Many studies have demonstrated the role of healthy lifestyle behaviors in increasing life expectancy. Specifically, the adoption of a healthy diet, maintenance of a healthy weight, cessation of smoking, consumption of alcohol in moderation, and regular exercise are associated with significant reductions in mortality. The researchers discovered that the adoption of each healthy lifestyle behavior contributes to an improved quality of life and enhanced longevity [1].

The number of health apps designed to assist individuals in adopting healthier behaviors has continued to grow steadily, with new apps reaching the market every year. In 2018, there were over 35,000 health apps available for download [2]. The advent of smartphone apps has created new opportunities for engaging in health-promoting behaviors. These apps offer immediate access to health-related information, medication reminders, and tools for monitoring progress, which can collectively support healthier lifestyle adoption [3].

Several scales have been developed to assess the quality of mobile health (mHealth) apps, such as the Mobile App Rating Scale [4] and the App Behavior Change Scale [5]. A key feature shared across these frameworks is the inclusion of personalization as a critical quality dimension. Personalization is widely acknowledged as essential in the design of behavioral interventions. For example, 1 study has demonstrated that messages tailored to the user’s specific characteristics are more likely to be read, recalled, and retained in memory, and are perceived as more personally relevant than untailored messages. Tailored messages also attract greater attention, stimulate more discussion, and are shared more frequently compared to nontailored messages [6].

Personalization of mobile apps can be done according to the user’s profile. In previous studies [7], we identified preference relations between personality traits and specific mechanisms (components that can be integrated into an mHealth app) based on a scoping review of the literature. These findings led to the development of a preference matrix indicating the types of mechanisms preferred according to a user profile. The preference matrix can be used as a framework for the personalization of mobile apps with the aim of encouraging behavioral change. When designing a mobile app intended to support health-related behavior change, designers can draw on this preference matrix to identify the mechanisms most likely to be effective for their target users. For instance, if an individual is classified as extraverted according to the Big Five personality traits, mechanisms that facilitate social comparison or collaboration between users may be particularly relevant [8].

The preference matrix includes 15 mechanisms associated with the Big Five personality traits, which are presented in detail in the table found in Multimedia Appendix 1. For personality profiling, we used one of the most widely applied classification systems, namely the Big Five personality model [8-16].

One of the most widely used instruments for the assessment of personality is the Big Five model. The Big Five model describes an individual’s personality based on 5 distinct traits, as summarized in Table 1.

This model has the advantage of being widely adopted in academic research. Between 1990 and 1994, the number of publications on this topic was approximately 400, while between 2005 and 2009, this figure rose to approximately 1600 [17].

The other advantage of this model is its cross-cultural universality. The Big Five model has been shown to be universal across human populations [18]. Indeed, the authors translated their personality questionnaire, the Revised NEO Personality Inventory (NEO-PI-R), into 6 languages belonging to the same language family as English (German, Portuguese, Hebrew, Chinese, Korean, and Japanese), and the Big Five factors consistently emerged across these versions.

In a subsequent study, researchers examined whether the Big Five personality dimensions, as measured by the NEO-PI-R, were universal across 50 cultures spanning American, European, Arabic, Asian, and African regions [19]. The NEO-PI-R was translated into the local language of each participating country. The participants were primarily students from the country in which the test was administered. Factor analysis demonstrated that the American structure of the NEO-PI-R self-report questionnaire (Form S) could be replicated in all cultures, with 94.4% of factors replicated. Moreover, the general factor structure remained recognizable across all cultural contexts included in the study.

Based on the hypothesis that individuals with different profiles exhibit distinct preferences regarding components of mHealth apps designed to facilitate behavior change, a review of the literature identified 15 mechanisms associated with user profiles [21]. In our context, the term “mechanism” is used to denote the entirety of components that can be incorporated within an mHealth app with the objective of promoting behavioral change. We classified the mechanisms into 2 categories: the mechanisms linked to behavior change techniques and the mechanisms linked to game elements. For further details and definition of mechanisms, see Table 2.

