reasons”. Acceptability The acceptability of the intervention was assessed using the Theoretical Framework of Acceptability (TFA) [17, 36]. On a 5-point Likert scale ranging from 1 to 5, participants provided information on their affective attitude, experienced burden, the intervention’s ethicality and coherence, perceived self-efficacy, and opportunity cost. Besides these subdomains, participants evaluated the overall acceptability of the intervention as well. To allow a nuanced understanding of participants’ perceptions, all TFA subdomains as well as the overall acceptability rating were analyzed independently. Preliminary effectiveness outcomes Life evaluation The evaluative component of SWB was assessed with the Cantril Ladder [37, 38]. This single-item question asks participants to imagine a ladder with steps numbered from zero at the bottom to ten at the top. The bottom of the ladder represents the worst possible life they can imagine, whereas the top of the ladder represents the best possible life. Participants are then asked on which step they feel they personally stand at the present time. Positive and negative affect The Positive and Negative Affect Schedules (PANAS) was used to measure the affective component of SWB [39]. This validated scale encompasses 10 positive and 10 negative emotional states. Participants rate how often they experience each emotional state on a Likert scale from 1 (never) to 5 (always). Positive affect and negative affect were calculated by taking the average of the positive emotion items and negative emotion items, respectively. The positive emotion items showed excellent reliability at baseline (α = 0.92) and follow-up (α = 0.89). For the negative emotion items, Cronbach’s alpha equaled 0.85 at baseline and 0.87 at follow-up. Meaningfulness In line with Vandepitte et al. [35], meaningfulness was assessed using two items: “my life has purpose” and “what I do in life is meaningful”. These items were scored on a Likert scale ranging from 1 (completely disagree) to 7 (completely agree). A score for meaningfulness was calculated by taking the average of both items. In our sample, Cronbach's alpha was equal to 0.85 and 0.77 for pre- and post-assessments, respectively. Peace of mind Peace of mind (internal state of peacefulness and harmony) was identified with the 7-item Peace of Mind Scale [40]. Example items are “My mind is free and at ease” and “I feel content and comfortable with myself in daily life”. Each item is rated on a 5-point Likert scale ranging from 1 (not at all) to 5 (all of the time). Peace of mind was equal to the average of the items. The scale demonstrated excellent internal consistency at baseline (α = 0.92) and at follow-up (α = 0.93). Control variables The choice of control variables were informed by the findings of Vandepitte et al. [35] who examined the factors related to SWB in a large sample of the general adult population in Belgium. A detailed description of how we addressed confounding is provided in Appendix 3. Sociodemographics Information about the following sociodemographic variables was obtained: age, gender, educational level, and occupational status. Educational level was divided into 2 categories: low or middle education (from early childhood education to post-secondary non-tertiary education) and high education (from short-cycle tertiary education to doctoral degree or equivalent). Occupational status encompassed the following categories: professionally active, not professionally active (i.e. unemployed, unable to work, and housewife/husband), retired, student, and other professional situation. Satisfaction with financial situation We asked participants how satisfied they were with their financial situation on a Likert scale ranging from 0 (totally unsatisfied) to 10 (totally satisfied) [41]. Subjective health Based on the EuroQol Visual Analogue Scale (EQ-VAS), subjective health was assessed with a Likert scale ranging from 0 to 10, whereby ‘0’ represented the worst imaginable health and ‘10’ the best imaginable health [42]. Sleep quality To collect data on participants’ sleep quality, the Single-Item Sleep Quality Scale was used [43]. On a Likert scale, participants are asked to rate their sleep quality from 0 (terrible) to 10 (excellent). Basic psychological needs The Self-Determination Theory identifies three basic psychological needs essential for human well-being and flourishing: autonomy (the experience of psychological freedom), competence (sense of mastery and personal effectiveness), and relatedness (feeling connected to important others) [44]. These needs were assessed with a shortened 12-item version of the Basic Psychological Need Satisfaction and Frustration Scale (BPNSF) [45, 46]. This scale combines both frustration (negatively formulated) and satisfaction items (positively formulated), scored on a 5-point Likert scale ranging from 0 (never) to 4 (always). For each psychological need, a score was calculated by subtracting the negative items from the positive items, resulting in a score ranging from −8 to + 8. Negative scores reflect greater frustration relative to satisfaction, whereas positive scores indicate greater satisfaction relative to frustration for the specific need. The internal consistency of the autonomy items was acceptable at pre- (α = 