Digital health interventions are complex interventions, operating within complex systems [46]. Paired has several dimensions of complexity; for example, users “receive” the intervention differently because it has multiple components that can be engaged with in different ways and because users engage with it for as long and as frequently as they choose [47,48] (vs completing a course). It may lead to improvement in relationship quality through complex causal pathways, and its use and effectiveness may be shaped by the social and relationship contexts and settings in which it is used [49]. It is also complex because it requires unaided use of technology [50] (ie, users’ devices and the app), relying on digital literacy skills that may differ between users.

To evaluate Paired, we drew on established guidance for evaluating complex interventions [50]. Evaluations can address a range of related questions, such as whether, in what contexts, and for whom an intervention works [51]; how it works; and how it may be further developed. Mixed methods and interdisciplinary approaches are recommended for evaluating complex interventions [52], specifically digital interventions [53-55]. The integration of qualitative and quantitative approaches can aid an understanding of user behavior and issues affecting intervention success [56] and strengthens the conclusions that can be drawn [57]. Qualitative research aids the development of causal explanations by describing the processes that produce an outcome [58].

Relationship quality, our main outcome, is measured in various ways [59]. We believe it is multidimensional [60] (ie, it entails factors such as communication quality and how couples deal with conflict) and is best assessed as such (information received from Di Martino et al [email, January 16, 2025]). In this study, we use a broad definition of relationship quality: “how positive or negative individuals feel about their relationship” [61], which acknowledges that it is subjective yet measurable at the individual level.

We developed a provisional theory of change to guide our evaluation, representing how we expected that Paired might “work” based on findings from relationships research and behavioral science. This included the following 3 strands:

We sought to develop this provisional theory through our research. We consider the role of Paired within relationships from a digital sociology perspective in the Discussion section.

This study aimed to explore the potential for an mHealth intervention to benefit relationships and how it might do so, using Paired as an exemplar. The objectives were (1) to assess the app’s impact on relationship quality by examining associations between relationship quality and duration, frequency and intensity of use, and by comparing the perceived impact of Paired among its users (as indicative of the direction of causation); and (2) to develop a refined, empirically informed understanding of how Paired may deliver improved relationship quality. To put it simply: (1) “Does it appear to work?” (2) “How does it work?”

At 3 time points with 1-month intervals, starting October 30, 2020, the Paired weekly quiz was replaced by a 5-item in-app survey, identified to users as university-led research. This brief survey asked for agreement or disagreement (on a 5-point Likert scale) with 4 statements about aspects of relationship quality (ie, communication, emotional connection, conflict, and sex and intimacy). One further statement concerned participants’ perception of the impact of Paired on their relationship communication. Researchers obtained a deidentified extract of data from Paired users who had completed at least 1 brief survey. This included data from a “relationship check-up” that users can complete when they first download the app, containing 4 statements similar or identical to those in the brief survey, which we used as baseline measures (the data extract also included demographic details for a small minority of users, which were not used). From 3717 unique participants, the sample was restricted to 440 (11.84%) individuals who completed the relationship check-up in October 2020 and all 3 brief surveys thereafter (4 data points), with at least 7 days between each data point to allow for a realistic prospect of change (in explanation, the data showed that some users delayed completion of the brief surveys; if they completed, say, November’s survey just before completing December’s survey, we would not expect to detect much change). This gave complete longitudinal data spanning approximately 3 months (October-December 2020), referred to as “brief survey data” for brevity.

A 30-item survey, administered in December 2020 and hosted securely outside the app, collected quantitative and limited free-text data. Paired users were invited to complete it via 3 in-app messages containing the survey link. Those willing to be contacted regarding a research interview provided email addresses.

The survey collected demographic data but not race or ethnicity, religiosity, income, or socioeconomic status (despite well-documented associations with marital satisfaction and quality in the United States [63-65]). These constructs are differently defined and delineated in different cultures, posing challenges for our relatively small-scale international study.

From the web-based survey data, we obtained self-reported duration and frequency of Paired use, as well as relationship quality, measured by (1) the Multidimensional Quality of Relationship Scale (MQoRS) and (2) direct questions about relationship quality change since using the app. The MQoRS, developed for this study, combines and weights responses to 17 statements on 5 aspects of relationship quality and is expressed on a scale ranging from 0 to 10. Suited to smartphone-based self-completion (ie, with relatively few survey items comprising simple statements and with responses using 5-point Likert scales that fit on the smartphone’s screen, thus requiring minimal scrolling), it overcomes the shortcomings of unidimensional relationship quality measures and is described in detail elsewhere (Multimedia Appendix 2).