The preference relations between the mechanisms and the Big Five traits identified in the literature enable us to construct a preference matrix summarized in Multimedia Appendix 1. Each cell of the matrix indicates the number of studies in the review reporting a preference relation between a given user profile and a specific mechanism [21]. Most of the preference relations were positive (53/75, 71%), indicating that a particular personality profile tends to prefer a given mechanism (represented by a plus sign in Multimedia Appendix 1). Negative relations, representing an aversion toward a mechanism (represented by a minus sign in Multimedia Appendix 1), accounted for 12% (9/75) of all potential relations. It is important to note that the relation matrix remains incomplete, as only 73% (55/75) of all potential relations were identified in the literature review.

This study aims to corroborate or enrich this preference matrix by experimentally identifying new preference or nonpreference relations between mechanisms and personality traits.

This study was approved by the University of Geneva Research Ethics Committee (CUREG_2021-04-38). This section is based on our previously published protocol study [25]. Participants were required to complete a consent form that detailed the purpose of the study, the procedure to be followed, and their right to withdraw from the study at any time. Confirmation that they had read, understood, and agreed to the terms of the consent form was mandatory before accessing the questionnaire and was required for the investigators to use the submitted data for research purposes. Participation was voluntary, responses were collected anonymously, and only deidentified data were used for research purposes in accordance with applicable data protection regulations. Participants did not receive any financial compensation for their participation.

We conducted a cross-sectional study to address our research objectives. Participants completed an online questionnaire following the CHERRIES (Checklist for Reporting Results of Internet E-Surveys) guidelines (Checklist 1) [26].

The primary outcome was the identification of preferred mechanisms according to the user profile.

The study targeted French-speaking adults aged 18 years and older. Recruitment was conducted both within the University of Geneva community and through social media platforms (eg, Facebook and Twitter, subsequently rebranded X), enabling participation from a broader population beyond university affiliates. The survey was open to anyone meeting the language and age criteria.

We calculated the required sample size using a multiple regression power analysis in R (R Foundation for Statistical Computing). The parameters used were the number of predictors (u=3), effect size (f²=0.07), significance level (α=.05), power of 0.9, and estimated variance of 202.403. These estimates were based on the hypothesis that individuals with higher levels of agreeableness, as measured by the Big Five personality traits, show a preference for social networks [11,24].

To approximate the variance, we used data from a previous study [27] that assessed Big Five personality traits in relation to user preferences for social network posters. Specifically, we relied on the variance in Big Five trait ratings for altruistic participants (n=46) based on their average scores for a blood donation poster, which was evaluated on a 0-to-100 scale. Based on these parameters, we determined that a minimum sample size of 206 participants was required.

Participants completed an online questionnaire developed by the investigators using Qualtrics software (Qualtrics; Multimedia Appendix 2). First, participants were required to complete a consent form that detailed the purpose of the study, the procedure to be followed, and their right to withdraw from the study at any time. Participants were then asked to provide demographic information. To proceed to the remainder of the survey, they were required to confirm that they were aged 18 years or older. Eligible participants subsequently completed two main components of the questionnaire: (1) a standardized personality assessment measuring their Big Five profile, and (2) a task in which they reviewed 15 mechanisms, selected their 5 preferred ones, and rated the extent to which each selected mechanism would motivate them to engage in physical activity on a scale from 0 to 100.

A total of 214 participants completed the questionnaire, including 118 (55.1%) women, 89 (41.6%) men, 5 (2%) individuals identified as “other,” and 2 (1%) participants who did not report their gender. The mean age of the sample was 29.42 (SD 10.41) years. Additional demographic characteristics are presented in Table 3.

Personality trait scores were measured on a 1 to 5 scale. The distribution of the sample across the Big Five dimensions is presented in Figure 1, where the line depicts the median and the cross indicates the mean.

Overall, participants in this study tended to score higher on agreeableness (mean 3.71, SD 1.31), followed by neuroticism (mean 3.12, SD 1.23) and extraversion (mean 2.86, SD 1.17).

A total of 4 mechanisms were selected by more than half of the participants. These mechanisms were self-monitoring, progression, challenge, and quest (Table 4).