0.78) and post-assessment (α = 0.74). The relatedness subscale demonstrated good internal consistency: Cronbach’s alpha was 0.80 and 0.89 at pre- and post-assessment, respectively. Lastly, Cronbach’s alpha equaled 0.83 at baseline and 0.77 at follow-up for the competence items. Power calculation Prior to data collection, a sample size calculation was conducted for the primary outcome using G*Power. Based on a meta-analysis of positive psychology interventions, an effect size of 0.22 was defined [47]. For digital interventions, a recent meta-analysis reported a similar effect size of 0.23 on SWB [13]. To achieve a power of 80% and a significance level of 0.05, each treatment arm required 257 participants. However, despite extensive recruitment efforts, this target was not met. As such, we performed a post-hoc analysis to estimate the power of our study, which revealed that our study had a power of only 46% at baseline. Statistical analysis Statistical analyses were performed in R version 4.3.3 and SPSS version 29. To establish a clear profile of the study participants, descriptive statistics were conducted. Potential baseline differences between both study arms were assessed using independent sample t-tests for normally distributed continuous variables and Pearson’s chi-square tests for categorical variables. Since group allocation was not performed at random, inverse probability of treatment weighting (IPTW) was performed to reduce confounding when evaluating changes in outcomes 8 months later [48]. In a first step, a propensity score was calculated which represents the probability of exposure (i.e. participating in the intervention) given an individual’s characteristics. Next, each individual is assigned a weight that is equal to the inverse of their estimated probability of being in the intervention group. This procedure results in a pseudopopulation in which the measured confounders are balanced across both treatment arms [48]. To ensure appropriate adjustment of confounding, the propensity score model should include all relevant baseline characteristics that influence the relationship between the exposure and the outcome. Also, covariates that are associated with the outcome should be included. To inform the selection of variables in the propensity score model, the authors designed a causal Directed Acyclic Graph (DAG) [48, 49]. This DAG was developed after data collection was completed and was primarily inspired by previous research findings on predictors of SWB in a large sample of Belgian adults [35]. A detailed description of the development of the DAG that informed our propensity score model is provided in Appendix 3. Appendix 4 shows the developed DAG for each outcome variable. The included variables in the propensity score model were: age, gender, educational level, professional status, satisfaction with financial situation, subjective health, sleep quality, and the basic psychological needs of autonomy, relatedness, and competence. To handle complete separation by gender, resulting from the absence of non-binary or transgender participants in the intervention group, the propensity score model was calculated with Firth’s Bias-Reduced Logistic Regression with the ‘logistf’ package. Secondly, potential intervention effects at 8 months were investigated using weighted linear mixed models following an intention-to-treat (ITT) approach. As such, participants were analyzed in their originally assigned groups, regardless of their level of engagement or adherence to the intervention. Separate models were defined for each dependent variable, with time (baseline and post-intervention) and group (intervention or control), and their interaction as fixed factors. Participant ID was included as a random factor. Since IPTW weighting artificially inflates the sample size and induces a correlation within each individual, the entire procedure (including weights estimation and calculation of the linear mixed model) was bootstrapped with 1,000 resamples for every outcome to estimate the standard error, confidence intervals, and p-values. A p-value of 0.05 was considered significant. The ‘lme4’ package was used to fit the weighted linear mixed models. Besides preliminary efficacy, we explored the feasibility and acceptability of the intervention using descriptive statistics. Results Sample description and inverse probability of treatment weighting Figure 1 provides an overview of the participant flow. In total, 208 participants completed the baseline questionnaire. At baseline, no significant differences were found between the intervention and control group, except for age (t(206) = 2.52, p = 0.012). Participants choosing the intervention (M = 44.31, SD = 13.77) were slightly younger than those in the control group (M = 49.45, SD = 13.80). The baseline characteristics of the participants and the standardized mean differences (SMD) before and after weighting are presented in Table 1. After IPTW, all measured covariates were well-balanced across the intervention and control groups. Only for gender, the balance could not be improved for the category “X” as a result of complete separation, since no individuals in the intervention group identified as transgender or non-binary. Additionally, the distributions of all covariates were similar across the