Individual interviews were conducted on the web with purposively sampled web-based survey participants (January-April 2021). The interviewer (TW) identified himself as a university researcher, independent from Paired. The primary sampling criteria were gender and country, with initial targets of 5 women and 5 men each from the United States and the United Kingdom. The secondary sampling criteria, by which we sought diversity across the sample, were sexuality, relationship duration, cohabitation, and the presence or absence of children in the household. The interviews followed a topic guide and lasted 40 to 75 minutes. Similar descriptions of Paired use and its impacts were heard repeatedly toward the end of data collection, suggesting that saturation had been reached.

Quantitative data were analyzed using SPSS software (version 26.0; IBM Corp) for descriptive statistics and R software (R Foundation for Statistical Computing) for inferential statistical analyses and data visualization. Statistical significance was considered as P<.05.

Brief longitudinal survey data were analyzed using multilevel multinomial logistic regression models with cumulative logit link functions. Predicted values were used to generate odds ratios (ORs) and 95% CIs for change in each dimension of relationship quality over time compared to baseline (the initial “relationship check-up”).

Drawing on data from the web-based survey, MQoRS scores were compared between different categories of reported duration and frequency of using Paired, using 1-way ANOVAs with Welch approximation due to unbalanced sample sizes and post hoc Tukey tests. Responses to direct questions about change in relationship quality since using Paired were compared using chi-square tests to make binary comparisons between people reporting using Paired for ≥1 month versus <1 month and people reporting using Paired on 6 to 7 days per week versus ≤5 days per week (among those reporting <1 mo use). Of note, because we lacked the data to predict our sample and subsample sizes in advance, decisions on the cutoffs for these categories were not made a priori but were data driven and were also informed by our provisional theory of change.

Evidence from the 2 surveys on the association between relationship quality and the duration and frequency of Paired use was triangulated. The direction of causation of this association was explored qualitatively, informing the development of the theory of change (objective 2).

We compared the perceived effectiveness of Paired between demographic groups, using the proportions agreeing with statements in the web-based survey (detailed in the Results section and Multimedia Appendix 3). We used chi-square tests for association and univariable logistic regression to obtain ORs for demographic differences (using the characteristics we used for interview sampling and additionally age, reflecting widespread age-related assumptions regarding technology engagement). Analysis was restricted to the 514 participants who reported ≥1 month’s use of Paired (because it is perhaps less plausible that change in relationship quality among new users is attributable to Paired). Qualitative evidence to help explain quantitative findings was provided by interview responses to direct questions and spontaneous remarks about the app’s inclusivity.

TW and CA, both experienced qualitative researchers, conducted the qualitative analysis. We used reflexive thematic analysis [66] to describe how Paired seems to work, using a largely inductive process but also informed by the provisional theory of change. “Dosage” and its association with relationship quality were explored quantitatively (refer to the Objective 1: Impact of Paired on Relationship Quality subsection) and qualitatively.

Web-based survey results show that relationship quality, measured by the MQoRS, was 35.5% higher among those with >3 months’ Paired use compared to new users (5.19 among people who reported having used the app for ≤1 wk vs 7.03 among those who reported >3 mo use; 95% CI 31.1%-43.7%; P=.002; Figure 1). Supporting this, 59.5% (440/740) agreed or strongly agreed that their relationship felt stronger since using Paired (95% CI 56%-63%), and people who had been using the app for ≥1 month were more likely to report this than newer users (330/513, 64.3%, 95% CI 60.2%-68.4% vs 110/227, 48.5%, 95% CI 42%-55%; P<.001).

The longitudinal brief survey data support this dose-response finding, showing positive changes over time in a cohort of Paired users in 4 distinct aspects of relationship quality (Figures 2-5; for each data point in Figures 2-5 [x-axis], the percentages sum to 100%; an increase in the extent of agreement is particularly marked in Figures 2-4, where there is an increase in the percentages responding “strongly agree”; the decreasing percentages responding “agree” represent a shift toward this stronger level of agreement [and not a decline in overall agreement]). The web-based survey analyses also support this (Multimedia Appendix 3).