The initial matrix (M1) summarizes the preference relations identified in the literature between personality traits and mechanisms. A positive relation indicates that individuals with a given Big Five profile tend to prefer a particular mechanism, whereas a negative relation (aversion) reflects a lower interest or dislike for that mechanism among individuals with that profile. Mixed relations correspond to cases where both preference and aversion were reported across studies, while missing relations denote the absence of evidence. Overall, M1 presents 60% (42/70) positive preference relations, 1% (1/70) aversion relations, 10% (7/70) mixed relations, and 29% (20/70) missing relations (see M1 in Figure 2).

The second matrix (M2) represents the empirical data collected in this study. For each selection mechanism (eg, collection, challenge, or reward), a discriminative score was calculated. This score indicates how strongly a given personality trait differentiates between participants who selected a mechanism and those who did not. M2 shows that 5% (4/70) of mechanisms had a very high discriminative score (>4), 3% (2/70) high (3), 17% (12/70) moderate (2), and 74% (52/70) low (0‐1; see M2 in Figure 2).

The second matrix (M2) represents the empirical data collected in this study. For each selection mechanism—defined as one of the gamified motivational strategies participants could choose (eg, collection, challenge, or reward)—a discriminative score was calculated. This score indicates how strongly a given personality trait differentiates between participants who selected a mechanism and those who did not. M2 shows that 5% (4/70) of mechanisms had a very high discriminative score (>4), 3% (2/70) high (3), 17% (12/70) moderate (2), and 74% (52/70) low (0‐1; see M2 in Figure 2).

A separate logistic regression analysis was performed for each mechanism to determine whether the personality scores predicted the likelihood of selecting that mechanism. In addition, an ordinal logistic regression was conducted using the motivation ratings assigned to the selected mechanisms as the dependent variables and the BFI-10-Fr scores as the predictors. Across analyses, 3 mechanisms yielded statistically significant models.

The objective of this study was to validate the preference matrix derived from a previous literature review by examining whether similar relations could be observed empirically in our sample.

Four mechanisms were selected by more than half of the participants: self-monitoring (n=149), progression (n=127), challenge (n=112), and quest (n=111). No significant associations were found between these mechanisms and the Big Five personality dimensions. However, when comparing these findings with the validated preference matrix (M1), a more nuanced interpretation emerges. According to the matrix (M1), self-monitoring is generally appreciated across all Big Five profiles, with the exception of one study reporting lower preference among highly conscientious individuals. The progression mechanism also appears to be universally preferred, reinforcing its broad motivational relevance. By contrast, the preference matrix (M1) suggests that the quest mechanism is predominantly preferred by conscientious and extraverted individuals, whereas the challenge mechanism tends to be favored by users scoring high in agreeableness, conscientiousness, and extraversion. Taken together, these results indicate that certain gamification mechanisms, particularly self-monitoring and progression, possess strong, cross-profile motivational appeal. These mechanisms should, therefore, be considered foundational elements in the design of mHealth apps, providing a core layer of user engagement upon which more personalized, profile-specific interventions may be developed.

The comparison between matrices M1 and M2 highlights several developments in the identification of potential relationships between mechanisms and Big Five personality traits. In M2, new associations emerge that were not identified in M1, particularly between neuroticism and mechanisms such as prompts and cues, demonstration of behavior, punishment, quest, challenge, and avatar. Similarly, openness shows newly emerging association with prompts and cues, progression, quest, and challenge, while conscientiousness displays novel relations with social comparison and avatar, as well as a particularly strong association with the collection mechanism in M2. These findings reflect a refined and evolving understanding of how specific personality traits may align with distinct mechanisms when examined in an experimental context.

In addition, whereas M1 did not show a clear consensus regarding the relationships involving agreeableness and openness, M2 reveals emerging associations, notably between agreeableness and competition, and between openness and rewards. Conversely, several relationships previously identified in M1 do not reappear in M2. For example, extraversion was associated in M1 with a wide range of mechanisms, including prompts and cues, demonstration of behavior, self-monitoring, punishment, social comparison, social support, progression, competition, cooperation, collection, rewards, and quest, but these associations were not observed in M2. A similar reduction in associations is observed for agreeableness, which in M1 was linked to prompts and cues, demonstration of behavior, punishment, collection, challenge, and especially rewards. Despite these discrepancies, statistically significant results in this study were identified for only 3 mechanisms: collection, competition, and rewards. Of the 5 significant relationships observed, 3 were already present in the preference matrix M1 (conscientiousness-competition, conscientiousness-rewards, and neuroticism-rewards). Two relationships, those involving neuroticism-competition and agreeableness-competition, had shown inconsistent evidence in M1, with both preference and nonpreference relations reported in the literature.