treatment arms after IPTW. Flow of participants Table 1Baseline characteristics of the study participantsCharacteristicsContinuous: Mean (SD)Categorical: % (n)Before weightingSMD after weightingControl Group N = 140Intervention Group N = 68Total N = 208p-valueSMDAge49.45 (13.80)44.31 (13.77)47.77 (13.97).0060.37−0.011Gender Male28% (39)27% (18)27% (57).5900.02−0.06 Female71% (99)73% (50)72% (149)−0.040.04 X1% (2)0% (0)1% (2)0.120.14Education level Low or middle17% (24)13% (9)16% (33).5500.110.06 High83% (116)87% (59)84% (175)−0.11−0.06Professional status Professionally active69% (96)75% (51)71% (147).191−0.14−0.01 Not professionally active6% (9)12% (8)8% (17)−0.200.01 Retired15% (21)6% (4)12% (25)0.28−0.03 Student6% (8)6% (4)6% (12)−0.010.03 Other4% (6)1% (1)3% (7)0.160.12Satisfaction with financial situation6.64 (2.14)6.69 (1.86)6.66 (2.05).434−0.03−0.05Subjective health7.56 (1.44)7.50 (1.36)7.54 (1.41).3790.040.01Sleep quality6.62 (2.12)6.38 (1.76)6.54 (2.00).2110.12−0.02Autonomy2.30 (2.93)1.85 (3.30)2.15 (2.89).1620.15−0.05Relatedness4.26 (2.88)4.46 (2.93)4.32 (2.89).327−0.070.01Competence3.82 (3.00)3.16 (3.15)3.60 (3.06).0760.22−0.07 Baseline characteristics of the study participants Attrition Table 2 shows a detailed comparison of baseline characteristics between completers and non-completers. Post-assessment completion rates were 36.8% (n = 25) for the intervention group and 45.0% (n = 63) for the control group. Baseline levels of life evaluation (t(206) = −0.621, p = 0.268), positive affect (t(206) = 0.025, p = 0.480), and negative affect (t(206) = 0.319, p = 0.375) were not significantly different between those who dropped out and those who did not. Also, no statistically significant differences between baseline levels of peace of mind (t(206) = −0.871, p = 0.192) and meaningfulness (t(206) = −0.188, p = 0.425) were found between completers and non-completers. Completers and non-completers did not differ significantly in terms of gender, educational level, professional status, or age. Also, we found no significant differences for the other confounders included in our propensity score model. Table 2Baseline differences between completers and non-completers in the total sample and intervention groupTotal sampleIntervention groupVariableCompleters (n = 88)Non-completers (n = 120)Test statisticp-valueCompleters (n = 16)Non-completers (n = 82)Test statisticp-valueAge48.98 (14.09)46.88 (13.87)t(206) = −1.07.28645.94 (12.71)43.81 (14.16)t(66) = −0.54.592Gender Male30.7% (27)25.0% (30) χ2(2) = 2.18.33618.8% (3)28.8% (15) χ2(1) = 0.64.423 Female69.3% (61)73.3% (88)81.2% (13)72.1% (37) X0.0% (0)1.7% (2)0% (0)0% (0)Education level Low or middle15.9% (14)15.8% (19)χ2(1) < 0.001 .9986.3% (1)15.4% (8) χ2(1) = 0.89.346 High84.1% (74)84.2% (101)93.7% (15)84.6% (52)Professional status Professionally active67.0% (59)73.3% (88) χ2(4) = 2.22.69668.8% (11)76.9% (40)χ2(4) = 1.99.738 Not professionally active8.0% (7)8.3% (10)12.5% (2)11.5% (6) Retired15.9% (14)9.2% (11)12.5% (2)3.8% (2) Student5.7% (5)5.8% (7)6.3% (1)5.8% (3) Other3.4% (3)3.3% (4)0% (0)1.9% (1)Satisfaction with financial situation6.57 (2.26)6.72 (1.88)t(205) = 0.54.5926.44 (2.37)6.77 (1.69)t(66) = 0.62.536Subjective health7.50 (1.45)7.58 (1.38)t(206) = 0.38.7058.06 (1.69)7.33 (1.20)t(66) = −1.94.057Sleep quality6.59 (2.02)6.51 (2.01)t(206) = −0.29.7706.94 (1.98)6.21 (1.66)t(66) = −1.46.149Autonomy2.26 (3.13)2.08 (3.01)t(206) = −0.43.6652.94 (3.89)1.52 (3.06)t(66) = −1.52.134Relatedness4.58 (2.82)4.14 (2.94)t(206) = −1.08.2825.63 (2.66)4.10 (2.95)t(66) = −1.86.068Competence3.72 (3.03)3.53 (3.09)t(206) = −0.44.6584.13 (3.05)2.87 (3.14)t(66) = −1.41.163Life evaluation7.26 (1.46)7.13 (1.48)t(206) = −0.62.2687.56 (1.37)6.92 (1.25)t(66) = −1.75.085Positive affect3.62 (0.60)3.63 (0.67)t(206) = 0.03.9803.61 (0.67)3.51 (0.58)t(66) = −0.55.587Negative affect2.37 (0.58)2.40 (0.56)t(206) = 0.32.3752.16 (0.61)2.54 (0.56)t(66) = −2.32.023Peace of Mind3.44 (0.74)3.35 (0.75)t(206) = −0.87.1923.60 (0.81)3.21 (0.74)t(66) = −1.81.075Meaningfulness5.24 (1.20)5.20 (1.38)t(206) = −0.19.8515.31 (1.42)4.97 (1.27)t(66) = −0.92.363 Baseline differences between completers and non-completers in the total sample and intervention group To better understand the characteristics of those who completed the intervention, we also explored baseline differences between completers and non-completers within the intervention group. Within the intervention group, a dropout rate of 63.2% (n = 43) was observed between pre- and post-assessment. Additionally, nine participants indicated in the post-questionnaire that they had discontinued the intervention. Consequently, only 23.5% (n = 16) of the participants completed the intervention. Non-completers experienced more negative emotions at baseline (M = 2.54, SD = 0.56) compared to those who completed the intervention (M = 2.16, SD = 0.61) (t(66) = 2.32, p = 0.023). For life evaluation, a marginally significant difference was found between completers (M = 7.56, SD = 1.37) and non-completers (M = 6.92, SD = 1.25) (t(66) = −1.75, p = 0.085). Positive affect did not differ significantly between both groups at baseline (t(66) = −0.55, p = 0.587). Additionally, a marginally significant difference between baseline levels of peace of mind was found (t(66) = −1.808, p = 0.075), whereas