Reporting more frequent Paired use was associated with higher MQoRS scores: people using Paired on 6 to 7 days per week had 11.8% higher relationship quality than those using Paired on ≤1 day per week (MQoRS scores: 6.81 vs 6.09), and this difference was statistically significant (95% CI 0.19-1.42; P=.04; Figure 6; details in Multimedia Appendix 2). Supporting this finding, among people who had used Paired for at least 1 month, those who reported using it on 6 to 7 days in a typical week were more likely to agree or strongly agree that their relationship felt stronger since using Paired compared to those who used it less often (223/316, 70.6%, 95% CI 65.5%-75.6% vs 106/196, 54.1%, 95% CI 47.1%-61.1%; P<.001). However, we found no statistically significant association between the total amount of time spent using Paired in a typical week (intensity of use) and MQoRS scores.

Among those who reported having used Paired for ≥1 month, there was no statistically significant difference in the proportion of web-based survey participants who agreed or strongly agreed that “Paired is improving how we communicate as a couple” (overall 414/514, 80.5%), “Our relationship feels stronger since we’ve been using Paired” (overall 330/513, 64.3%), or “The longer I use Paired, the better my relationship gets” (overall 252/514, 49%) between men and women or by age group, sexuality, country, relationship status, or relationship duration (refer to Multimedia Appendix 3 for details of the demographic categories used). Participants without children (aged <18 y) in their household were more likely to agree that their relationship felt stronger since using Paired compared to those living with children (OR 1.49, 95% CI 1.00-2.21), but this was of borderline statistical significance (P=.048), and no differences were observed in agreement with the other statements.

Regarding the qualitative data, the inclusivity of Paired was described positively; for example, it was described as suitable for a range of ages (“kind of age agnostic” [UK02, woman, aged 58 y, heterosexual, relationship duration: ≤1 y, living apart, no children]) and as potentially benefiting couples in different ways over the course of their relationship (“as a prompt of finding out more about your partner early on, or [for] mature couples, the stuff you take for granted” [UK07, man, aged 36 y, heterosexual, relationship duration: >5 y, cohabiting, children]). While some commented that some of the app’s content did not apply to their relationship (eg, sharing domestic chores among noncohabiting couples), this was not necessarily perceived negatively. Interviewees valued questions and content that were broadly relevant to diverse couples. As an interviewee in a same-sex relationship explained, he had never come across a question which was not inclusive:

Another interviewee considered Paired “just so useful for anyone” but wondered whether some people would be reticent to start using it:

A total of 4 themes were identified from the qualitative analysis, 3 (75%) of which drew upon the provisional theory of change (refer to the Approach and Theoretical Basis subsection in the Introduction section). Subthemes and the theme Engagement (which describes how Paired use is sustained) were derived from the data. Theme and subtheme descriptions (subheadings) in this subsection summarize the refined theory of change.

The theme Communication described how Paired can cumulatively improve relationship quality through prompting meaningful conversations and providing topics, enabling couples to regularly practice communication and enhance their emotional connection. In the themes Engagement and Dailiness, we described how the use of Paired can be sustained and become embedded in couples’ daily lives. Supporting the quantitative findings, interviewees noticed gradual, incremental gains with regular, frequent use of the app:

The survey findings showing that the amount of time (per week) spent on the app was relatively unimportant may be explained by the qualitative findings that Paired facilitates relationship maintenance behaviors that may occur offline:

The findings provide proof of concept of mHealth for relationship care for supporting modest improvements in relationship quality, over a relatively short time frame, in a self-selected population of users of 1 app. In the context of limited research on fully digital couple interventions for relationship care [45] (ie, positive, preventive interventions that are not blended or therapist guided), our rigorous evaluation makes an important contribution to the evidence base. The findings fit with existing evidence and theory about the importance of regular daily “relationship work” in sustaining couple relationships [43] and demonstrate how this can be supported in practice. The interview accounts of how benefits can accrue from regular, positive interactions and moments of connection evoke the concept of “relationship banking” [70] and demonstrate how mHealth can prompt these interactions. We would not expect large changes in relationship quality, given the many contextual influences on relationships, the time frame, and the fact that the study population included users whose relationship quality was already good (among whom scope for improvement is limited).

Similar perceptions of the app’s effectiveness among people of different ages, sexualities, and relationship types and durations suggest that it is inclusive by these characteristics. Therefore, its appeal is likely to be broad, although ethnic and cultural inclusivity has not been explored. Evidence of the effectiveness of Paired is promising in the context of underuse of existing preventive interventions. The convenience and flexibility of mHealth suits the ebbs and flows of couple relationships because partners have control over when and how much they engage and apparently benefit without needing to use the app intensively. Barriers to uptake are low for digitally literate web-based populations, and digital interventions are scalable (eg, the work of creating new questions and content for Paired is the same, irrespective of the number of users), contrasting with face-to-face or “blended” courses or therapy, which require conjoint commitment and a professional’s input with each couple or group.