This study contributes to refining the preference matrix in two ways. First, it enriches the matrix by identifying a new significant preference relation between conscientiousness and the collection mechanism. Second, in cases where the literature reports incongruent findings, our results help clarify these relationships by supporting a nonpreference relation for neuroticism and agreeableness with competition.

In the following sections, we examine in greater detail the 3 mechanisms for which significant regressions were obtained. Figure 3 presents the mockups used in the online survey to illustrate the 3 mechanisms preferred by participants according to their personality traits. These visual representations supported participants’ understanding of each mechanism and facilitated more reliable self-selection during the experimental procedure.

In this study, the collection mechanism refers to the accumulation of virtual items or badges over time. Collecting has been identified as a motivational strategy that can support engagement in health-related behaviors, such as maintaining physical activity [28]. Our results show that participants scoring higher on conscientiousness were 1.87 times more likely to select the collection mechanism than those with lower levels of this trait.

These findings are largely consistent with the existing literature, which reports a general preference for collection-based mechanisms across most Big Five personality traits in the preference matrix M1 [9,24,29-31]. Notably, however, none of the previously reviewed studies identified a specific preference for collection among highly conscientious individuals. As such, the present result constitutes a novel contribution to the preference matrix by establishing a new, trait-specific association between conscientiousness and the collection mechanism.

In this study, the competition mechanism is defined as enabling users to engage in competitive interaction with others in order to encourage the target behavior [23].

According to the preference matrix M1, competition is associated with all 5 Big Five personality traits. However, our results reveal a more differentiated pattern. Individuals scoring high in conscientiousness (OR 3.22, 95% CI 1.73-6.00) and agreeableness (OR 1.93, 95% CI 1.08-3.45) were significantly more likely to select competition as a motivating mechanism for maintaining their health compared with participants scoring lower on these traits. In contrast, participants with higher levels of extraversion (OR 0.43, 95% CI 0.25-0.74) and neuroticism (OR 0.52, 95% CI 0.30-0.89) were significantly less likely to select this mechanism.

Consistent with these findings, the ordinal regression analysis showed that participants with higher conscientiousness scores rated the competition mechanism as more motivating (OR 1.63, 95% CI 0.08-0.89), whereas those with higher neuroticism scores assigned lower motivation ratings (OR 0.71, 95% CI 0.70-0.01).

These divergent results suggest a controversial or polarized preference for competitive mechanisms. Qualitative feedback indicated that some participants favored individual-based mechanisms that emphasize personal effort rather than social comparison or rivalry. This interpretation aligns with previous research, which reports similar reservations toward competitive features in health-related apps, particularly when users prefer self-directed rather than socially driven motivation [10].

While other participants said they found competition more motivating:

The preference matrix (M1) further indicates that individuals scoring high in agreeableness and conscientiousness tend to prefer competition [8,10,32]. However, it also reports a preference relation with extraversion [9,10,24] and neuroticism [33], whereas our results show that individuals high in these 2 traits selected the competition mechanism less frequently. This discrepancy is noteworthy, as the preference matrix (M1) is based on evidence from three studies reporting a positive association between extraversion and competition, which contrasts with the nonpreference pattern observed in our data. In addition, several studies examining competitiveness more broadly have documented a general attraction to competitive contexts among extraverted individuals, further emphasizing the unexpected nature of our findings [34,35]. One possible explanation lies in the heterogeneity of the extraversion construct itself: preferences for competitive environments appear to vary depending on specific subfacets of extraversion [36]. Individuals scoring higher on friendliness (“make friends easily” and “often feel uncomfortable around others”) or cheerfulness (“have a lot of fun” and “am not easily amused”) tend to show reduced interest in competitive dynamics. In contrast, those characterized by higher assertiveness (“take control of things” and “wait for others to lead the way”) or excitement-seeking (“love action” and “dislike loud music”) display stronger preferences for competition and achievement-oriented contexts [36]. These nuances suggest that the link between extraversion and preference for competitive mechanisms is not uniform and may be facet-dependent, an aspect that warrants deeper investigation in future research.

