The Psychology of Personalization in Digital Environments

The impact of digital media on individuals and societies is one of the key issues of our times. Some have argued that the rise of digital media has reduced individuals’ well-being, leading to an increase in depressive symptoms and suicides (e.g., Twenge, 2019^; Twenge et al., 2018), whereas others have questioned the evidence for this causal link (e.g., Appel et al., 2020^; Granic et al., 2020^; Orben & Przybylski, 2019). Against this background, and with digital platforms increasingly relying on artificial intelligence (AI) to boost user engagement and company revenues, it has become pivotal to understand the antecedents and processes underlying beneficial or maladaptive digital media use. Our focus here is on the personalization of experiences in the digital realm and its influence on the individual. We present a model based on Self-Determination Theory (SDT) that explains and predicts need satisfaction and well-being in personalized digital environments. This model is applied to reconstruct extant scientific evidence on personalization experiences in three fields: news feeds and websites, music streaming, and online dating. We discuss how our model could be used to improve human well-being in AI-based personalized environments.

The Personalization of Digital Environments

Digital environments can broadly be described as spaces and contexts created by electronic and digital technologies that we use for communicative, industrial, commercial, and recreational purposes (Buongiorno, 2021). The personalization of these environments is becoming ubiquitous, thanks to the rise of AI-based algorithms and recommender systems (e.g., Burr & Floridi, 2020^; Sundar, 2020). By analyzing user data, news feeds and websites filter their content, search engines adjust which results to display, companies tailor ads to their customers, dating platforms suggest potential matches, music streaming services are able to recommend albums and artists, health service providers show different patients different information, and online learning environments adapt to the learners’ progress and interests—to name just a few examples. In contrast to user-initiated customization, personalization refers to system-initiated processes (Sundar & Marathe, 2010; see also Fan & Poole, 2006^; Arora et al., 2008). That is, personalization is not about the user making deliberate decisions regarding the kind of content that is displayed; it is about all those “automatic changes [that are made to digital environments] to accommodate individual user’s need, interests, knowledge, goals, or tasks” (Serino et al., 2005, p. 1).

This kind of automatic, system-initiated, and algorithm-driven adaptation of digital environments is something fundamentally new: While people have always decided to focus on some aspects of their environment while neglecting others, the environment has remained essentially the same for everyone. However, personalization changes the environment itself: The virtual world is different for each individual—and it constantly changes in different ways for different individuals. For instance, the reader of a printed newspaper may skip certain articles, but the skipped articles remain a visible part of the environment. On a personalized news website, an algorithm can potentially filter out the articles the reader would likely skip before the reader even gets to see them. Arguably, this technological development has far-reaching consequences not only for human–computer interactions, but also for psychological well-being and ultimately for society in general. But is personalization a good or a bad thing?

This question can be answered from a wide range of different perspectives. For instance, the fact that personalization relies on collecting an abundance of personal data has raised ethical and legal concerns regarding privacy, trust, and discrimination (e.g., Bozdag, 2013^; Burr et al., 2018^; Fast & Jago, 2020^; Matz et al., 2020^; Milano et al., 2020^; Mittelstadt et al., 2016^; Taddeo & Floridi, 2018). Here, we are mainly concerned with understanding the impact of personalization on individuals. In this context, some argue that personalization enhances human agency by supporting better decision-making and reducing information overload, and that it may ultimately increase the individual’s well-being (e.g., by fostering healthier lifestyles and lifelong learning; Spanakis et al., 2014). Others contend that personalization excludes human supervision, potentially enabling coercion, manipulation, and the formation of echo chambers as well as undermining autonomy and control (e.g., Mejtoft et al., 2020^; Newell & Marabelli, 2015). Against this background, it seems reasonable to hypothesize that users differ regarding the perceived benefits and risks of personalization (see, e.g., Amarnath & Jaidev, 2020^; Lee & Lee, 2009^; Liang et al., 2006^; Salonen & Karjaluoto, 2016^; Treiblmaier & Pollach, 2007) and that personalization is likely to have complex influences on individuals.