meaningfulness was not significantly different at baseline for completers and non-completers (t(66) = −0.916, p = 0.363). Within the intervention group, there were no significant differences between completers and non-completers in terms of gender (χ2(1) = 0.641, p = 0.423), educational level (χ2(1) = 0.889, p = 0.346), professional status (χ2(4) = 1.989, p = 0.738), or age (t(66) = −0.538, p = 0.592). Lastly, regarding the other confounders of our propensity score model, we found a marginally significant difference between completers and non-completers for subjective health (t(66) = −1.938, p = 0.057) and relatedness (t(66) = −1.855, p = 0.068). Acceptability A comprehensive overview of participants’ ratings of the intervention acceptability is provided in Fig. 2. Among those who completed the intervention (n = 16), the intervention was generally perceived as acceptable (M = 4.19, SD = 0.75). The majority (81.3%; n = 13) said they liked the intervention (responding “liked” or “strongly liked”). The perceived burden was moderate (M = 2.31, SD = 1.14), although 18.8% of participants (n = 3) reported that the intervention required substantial effort. More than half of the participants (56.6%; n = 9) considered the intervention suitable for the general adult public (responding “agree” or “strongly agree”), and half of the participants (n = 8) indicated understanding how the intervention could contribute to their overall well-being (responding “agree” or “strongly agree”). Additionally, most participants (75.0%; n = 12) expressed confidence in their ability to engage in the intervention activities (responding “confident” or “very confident”). Lastly, opinions regarding the opportunity cost varied (M = 2.62, SD = 1.5), and 31.3% of participants (n = 5) reported that the intervention interfered with other priorities (responding “agree” or “strongly agree”). Mean, SD, and proportion for intervention acceptability (n = 16) Engagement Post-assessment data revealed low engagement with the intervention activities (see Table 3) as 11 out of 25 participants (44.0%) reported virtually no engagement in their activity of choice. Moreover, only 3 participants (12.0%) reported engaging often in their activity, and 4 (16.0%) reported engaging (almost) daily. Table 3Engagement rates in the intervention group (n = 25)How often did you engage in your activity of choice?% (n)(Almost) never44.0% (11)Rarely12.0% (3)Sometimes16.0% (4)Often12.0% (3)(Almost) daily16.0% (4) Engagement rates in the intervention group (n = 25) Reasons for non-uptake Engagement in other activities was the primary reason for non-uptake. In particular, 27 out of 47 participants in the control group (57.5%) reported this barrier. Additional frequently reported barriers were already feeling happy (n = 14; 29.8%) and lack of alignment with priorities (n = 11; 23.40%). Additionally, a smaller proportion (n = 7; 14.9%) indicated that the activities did not appeal to them. Financial barriers (n = 4; 8.5%), time constraints (n = 3; 6.4%), and engagement in similar activities (n = 3; 6.4%) were less frequently mentioned. However, no baseline differences were observed in self-reported engagement in these activities between participants who chose to participate in the intervention and those who did not (see Appendix 5). Lastly, 2 participants (4.3%) cited an unspecified reason for non-uptake. Preliminary effectiveness of the online intervention Table 4 presents the results of the weighted linear mixed models for the outcomes and changes in predicted scores are illustrated in Fig. 3. For life evaluation (the evaluative component of SWB), the interaction effect between group and time was not statistically significant (β = 0.15, t(293) = 1.00, p = 0.317), suggesting comparable changes over time in both the intervention and control group. Also for positive and negative affect (the affective component of SWB), the interaction effect was not statistically significant. However, for positive emotions, results revealed a small but significant effect of time (β = −0.99, t(293) = −2.08, p = 0.038), indicating that 8 months later both treatment arms experienced, on average, fewer positive emotions. For negative emotions, the time effect was marginally significant (β = −0.09, t(293) = 1.87, p = 0.063). Moreover, our results show a significant group x time effect for meaningfulness (β = 0.55, t(292) = 2.44, p = 0.015). This suggests that individuals in the intervention group experienced more pronounced improvements in how meaningful they perceived their lives to be compared to those in the control group. Lastly, no significant interaction effect between group and time was found for peace of mind (β = 0.16, t(292) = 0.97, p = 0.333). Table 4Results of weighted linear mixed models for life evaluation, positive and negative affect, peace of mind, and meaningfulnessFixed effectsRandom effectsFixed factorEstimateSEt (df)p95% CICluster groupsVarianceICCLife evaluationIntercept7.220.1072.24 (293) <.001[7,03; 7,42]Subjects1.350.54Group−0.110.15−0.77 (293).443[−0,41; 0,18]Residuals1.14Time−0.140.111.20 (293).233[−0,09; 0,36]Group*Time0.150.231.00 (293).317[−0,14; 0,44]Positive affectIntercept3.650.0485.97 (293) <.001[3,57; 3,74]Subjects0.240.46Group−0.110.07−1.60 (293).112[−0,25; 0,03]Residuals0.28Time−0.990.05−2.08 (293).038[−0,19; −0,01]Group*Time0.270.191.39 (293).166[−0,11; 0,64]Negative affectIntercept2.360.0466.18 (293) <.001[2,27; 2,43]Subjects0.240.61Group0.090.061.40 (293).162[−0,04; 0,22]Residuals0.16Time0.090.051.87 (293).063[−0,01; 0,19]Group*Time−0.140.09−1.56 (293).121[−0,32; 0,04]Peace of MindIntercept3.420.0573.02 (292) <.001[3,33; 3,51]Subjects0.430.64Group−0.110.08−1.31 (292).190[−0,28; 0,06]Residuals0.24Time−0.050.05−1.02 (292).310[−0,15; 0,05]Group*Time0.160.160.97 (292).333[−0,16; 0,48]MeaningfulnessIntercept5.280.0961.44 (292) <.001[5,11; 5,45]Subjects1.060.51Group−0.180.14−1.33 (292).183[−0,45; 0,09]Residuals1.02Time−0.150.13−1.19 (292).238[−0,41; 0,10]Group*Time0.550.222.45 (292).015[0,11; 0,99] Results of weighted linear mixed models for life evaluation, positive and negative affect, peace of mind, and meaningfulness Standardized predicted levels and 95% confidence intervals for well-being outcomes at pre- and post-intervention per treatment arm Discussion The aim of this study was to investigate the feasibility, acceptability, and preliminary efficacy of an unguided web-based intervention for the general adult population designed to promote their subjective well-being (SWB). In line with previous research, our results demonstrate that low uptake and adherence pose a great challenge to the potential of digital interventions for promoting well-being [15, 16]. Of the 208 participants who signed up for our study, only 68 (32.7%) chose to participate in the intervention. In accordance with prior research [50], our findings demonstrate reluctance toward the uptake of digital intervention within a sample of non-clinical adults. A major barrier to uptake was the presence of competing commitments. In particular, engagement in other activities was reported by 57.5% of the control group as a reason for non-uptake, while conflicting priorities were mentioned by 23.4%. Also, a notable proportion (29.3%) questioned the intervention’s relevance as they declared being already happy. Our result showed no significant differences between those willing to participate in the intervention and those who were not, except for age. The intervention group appeared to be slightly younger than the control group in our study. This finding suggests that digital well-being interventions could appeal to a broad audience rather than a specific subgroup, but are likely to attract a younger public. While some prior studies indeed suggest that older age could be a barrier to the uptake of digital mental health, older individuals typically show better engagement rates [21, 51, 52]. Alongside low uptake, adherence remains a major concern according to our findings, which show a dropout rate of 63.2% within the intervention group. In our study, drop-out was thus considerably higher than the average rate estimated in a meta-analysis of digital well-being-promoting interventions for the general public [13]. Remarkably, while participants who completed the intervention (n = 16) found it generally acceptable, actual engagement levels were far from optimal. Specifically, 44.0% of the participants (n = 25) reported virtually no engagement, and an additional 12.0% engaged rarely in their activity of choice. These low engagement levels are consistent with findings from prior e-mental health research, in both clinical and non-clinical samples [13, 16, 53]. Although our study was not powered to detect significant changes in SWB, the direction of the observed associations was in line with our expectations. Albeit not statistically significant, the intervention group demonstrated improvements across all outcomes. Only for meaningfulness, our results showed that the intervention group experienced statistically significant improvements in their sense of meaningfulness 8 months later compared to the control group. In line with previous research [13], these findings highlight the potential for digital interventions to promote the well-being of the general adult population. It is plausible that with a larger sample size, these effects would have reached statistical significance. Overall, the results of this study show critical obstacles to the feasibility of digital interventions to promote mental health and subjective well-being among the general adult public. Several lessons can be learned from our findings that can inform e-mental health research and practice. Lesson 1: barriers towards uptake must be considered in intervention design While digital interventions have the potential to reach wide audiences, the results indicate reluctance among our sample of Belgian adults to use such tools, which is consistent with prior research [23–26]. Unlike clinical populations, the general adult population does not experience an immediate need for mental health support, potentially limiting the uptake of digital resources [21]. Moreover, competing daily demands present an additional barrier towards uptake. Together with prior evidence [23–26], our study highlights a critical misalignment between what is needed among the broader population and what is offered to them. To address this gap, e-mental health initiatives should be designed with the target population in mind to ensure that content is perceived as relevant and valuable [54]. Co-creation and user participation offer promising strategies