By contrast, the preference matrix (M1) also reports a nonpreference relation for individuals with higher levels of neuroticism profiles [32], a pattern that is corroborated by the results of our experiment.

The reward mechanism involves the provision of virtual incentives to users for completing a target behavior [23]. It differs from the collection mechanism in that rewards are not necessarily accumulated as part of a predefined or structured set of items.

In our study, individuals with higher levels of neuroticism (OR 1.97, 95% CI 1.36-2.86) and conscientiousness (OR 2.36, 95% CI 1.53-3.63) were significantly more likely to select the reward mechanism. In addition, these participants assigned higher motivation scores to this mechanism (OR_{Conscientiousness}1.66, 95% CI 0.16-0.85; OR_Neuroticism1.38, 95% CI 0.03-0.61).

These findings are consistent with the preference matrix (M1), which reports established preference relations between the reward mechanisms and both conscientiousness and neuroticism across multiple studies [24,30,32,33,37,38].

Notably, conscientiousness emerged as a significant predictor across all 3 mechanisms identified in this study. In contrast, no significant associations were observed for openness, despite prior literature suggesting preferences for certain mechanisms, such as collection and competition, among individuals high in this trait.

Similarly, no significant preference results were identified for extraversion, even though the preference matrix derived from the literature (M1) reports associations between extraversion and multiple mechanisms. A comprehensive summary of the experimental findings alongside the preference matrix is presented in Table 10.

Implementing the preference matrix involves 2 main steps: identifying the user profile and tailoring the app content accordingly. User profiling traditionally relies on validated personality questionnaires (eg, BFI-10 and NEO-PI-R); however, these instruments can be time-consuming and may discourage user participation. As an alternative, automated profiling approaches based on smartphone usage patterns [39,40] or social media data [41-43] have shown promising results, although privacy concerns remain a major limitation. Such approaches require access to personal data, sometimes including sensitive inputs such as audio or video, which many users are reluctant to share [44]. Moreover, extensive data collection may increase perceptions of intrusiveness or manipulation, potentially undermining user trust and the perceived validity of behavior-based personalization [44]. To address these concerns, we recommend making personalization optional, offering users the choice to complete a brief, validated personality questionnaire if they opt in, and implementing strict ethical safeguards to ensure responsible data use. In particular, personality-related data should be collected solely for the purpose of enabling individualized personalization within the app, should not be repurposed for any secondary uses or shared with third parties, and should be handled with full transparency. Users must be clearly informed about what information is being collected, how it will be processed, and for what specific purpose (to tailor the app’s features to their personality profile), thereby promoting informed consent, user autonomy, and trust.

Once the profile is established, app content can be adapted by selecting mechanisms aligned with the user’s dominant personality traits, as defined by the preference matrix. For example, a user scoring high in conscientiousness might be shown mechanisms emphasizing rewards, competition, and collection, while competitive features could be intentionally avoided for a more extraverted user. This automated adaptation enhances user engagement by aligning motivational strategies with intrinsic preferences derived from psychological profiling.

However, to ensure that personalization remains user-centered and flexible, the system should also allow users to manually adjust or refine these automatically suggested mechanisms. After reviewing the proposed configuration, users can confirm, modify, or reject specific mechanisms based on their subjective preferences or contextual needs. This hybrid approach, combining data-driven profiling with user autonomy, maximizes both the accuracy and acceptability of personalization. It also mitigates potential mismatches between predicted and actual motivation, ensuring that the final app configuration genuinely reflects the user’s individual preferences.

The requirement for participants to select exactly 5 mechanisms out of 15 constitutes a methodological limitation. Some participants may have selected mechanisms they did not find particularly motivating to reach the required number, while others may have wished to select more than 5 mechanisms but were constrained by the study design. This forced-choice approach may, therefore, have influenced the distribution of selected mechanisms and attenuated individual preference differences.