In this paper, we propose a psychological model of the effects of personalization, one that considers individual motivational differences and that can explain both the positive and the negative effects of personalization. By integrating a wide range of findings from both inside and outside psychology, the model also provides a useful starting point for organizing future research. Our model is based on Self-Determination Theory (SDT; Deci & Ryan, 2000), an established model of psychological needs. This has a crucial advantage of allowing us to identify the psychological impact of personalization in digital environments, a novel phenomenon, by building on a strong foundation of knowledge regarding basic psychological needs. Focusing on the psychological processes addressed by SDT also enables us to base our analysis on mechanisms and individual differences that are deeply rooted within the human psyche. In other words, we decided to analyze personalization within a well-established framework designed to capture a wide range of human behavior, rather than outlining a model that can be used exclusively to explain the effects of personalization: This allows us to describe the impact of personalization using known terms and concepts but also to state explicitly what distinguishes personalization from other mechanisms shaping our environment. In this context, note that SDT has already been used to tackle the topic of well-being in human–computer interactions in general (e.g., the METUX model proposed by Peters et al., 2018). Although such broad conceptualizations provide a valuable starting point for further investigations, our model, which focuses specifically on personalization processes, allows for a more nuanced and detailed analysis of a phenomenon that drives many of the contemporary changes in the digital realm.

The Model: Motivation, Basic Psychological Needs, and Well-Being

Self-Determination Theory is a macro theory of human motivation, which connects motivational tendencies with innate psychological needs and well-being (for an overview, see, e.g., Deci & Ryan, 2000^; 2008^; Ryan & Deci, 2017). As far as motivation is concerned, SDT assumes that “the type or quality of a person’s motivation [is] more important than the total amount of motivation” (Deci & Ryan, 2008, p. 182) and proposes an autonomy-control continuum. If a behavior is autonomously motivated, the person experiences volition and identifies with the behavior’s value. The paradigmatic example for autonomous motivation is intrinsically motivated behavior, which is pursued without the necessity of specific consequences, that is, behavior that is interesting and pleasurable in and of itself. In contrast, behavior is controlled when it is regulated by external contingencies (i.e., when behavior is aimed at gaining a reward or avoiding a punishment) or when it “is energized by factors such as an approval motive, avoidance of shame, contingent self-esteem, and ego-involvements” (Deci & Ryan, 2008, p. 182). Put differently, controlled motivation comprises those forms of extrinsically motivated behavior that aim at obtaining some instrumentally separable consequence. When extrinsically motivated behavior is perceived as personally important, however, it will be experienced as relatively autonomous. Finally, a complete lack of self-determination and intention is called amotivation. When being amotivated, people feel unable to attain desired outcomes and experience a lack of interest and relevance, sometimes accompanied by feelings of helplessness (see, e.g., Pelletier et al., 1999).

While the degree to which behavior is autonomous or controlled may vary from context to context (i.e., is state-dependent), SDT also proposes that there are relatively stable individual differences in the form of causality orientations (see Deci & Ryan, 1985). The three causality orientations—the autonomy orientation, the controlled orientation, and the impersonal orientation—“parallel at a more global level the concepts of autonomous motivation, controlled motivation, and amotivation” (Ryan & Deci, 2017, p. 20). While an autonomy orientation mirrors the general tendency to take one’s own values and interests as a guideline when organizing behavior, a controlled orientation is the general tendency to adjust one’s behavior based on expected reward contingencies and social controls. In addition, if one generally experiences feelings of incompetence and a lack of control over one’s environment, one has an impersonal causality orientation.