to identify important barriers and needs to consider in the development of e-mental health initiatives for the general public [55]. Future research should explore how personalized and tailored intervention strategies and co-creation initiatives improve the uptake and long-term adherence to digital interventions among non-clinical users. Lesson 2: motivation is a dwindling spark In our study, allocation to the intervention group was based on participants’ preferences. Despite their initial willingness to participate in the intervention, actual engagement rates, defined as the frequency of practicing the activity of choice, were low, and substantial drop-out was observed in the intervention group. Our results thus clearly illustrate a discrepancy between participants’ intentions and their actual behavior. This intention-behavior gap could be related to the quality of the original intention as well as self-regulatory challenges at different stages of goal pursuit, including initiation, persistence, and completion [56]. It is also essential that intervention content itself actively supports behavioral change, as therapeutic persuasiveness is key to user engagement [57]. The low engagement rates in the present study may thus reflect insufficient facilitation of behavioral change through the intervention content. In the context of digital interventions, the absence of structured guidance could further enlarge the gap between intention and behavior. Research consistently shows that guided interventions achieve higher engagement rates than unguided ones [21, 58], but offering guidance is more resource-intensive. A promising alternative is the use of automated reminders [59, 60] and gamification [61], which may help sustain participation at a lower cost. Future research should determine the most effective approaches to foster engagement and bridge the intention-behavior gap at different stages of behavior change in self-guided interventions. In this regard, it should be increasingly recognized that self-guided interventions place considerable demands on self-regulated learning as they require users to navigate their own learning trajectory independently [27]. When this capacity to actively direct one thoughts, motivations, and behaviors to improve desired outcomes is limited, sustained engagement with DMHIs is likely compromised. To mitigate this risk, intervention design should aim to minimize cognitive burden and enable mastery experiences through proactive goal setting, progress monitoring, and adaptive feedback [27]. At the same time, it is increasingly recognized that engagement tends to fluctuate. Indeed, engagement is driven by multiple factors related to the individual, the intervention, and the broader context [28]. Changes in any of these factors may result in varying levels of engagement [62]. Such fluctuations warrant consideration in intervention design through, for example, just-in-time support. While sustaining user engagement is considered one of the major challenges for digital interventions, higher engagement is not consistently associated with better outcomes [63]. Therefore, determining the optimal level of engagement necessary to achieve meaningful benefits remains an important area for future investigation. Moreover, beyond their intended benefits, digital well-being interventions could evoke unintended adverse effects. The additional demands placed on participants could contribute to stress, potentially undermining user engagement. As such, it could be interesting to expand the current evidence base with robust qualitative or quantitative research into the potential negative effects of e-mental health. Lesson 3: beware of Matthew The Matthew Effect describes a process of cumulative advantage, whereby better-off individuals continue to prosper whereas disadvantaged individuals are likely to fall further behind [64]. Consequently, disparities between advantaged and disadvantaged individuals will aggravate over time. A similar dynamic can occur in the context of well-being promotion as individuals with higher baseline well-being are more likely to engage with and benefit from such initiatives. In line with prior evidence [21, 65–67], we found that individuals with higher baseline negative affect are less likely to complete the intervention. While it is well-known that low mood and depressive symptoms could hamper adherence to e-mental health [21, 65–67], comprehensive guidelines to mitigate this risk are currently lacking. Future research should therefore aim to establish practical strategies for enhancing intervention usage and adherence among individuals with lower well-being. One potential avenue worth exploring is whether tailoring intervention content and activities based on participants’ mood can enhance engagement and adherence [61]. A greater person-activity fit could indeed contribute to greater well-being gains [68]. Health behaviors are shaped by both individual characteristics as well as broader structural contexts which individuals operate in [69]. Future DMHIs should therefore move beyond targeting individual-level behaviors and instead dynamically capture, interpret and respond to users’ social, spatial, and structural contexts in real time. Promising directions could include the use GPS-based activity