Where mHealth interventions incorporate tailoring to individual users’ characteristics, this can increase the personal relevance of content and messaging, which in turn can increase engagement with the intervention and improve learning [71]. Blended interventions may achieve similar effects through personalized messaging from a coach to each individual user. We have shown that an intervention without these features can be personally relevant because partners effectively create content for each other. They “receive” an intervention that has a unique human touch (their partner’s) that is not only personalized but intimately personal. Peer-to-peer digital communication ordinarily requires resource-intensive moderation, but this does not apply to Paired because communication is between partners only.

Theories of behavior change tend to be individualistic, although some acknowledge the social dimensions of learning [31,72]. The refined theory of change that we developed is novel in encompassing the dyadic nature of behavior change within relationships, and in emphasizing how human-technology interactions and between-partner interactions support this change and sustain engagement. Paired engagement is sustained directly (eg, in-app messages) and indirectly (partners remind each other to answer questions; they experience relationship improvements and are motivated to continue to engage).

A combination of fun and meaningfulness seemed to help maintain engagement, and the app prompted regular daily “relationship work.” These findings may inform self-help and professionally delivered courses and therapy, both digital and face-to-face, including future iterations of Paired.

The use of apps to score and rate users’ sexual lives has been described by Lupton [73] as reinforcing reductive and normative perspectives on what is “good” within an area of intimate life. Levels of agreement in Paired quizzes are undoubtedly reductive assessments of how “good” a relationship is, but in contrast to the apps reviewed by Lupton [73], these comparisons are not designed for ranking or sharing outside of the couple. Instead, they stimulate within-couple discussion and interpretation. Daily questions elicit free-text responses, eluding quantification, but “streaks” are comparable between partners. The work conducted by Lupton [55,74] explains how mHealth apps become more-than-human individual-app assemblages, with their own agency. We suggest, based on our interviews, that Paired creates an individual-partner-app assemblage, functioning within and as part of couples’ intimate lives [75]; for example, daily question completion indicates to one partner that the other is thinking of the relationship, both documenting and becoming part of couples’ relationship work; differences in streak length may signify lower commitment to the relationship. App-based indicators of conjoint accountability may cause difficulties for couples where problematic relationship dynamics already exist [42]. In a study of young people’s communication regarding their Snapchat streaks, Hristova et al [76] found that not maintaining a streak could have significance for peer relationships similar to what we found for couples (unlike Paired streaks, which are individualized, a Snapchat streak is a continuous period of at least daily “snaps” exchanged between a pair of users).

We evaluated 1 app, within the first 3 months of its availability, as an exemplar to explore whether mHealth can benefit relationships. Choosing Paired was fortuitous, given the rapid turnover of apps, most of which fail (an estimated 99.5% [77]). Paired became the leading relationships app worldwide in terms of revenue and downloads in 2021 and 2022 [78]. It expanded from 10,000 active users during our study to >1 million monthly active users in November 2024 (information received from Paired [personal communication, December 10, 2024]). Therefore, the findings inform an app with considerable reach. Paired became more sophisticated since data collection, but our findings remain relevant because core features remain, including the daily questions that were so important. The app’s impact may be affected by how new technologies’ user populations tend to be differently constituted over time [79]. Our findings should not be generalized uncritically to apps with different functionality, features, and content. We make no claim that our findings are generalizable to all couples or all users of Paired. We have not explored what happens to relationship quality beyond 3 months’ use of Paired or whether changes are sustained if couples stop using it.

This research took place during the COVID-19 pandemic, and couples may have turned to Paired during relationship strain; however, the data are not suggestive of severe relationship problems among participants. Media predictions of large increases in relationship breakdown and divorce during the pandemic did not occur [15,80]. Instead, impacts on relationships were heterogeneous—and positive for some couples [17,81]—although long-term impacts remain unknown [82,83]. However, couples’ access to formal and informal relationship support was reduced or interrupted. Some people invested in personal growth and relationships during lockdowns and quarantines [84], often supported by digital resources. Couples may use Paired differently now, perhaps preferring to engage with relationship care through other means or not at all. Nevertheless, if the digital engagement with services, social lives, and entertainment that accelerated during COVID-19 is here to stay, interest in digital relationships care may persist—as Paired subscription patterns suggest.