Crucially, motivation is linked to the three innate psychological needs for autonomy, competence, and relatedness. Autonomy denotes the need “to self-organize and regulate one’s own behavior,” while competence refers to the need “to engage optimal challenges and experience mastery or effectance in the physical and social worlds,” and relatedness is the need “to seek attachments and experience feelings of security, belongingness, and intimacy with others” (Deci & Ryan, 2000, p. 252). As far as the link between motivation and need satisfaction is concerned, autonomously motivated behavior is more likely to result in a satisfaction of all three basic needs than controlled behavior. Moreover, the satisfaction of the three psychological needs leads to well-being. Note in this context that the term well-being can refer to two different concepts: hedonic and eudaimonic well-being (e.g., Ryan & Deci, 2001). Hedonic well-being is achieved when pleasure is maximized and pain is minimized, while eudaimonic well-being denotes the full functioning of the individual. SDT is mainly concerned with eudaimonic well-being and states that people experience eudaimonic well-being when they achieve their potential and have the opportunity for personal growth as indicated by a satisfaction of their basic needs. Reversely, need thwarting is associated with less than optimal development and functioning, potentially causing psychological and physical health problems (see, e.g., Ryan et al., 2006). Note also that need frustration and reduced well-being predict motivational depletion (Ryan & Deci, 2008), that is, need frustration and reduced well-being make it less likely that a person shows autonomously motivated behavior, potentially creating a vicious circle.

According to our model, the main components of SDT can be taken as a basis for analyzing the impact of personalization in digital environments from the perspective of the individual (see Figure 1). The effects of personalization on a user with a controlled motivation will crucially depend on whether the personalized recommendations help the user to attain the desired goal (e.g., finding the information on a website that one is looking for). However, when considering individuals with autonomous motivation, that is, individuals whose interest is not primarily in achieving an instrumental goal but who identify with their behavior’s value and for whom the behavior is interesting and pleasurable in and of itself, matters become more complex. Following SDT, we assume that the impact of personalization on autonomously motivated individuals mainly depends on how it affects the current experience. Hence, autonomously motivated individuals may perceive personalization as helpful insofar as it reduces information overload and filters out irrelevant options. However, they may also perceive personalization as intrusive and limiting insofar as it disrupts engagement with the activity.

Open multimedia modal

Especially the latter considerations regarding autonomously motivated individuals seem to suggest that personalization in digital environments can ultimately have both positive and negative influences on the satisfaction of our basic needs for autonomy, competence, and relatedness. For instance, being freed from irrelevant options may help foster autonomy and competence as it enables users to focus on those aspects that they find interesting and engaging when navigating through a digital environment. At the same time, not knowing which options are being filtered out and for what reasons, may undermine this effect, potentially leading to need frustration and reactance (Brehm, 1966^; Brehm & Brehm, 1981). With respect to relatedness, personalization could help to create common environments for individuals with common interests. At the same time, being restricted to one’s personalized environment may also disrupt social interactions, as there are fewer shared experiences with other individuals who do not have the same interests. Ultimately, this could also have effects on well-being. As one may argue, a well-designed personalization algorithm, which is able to capture the individual’s interests and goals and to recommend content that fits these interests and goals, will help to achieve hedonic well-being. Achieving eudaimonic well-being, however, requires more than good recommendations that work as gratifications: It requires respecting the needs for autonomy, competence, and relatedness. These considerations add a normative dimension to our model: If personalization in digital environments is neither a good nor a bad thing per se, but has distinct effects not only on user experience, but also on the individual’s well-being, it seems desirable to design personalized environments and recommender systems in a way that supports the satisfaction of our basic psychological needs (see, e.g., Calvo et al., 2020^; Peters et al., 2018). Before discussing these implications, we will review findings from three areas of application—news feeds and websites, music streaming, and online dating—based on the model that we have just described.

Personalization of News Feeds and Websites

Most information that we have about the world is not based on direct, personal experience, but rather on mediated information that we have gathered from different sources. News media have a particularly important role, as they provide information about the outside world with the truthful depiction of events and background facts as the guiding principle (APME, 2019). Therefore, news media are a constituent element of modern democracies: If and only if citizens have access to diverse information, they have a fair chance of building informed opinions. For instance, that may also explain why the right to freedom of expression as formulated by the European Convention on Human Rights does not only include the freedom to hold opinions, but also the right to receive information “without interference by public authorities and regardless of frontiers” (Article 10, paragraph 1; see also Eskens et al., 2017).