patterns to understand environmental exposures or peer support features designed to strengthen social embeddedness of the intervention. Incorporating social determinants of health behavior into intervention design could prevent disengagement, particularly among individuals prone to it. Strengths and limitations To correctly interpret the results discussed above, some strengths and limitations need to be addressed. A first limitation is that our sample size was not sufficient to detect statistically significant changes in SWB despite our a priori power analyses and extensive recruitment efforts. Recruitment took place during the summer of 2023, which may partially explain the suboptimal enrolment in our study. Nevertheless, the direction of the observed trends was in line with our expectations as the intervention group demonstrated improvements across all outcomes, albeit not statistically significant. Small sample sizes are common in feasibility studies, including our own. However, the resulting low statistical power limits the validity of the observed effects, as it increases the likelihood of both Type I and Type II errors [70]. Despite the limited power, we decided to report the preliminary intervention effects because we rigorously controlled for baseline differences between the study arms using IPTW. Our detailed reporting of this method may therefore inform similar intervention studies where randomization is infeasible or undesirable, and help advance the field by increasing familiarity with analytical approaches for causal inference when randomization is absent, such as DAGs and IPTW. Nevertheless, future studies should include a bigger sample size to support our preliminary effectiveness results. Alongside the small sample size, the present study included self-selected participants whose intrinsic motivation and anticipation of well-being improvements may have influenced their reported well-being gains [71, 72]. As such, a sample representative of the general population would improve the generalizability of the observed findings. Third, there was a considerable drop-out in both treatment arms, which may have introduced bias in our results, particularly given that intervention drop-outs reported higher baseline negative affect. Drop-out and low adherence remain critical challenges for digital health, and ongoing research efforts are needed to tackle this issue effectively [15, 73]. To mitigate the impact of missing data, preliminary effectiveness was investigated using weighted linear mixed models since this approach can yield unbiased estimates under the assumption of missing at random (MAR) [74]. However, the validity of this assumption remains uncertain, and future studies could consider additional sensitivity analyses to further assess the robustness of their findings. Fourth, our study design could have introduced selection bias. In particular, group allocation was based on the preferences of the participants. Those who opted for the intervention were thus likely already favorably disposed towards DMHIs, which may have inflated our feasibility and acceptability findings. The acceptability of the intervention may be further overestimated since only participants who completed it provided feedback. As such, the perspectives of those who disengaged remain unknown. This limitation is particularly relevant given the low engagement rates observed in our studies, which suggest the acceptability of our web-based intervention was far from ideal. It is furthermore difficult to interpret these results in the absence of established benchmarks for what is generally considered acceptable. In light of these limitations, it should thus be noted that our findings regarding the acceptability of the intervention should be interpreted with caution. While it is certainly a strength that a validated questionnaire was used to measure acceptability [36], future research could consider a mixed-method design to explore user experiences in greater depth, particularly among those who disengaged early. These qualitative insights could inform the development of more user-centered e-mental health interventions for the general public [55]. Fifth, engagement was measured with self-report data, which can induce social desirability bias or recall bias. This approach was justified since our intervention included offline components such as physical activity. Moreover, as the intervention directed participants to pre-existing, commercial digital tools, we were unable to objectively monitor their engagement. To obtain a more detailed picture of user engagement patterns, future studies could supplement self-report data with objective usage metrics (e.g. login frequency, time spent on the website or app, number of modules completed, etc.). When planning future studies, researchers are advised to select measures of engagement based on their conceptualization of the construct. While objective metrics capture behavioral aspects of engagement, such as frequency, duration, and the variety of features used, qualitative and self-report data can provide valuable insights into the mental state and lived experiences of users while engaging with the intervention [22]. Sixth, the long-term follow-up in the current study is definitely a strength, especially since prior research on