In addition, our study sample was not fully representative of the general population, as it included a high proportion of university students, who are typically younger (mean_age 29.42, SD_age 10.41). Such demographic homogeneity may restrict the extent to which these findings can be generalized to other age groups or populations. Nevertheless, prior research on digital behavior changes and web-based health interventions suggests that age does not substantially moderate the effects of these interventions on health-related behaviors [45,46]. Systematic reviews have similarly reported that demographic variables such as age, sex, and education tend to have weak or inconsistent moderating influences on the effectiveness of these techniques [45]. Moreover, individuals most inclined to engage with health apps are generally younger and possess higher levels of education and income [47], which aligns closely with the characteristics of our sample and supports the external validity of our findings. Future research should, however, aim to include older adults and individuals less familiar with digital technologies to examine whether similar patterns of preference and engagement are observed in these groups. This population is particularly relevant to investigate, as older adults also represent an active and growing user base of mHealth apps, often using them for health monitoring, disease prevention, or physical activity tracking. Recent evidence from a comprehensive scoping review underscores the growing importance of this demographic and highlights that addressing key barriers, such as evolving capability requirements, cost, privacy concerns, and age-related design assumptions, can substantially enhance meaningful engagement. It also emphasized the importance of empowering older adults through learner-centered, need-based digital education [48]. Building on these insights, future studies should explore how mHealth apps can be designed to support sustained digital engagement among older adults and to better capture the diversity of their experiences and values in technology use.

Another limitation concerns the personality assessment instrument used in this study. We used the BFI-10-Fr, an ultra-short version of the Big Five Inventory, to minimize participant burden given the overall length of the online questionnaire. Although the BFI-10-Fr provides a practical and time-efficient measure, previous research has noted that its internal reliability can be relatively low due to the limited number of items per trait [49]. Nevertheless, according to the original validation study by Rammstedt and John [49], the BFI-10 demonstrates satisfactory test–retest stability and acceptable construct validity, supporting its suitability for large-scale or time-constrained surveys. Future studies should, however, consider replicating these analyses using a more comprehensive personality inventory, such as the Big Five Inventory-44 or NEO-PI-R, to ensure higher internal consistency and further validate the robustness of the associations observed.

We also observe that, despite our sample size calculation, it appears that it did not allow for proper discrimination, as we obtained significant results for only 3 out of 14 mechanisms. Therefore, this calculation should be reconsidered.

It was not feasible to assess the customization mechanism, given its expansive scope and the consequent challenges in representing it in the format of a conventional mockup screen.

Although the regression analyses yielded statistically significant results for several mechanisms, the explanatory power of the models remained relatively modest. For instance, the proportion of variance explained ranged from 6% to 10% for the collection mechanism, and from 5% to 9% for the ordinal regressions predicting motivation scores. Even for the most predictive models, such as competition (15%‐33%) and Rewards (16%‐27%), a substantial portion of the variance remained unexplained. This indicates that while personality traits, particularly conscientiousness and neuroticism, appear to play a meaningful role in the selection of gamification mechanisms, other unmeasured variables likely contribute to these motivational choices. Therefore, the practical significance of these findings should be interpreted with caution: the Big Five dimensions alone may not fully account for users’ preferences, which are likely shaped by situational, contextual, or motivational factors beyond personality. Future studies could aim to include such variables, or use alternative modeling approaches (eg, interaction terms or structural models) to capture more of the underlying variance in mechanism selection.

The entire preference matrix could not be validated with experimental data. Significant relations were identified for three mechanisms only: collection, competition, and rewards. Most of these preference relations are consistent with those reported in the literature-based preference matrix (M1), apart from the preference relation for the collection mechanism and conscientiousness and the nonpreference relation for competition regarding the extraversion trait, which is significant in this study but, according to the preference matrix M1, is a preference relation. The neuroticism trait is significant for competition as nonpreferred, yet the preference matrix studies demonstrated both a preference and a nonpreference relation for this trait. Considering these findings, it can be concluded that a nonpreference relation is corroborated. It would be interesting to replicate this study with a larger, less student-dominated panel to achieve better representation and more significant results on mechanism preference relations by Big Five traits.