Against this background, it seems understandable why the personalization of news feeds and websites, which has become increasingly common in recent years (see, e.g., Karimi et al., 2018^; Thurman & Schifferes, 2012), is met with serious reservations. In particular, it has been argued that news personalization leads to the formation of filter bubbles (Dylko et al., 2017^; Flaxman et al., 2016^; Pariser, 2011) and echo chambers (Quattrociocchi et al., 2016^; Sunstein, 2009), which may ultimately result in the gradual dissolution of the shared public sphere (Helberger, 2019). In other words, the main concern is that people tend to prefer information that is in line with their attitudes, their political convictions, and their worldview in general—and that this tendency is reinforced by personalization algorithms, which typically recommend content that is similar to content users have previously engaged with. If this were the case, users would be exposed to a highly biased selection of information—maybe even without noticing it. Indeed, it has been noted that many users are unaware of the fact that news feeds and websites are being personalized (Eslami et al., 2015^; Fletcher & Nielsen, 2019^; Powers, 2017^; Swart, 2021).

Whether or not personalization algorithms nurture filter bubbles and echo chambers is the matter of an ongoing debate. News platforms argue that, in addition to personalizing content, they add more diverse content to prevent boredom and to foster the impression of a proper journalistic product (Bodó, 2019). Likewise, user data suggest that the effects of news personalization are complex. Recent data from the UK (Fletcher et al., 2021) show that news platforms that use personalization algorithms (e.g., Facebook, Twitter, or Google News) provide exposure to more diverse sources, as compared to non-personalized, self-selected news exposure, which, in the UK, often starts and ends with information provided by the BBC. These diverse sources, however, involve more partisan sources (i.e., leaning to the left or to the right of the political spectrum, such as breitbart.com or thecanary.co; see also Hasell, 2020^; Muller & Schulz, 2021).

Typically, the above-mentioned concerns regarding echo chambers and filter bubbles are described from an “outside” perspective that views users as passive recipients of news. However, as previous research has demonstrated, users hold nuanced views on potential advantages and disadvantages of news personalization when asked about it explicitly (e.g., Bodó et al., 2019^; Fletcher & Nielsen, 2019^; Joris et al., 2021^; Monzer et al., 2020): While they acknowledge that news personalization can help reduce information overload and leads to receiving more relevant news, they express concerns about being manipulated and caught up in filter bubbles. As they do often not know, which content is being filtered out and for what reasons, users also tend to experience a lack of agency and control when confronted with the fact that their news feed is personalized. This seems particularly important as users are very sensitive to threats to their autonomy in the context of news (Groot Kormelink & Costera Meijer, 2014; see also Kozyreva et al., 2021). That is, users do not necessarily read all articles or watch all videos, but they still want to be able to access all content whenever they want. Note, however, that there seem to be important individual differences when it comes to evaluating the desirability of news personalization. It has been observed that individuals with a higher interest in news are more skeptical about news personalization (Joris et al., 2021). More specifically, this seems to be true for individuals who have a high interest in “hard” news topic and a rather low interest in “soft” news topics (Fletcher & Nielsen, 2019; see also Kozyreva et al., 2021). In contrast, people with lower education and people who experience less political efficacy place a higher value on news personalization (Bodó et al., 2019).

The findings described so far can be understood through the lens of the proposed model. As it seems, users generally tend to be autonomously motivated in the context of news: Being well-informed about current events and political affairs is considered a value in itself or at least an important precondition for becoming a responsible citizen by many people. Therefore, individuals desire autonomy and control over the personalization process, while simply receiving information that is tailored to one’s interests is considered insufficient. However, inter- and intraindividual differences need to be taken into account. Some individuals seem to be regularly driven by controlled motivation: They do not place such a high value on being well-informed, but rather see navigating through news feeds and websites as a means to get the information they are looking for. Hence, they are more open to news personalization as long as it leads to the desired outcomes.