digital well-being interventions reported short-term intervention effects. However, additional follow-up assessments, including in the short run (e.g. after 1 month), would help clarify whether the intervention effects persist over time or gradually decline. Alongside changes in intervention effectiveness, future research would benefit from examining fluctuations in engagement over time and how these fluctuations are shaped by the characteristics of the individual, the intervention and the broader context and changes therein. Finally, to guide policymakers in making informed decisions on whether to adopt e-mental health into public mental health programs, cost-effectiveness analyses of digital well-being promotion as well as budget impact analyses should be conducted in future research [75]. Besides these limitations, our study had some important strengths. First, a quasi-experimental design was chosen for both practical and methodological reasons. As previously noted, randomizing participants against their preferences may lead to disengagement [29]. An additional benefit of this design is that we could shed light on the characteristics of individuals who voluntarily participated in the online intervention and assess whether they differed from those who did not. Gaining insights into these characteristics is essential for optimizing the development and implementation of digital well-being interventions in order to attract a broad public. Second, to mitigate the bias and confounding inherent in quasi-experimental designs without random assignment, IPTW was employed. This approach allowed us to balance key baseline characteristics between both groups, thereby approximating the conditions of a randomized controlled trial. By improving comparability, IPTW strengthens causal inferences regarding intervention effects. Nevertheless, it is important to acknowledge that not all theoretically relevant confounders were included in our propensity score model. The DAGs guiding variable selection were developed after data collection. As a result, some confounders were not measured in our study. Additionally, in light of our relatively small sample size, we limited the number of variables in our propensity score model to avoid overfitting. As such, residual confounding may still be present. Third, the focus on a non-clinical sample is a key strength of this study since the acceptability and uptake of e-mental health in the broader population remain underexplored. While clinical treatment is crucial, strengthening mental health promotion and prevention are equally essential for fostering the overall well-being of the population. In conclusion, the present study provides preliminary insights into the potential of an unguided online well-being intervention for the general adult population. While the intervention group showed significant improvements in meaningfulness compared to the control group, no significant changes were observed in other outcomes (i.e. life evaluation, positive affect, negative affect, and peace of mind). Moreover, low uptake, engagement, and adherence remain critical challenges that could jeopardize the success of digital well-being promotion. Addressing these barriers requires additional research efforts to optimize the design and implementation of e-mental health for the general adult public. Replication of these findings in larger samples is furthermore warranted. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Acknowledgements The authors would like to thank Steven Wallaert for his contribution to the data analysis method. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the author(s) used Grammarly in order to control for and address language issues as non-native speakers. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article. Abbreviations Basic Psychological Need Satisfaction and Frustration Scale Directed Acyclic Graphs EuroQol Visual Analogue Scale Inverse Probability of Treatment Weighting Intention To Treat Missingness At Random Positive and Negative Affect Schedules Standardized Mean Difference Subjective Well-being Theoretical Framework of Acceptability Authors’ contributions SC wrote the main manuscript, performed the data analysis, and prepared all figures and tables. EC and LA did supervision. LA did project administration and funding acquisition. All authors conceptualized the paper and reviewed the manuscript. Funding This study obtained funding from NN Belgium. The funding resources have no role in the development of the design, conduct, analysis of this study, and final publication decisions. Data availability The datasets generated and/or analyzed during the current study are not publicly available due to privacy laws, but are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This study has obtained ethical approval from the ethical committee of the University of Ghent (BC-10848) and was conducted in strict accordance with the principles of the Declaration of Helsinki. All participants signed an informed consent before participation. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 5. Data Availability Statement The datasets generated and/or analyzed during the current study are not publicly available due to privacy laws, but are available from the corresponding author on reasonable request.