Based on the assumption that functioning democratic societies need open-minded news recipients who value diversity, one may argue that autonomous motivation is highly desirable in the context of news from a societal perspective. As news personalization could undermine diversity, there is reason to treat it with caution. This point of view seems to be supported from the perspective of individual well-being: Ultimately, news personalization may not only compromise autonomy, but also the individual’s need to feel knowledgeable and competent—not to forget that the anticipated fragmentation of the public sphere could potentially also endanger the experience of relatedness to other human beings. According to our model, such need thwarting would result in reduced well-being, which would in turn undermine autonomous motivation. However, it should be noted that these predictions regarding needs and well-being have not been addressed explicitly in previous research. That is, while the evidence presented above allows at least preliminary conclusions regarding the role of motivational differences in the context of news feeds and websites, the long-term effects of personalization algorithms as hypothesized by our model remain to be tested.

In case it should turn out that receiving unbiased information is not only important for political debate and social cohesion, but also for need satisfaction, well-being, and ultimately self-development (see also Eskens et al., 2017), it would be important to think about finding ways to give back control and autonomy to users in the context of news personalization. One idea might be to rely on customization instead of personalization, that is, to make adjustments in news feeds and on news websites dependent on the active choices that users make (Harambam et al., 2018^; Sullivan et al., 2019; for experimental evidence, see Zhu & Lee, 2020). Other ideas include increasing transparency about the kinds of data that are used to create personalized news feeds (Sullivan et al., 2019) and to raise awareness among users about the way personalization algorithms are operating (Harambam et al., 2018). In case participants feel unable to gain control over the personalization process, they may decide to alternate between different platforms or to obfuscate their digital footprint in order to reclaim autonomy (Harambam et al., 2018).

Music Streaming

Listening to music is not only one of people’s favorite leisure activities (Rentfrow & Gosling, 2003), it can have a profound impact on subjective well-being (for an overview, see Hallam et al., 2017): People use music to regulate their moods and emotions, and to reminisce about past experiences. Thus, music streaming services may appear as the “celestial jukebox” that Goldstein (1994) had in mind—as a tool that offers listeners effortless and immediate access to any music imaginable whenever they want and wherever they are. However, the currently available music streaming services do not simply offer access to any music imaginable, but adjust the displayed artists, albums, and songs to the user’s taste and preferences. This kind of personalized celestial jukebox is the result of a complex interplay between different algorithmic strategies (for an overview, see, e.g., Knees et al., 2019^; Schedl et al., 2015). Ultimately, the personalization of music streaming can be described as the “datafication of listening” (Prey, 2016) as “all listening time is data-generating time” (Prey, 2019, p. 11), that is, as all actions that a user takes within the music streaming service will be analyzed and used to adapt the displayed content.

Overall, users seem to enjoy the personalization of music streaming services (e.g., Hagen, 2015^; Kamehkhosh et al., 2020^; Liikkanen & Åman, 2016). However, there are also potential problems and sources of dissatisfaction. First, user needs and goals do not necessarily align with the financial interests of the companies, which offer the music streaming service (e.g., Morris & Powers, 2015^; Prey, 2018^; 2020^; Sun, 2019; see also Karakayali et al., 2018). While users may desire unlimited access to the “celestial jukebox,” for instance, the companies may profit from restricting certain features to paying customers (so-called “freemium model,” see, for example, Holm & Günzel-Jensen, 2017^; Voigt et al., 2017). Second, and maybe even more importantly in the present context, the music recommender systems do not always produce the results users are hoping for. In this context, it has been observed that some people stop using music streaming services because of a perceived lack of fit (Kjus, 2016) and that the behavior preferred by the users is not necessarily the behavior that is optimally supported by the recommender system (Lüders, 2019). Reversely put, the perceived match between the recommended content and the user’s motivations and expectations is an important predictor for overall satisfaction and continued use (see, e.g., Jones, 2010).

Interestingly, it has been observed that users engage in a wide range of different activities when interacting with a music streaming service. For instance, users rely on recommendations to play music in the background that fits their current activity (such as working or exercising) or to find new music for later, that is, to quickly collect tracks that they plan to play in other contexts such as a house party (Garcia-Gathright et al., 2018). In addition, the general possibility to access specific recordings whenever and wherever they want (i.e., ubiquity), is a key gratification for some users (Mäntymäki & Najmul Islam, 2015). Apart from that, it has been noted that users rely on recommendations to engage with new music (Garcia-Gathright et al., 2018) and that they may perceive the experience of interacting with the music streaming service as enjoyable in its own right (Mäntymäki & Najmul Islam, 2015). Moreover, recommendations can support the discovery of new music and the broadening of one’s musical taste (Mäntymäki & Najmul Islam, 2015), which may especially be important for users who want to immerse themselves in the listening experience, that is, for users who interact with the music streaming service, because they want to listen to new music (Garcia-Gathright et al., 2018).

As users differ regarding their preferred activities (i.e., their reasons to use a music streaming service), researchers have proposed different user personas (Lee & Price, 2015; see also Fuller et al., 2016^; Lee et al., 2019). Note, however, that these personas are not necessarily conceptualized as stable personality traits: While an individual may identify with a specific persona, it is also possible to show different personas—or a combination of personas—at different points in time. Among the seven proposed personas, there are three, which mirror the tendency to avoid the use of recommender systems, either because music is considered to be something very personal (music recluse) or because one does not believe that recommender systems provide adequate and helpful recommendations (non-believer, music epicurean). Both the wanderer (who likes to be surprised and to discover new music) and the active curator (who puts a lot of effort into curating music) have a high interest in engaging with different kinds of music. On the other hand, both the guided listener (who does not have a strong musical taste and is therefore willing to hand over control to the music streaming service) and the addict (who listens repeatedly to favorite songs and albums) have a comparably low interest in engaging with different kinds of music.

However, how can the different activities and the different user personas be analyzed from the perspective of the proposed model? As far as the different activities are concerned, some seem to be driven by controlled motivation as they are aimed at achieving a particular goal (e.g., playing music in the background, finding new music for later); in contrast, others seem to be driven by autonomous motivation as they are more about the process of interacting with the music streaming service (e.g., engaging with new music, discovering new music and broadening one’s musical taste). The differentiation between autonomous and controlled motivation can also be used to distinguish the user personas from one another. Note also that this allows for a broader perspective, as the opposition between controlled and autonomous motivation is—other than the enumeration of different personas—not limited to the context of music streaming.

Clearly, the wanderer and the active curator, who have a high interest in engaging with new music, seem to represent instances of autonomous motivation, while the guided listener and the addict represent instances of controlled motivation. Crucially, all of these personas do not mind relying on recommendation algorithms—although they may have different expectations regarding their operating characteristics. However, the opposite is true for the remaining three personas—the music epicurean, the music recluse, and the non-believer—who genuinely dislike and avoid music recommendations. This is particularly interesting, as the music epicurean and the music recluse also place a high value on engaging with music, that is, as they also represent instances of autonomous motivation. Overall, this seems to indicate that autonomous users may have ambivalent feelings with respect to music recommender systems.

In short, previous research has demonstrated that humans differ regarding their motivation to use music streaming services and that their satisfaction with these streaming services consequently depends on different aspects of the personalization process. The existing findings can be reinterpreted using the proposed model. However, there are some components, which have not been explicitly investigated empirically so far. In particular, user satisfaction was typically not operationalized as the satisfaction of the basic psychological needs, but rather as an overall evaluation of the system or as the intended long-term adoption of the system (see, e.g., Jones, 2010). Consequently, we know only little about the differential impact of personalization in the context of music streaming on the satisfaction of the three basic psychological needs for autonomy, competence, and relatedness. While the above-mentioned evidence makes it plausible to speculate that experiencing a lack of autonomy is what commonly upsets users with respect to the personalization of their music streaming experience, competence and relatedness have received limited attention in previous research. With respect to competence, one may at least suspect that those autonomously motivated users that do not want to rely on recommender systems (such as the music epicurean or the music recluse) feel that algorithmic selection undermines their competence. How the personalization of music streaming affects social relations, however, remains relatively unclear. One could hypothesize that personalization ultimately disrupts the shared social sphere, which might be particularly problematic for autonomously motivated music listeners who place great emphasis on sharing and discussing music with others or who use music as a means to achieve social distinction (e.g., Webster, 2021). However, personalization could also enable the formation of new social groups who share an interest in the same kind of music (e.g., Park & Kaneshiro, 2021). Against this background, research in the context of music streaming that explicitly addresses the impact on the basic psychological needs is highly desirable. Apart from that, it seems worth exploring to what extent the motivation to use a music streaming service remains stable over time and to what extent it depends on the specific context. Moreover, one may investigate what distinguishes those autonomously motivated users who rely on music recommendation algorithms from those who do not.

Online Dating

Using online services has become a part of dating for many people. For instance, about one third of U.S. adults state that they have used a dating site or app—among young adults it is even half of the population (Pew Research Center, 2020). Overall, these online services are evaluated rather positively and the number of couples who have met online has been increasing over the last years (Pew Research Center, 2020^; Rosenfeld & Thomas, 2012^; Rosenfeld et al., 2019). In fact, more than one third of American couples that married between 2005 and 2012 had met online (Cacioppo et al., 2013). From a theoretical point of view, it has been argued that online dating differs from offline dating in three crucial respects (Finkel et al., 2012). First, online dating services provide convenient access to an unprecedented number of potential partners: The number of people that one can get in touch with exceeds the number of people that one meets in one’s everyday normal life by far. Second, online dating services enable users to engage in computer-mediated communication before meeting in person. Third, many online dating services use matching algorithms that use the data provided by the users to suggest potential partners, typically based on the principle of either similarity or complementarity (see also Fisher, 2009^; Schwartz, 2006^; Warren, 2002).

At first glance, this sounds promising—and it certainly is in some cases and for some individuals. Indeed, it has been reported that couples who have met online show slightly lower separation and divorce rates as well as higher marital satisfaction compared to couples wo met offline (Cacioppo et al., 2013; but see Joel et al., 2017). At second glance, however, online dating services are also associated with various downsides that flow from the combination of features that these services offer. First, having access to more potential partners is not always experienced as positive. Rather the contrary, choice overload can undermine feelings of autonomy and control within the dating process (D’Angelo & Toma, 2017^; Lenton et al., 2008^; Lenton & Francesconi, 2010). Ultimately, the feeling of being overwhelmed by the available options, that is, by the sheer number of people that one could and may want to get in touch with, often leads users to making superficial and ill-advised decisions (Chiou & Yang, 2010^; Pronk & Denissen, 2020^; Wu & Chiou, 2009^; Yang & Chiou, 2010). Second, and maybe even more importantly in the present context, knowing that various kinds of information about the users of an online dating service are compared and matched using an algorithm may elicit “an evaluative, assessment-oriented mindset” (Finkel et al., 2012, p. 3). By offering access to an extremely high number of potential partners as well as a means to deal with this number, online dating services may encourage users to engage in what has been called “relationshopping” (Heino et al., 2010) or the “McDonaldizing of romance” (Ahuvia et al., 2009; see also Lawson & Leck, 2006). In the long run, such an objectification of potential partners can have negative effects on the dating life of the individual: If one is constantly hoping that “an even more congruent soul mate is waiting […] on the next website” (Coren, 2006), the willingness to commit to one particular person is likely to be undermined (see Finkel et al., 2012).

In addition to these general observations, it should be noted that different people use online dating services for different purposes and that this may also affect their user experience (Gatter & Hodkinson, 2016^; Hsiao & Dillahunt, 2017^; Lawson & Leck, 2006^; Sumter et al., 2017^; van de Wiele & Tong, 2014; see also Zytko et al., 2018). While online dating services are typically used to find a partner for a long-term committed relationship, users also look for new friends and people to talk to (e.g., to become part of the local community after moving to a new city) or for casual sex partners. Apart from that, there are users who do not aim at fulfilling such social relationship needs, but who use the websites or apps just for fun, that is, because they find it entertaining to browse the different profiles, because they are curious about the world of online dating, or because their peers use the service and they want to try it as well. Moreover, users may seek the thrill of excitement and think of online dating as some kind of adventure or look for self-worth validation trough receiving positive feedback and compliments from others.

In terms of the proposed model, this clearly shows that users of online dating services differ regarding their motivation. While the wish to get to know new people or a potential partner is likely to be driven by autonomous motivation, users who engage in online dating to receive approval from others or to improve their self-esteem are driven by controlled motivation. Arguably, these motivational differences also influence the perception of the matching algorithms. For a user with controlled motivation, the potential datafication and objectification of individuals should not pose a problem per se. If one is looking for positive feedback from others or for a casual sex partner, for instance, it does not matter whether the algorithm finds the “perfect” match or whether the algorithm reduces people to two-dimensional displays of information. Simply put, such a user will already be satisfied with the recommender algorithm if it leads to getting positive feedback from someone or the opportunity to have sex with someone, that is, if the algorithm helps to achieve certain instrumental goals. On the contrary, an autonomously motivated user will not be satisfied with any match. Rather, an autonomously motivated user desires at least some control over the dating process and refuses to see others as an accumulation of data points. Against this background, it is worrying that matching algorithms seem to encourage an assessment-oriented mindset: This could imply that the users’ autonomous motivation is being undermined by the very setup of online dating services. The marketization of romantic relationships that is reinforced by algorithmic filtering may hinder and disrupt the ideal dating process that many users have in mind. Interestingly and in line with this, it has been demonstrated that users are most satisfied with online dating services which combine active user choices with automatic algorithmic matching (Tong et al., 2016), that is, with online dating services that help the users to deal with the high number of other users, while still giving them the chance to freely interact with all of them. Again, it should be emphasized that these conclusions are in line with the existing literature, but that more research is needed to confirm the idea that the effect of matching algorithms is linked to the individual’s motivation. Once more, this particularly applies to the connection between personalization and the satisfaction of the basic psychological needs. The above-mentioned studies allow at least some empirically informed speculations. On the one hand, it seems that matching algorithms can undermine autonomy and competence in a field where these two needs are particularly important to many people. When it comes to finding a partner for a relationship, individuals desire control over the dating process as well as the feeling that they are capable of deciding who is good for them and who is not. On the other hand, the marketization of romantic relationships may also endanger the satisfaction of the need for relatedness. The algorithmically enforced notion that one is not interacting directly with other persons, but rather with abstract profiles that are compared and matched, could also explain why some people remain skeptical with respect to online dating. In view of the omnipresence of online dating in modern societies, further clarifying these hypothesized causal links and mechanisms is a crucial task for future research.

Discussion

The personalization of digital environments, which is based on adaptive algorithmic processes, is something fundamentally new: It creates a virtual world that is different for each individual—and that constantly changes in different ways for different individuals. Hence, it is of utmost importance to understand the consequences of this technological leap for human behavior and well-being. Against this background, we introduced a model with five crucial features. First, the model is focused exclusively on the impact of personalization processes. That is, instead of taking a concept such as “screen time” as a starting point for our investigation that is too broad to capture the complex processes underlying the interactions with digital environments (Granic et al., 2020^; Orben et al., 2020), we picked a specific phenomenon with specific operating characteristics that drives many of the contemporary changes in the digital realm. This gave us the opportunity to provide a detailed analysis of the mechanisms of personalization and their impact on the individual. Second, the model is based on an established theory of motivation, needs, and well-being: Self-Determination Theory (SDT; Deci & Ryan, 2000^, 2008^; Ryan & Deci, 2017). Using SDT as a starting point allowed us to identify the psychological impact of personalization in digital environments, a novel phenomenon, by building on a strong foundation of knowledge. This seems particularly important as investigating the psychological effects of personalization is still an emerging field of research in which many processes and mechanisms remain to be fleshed out by empirical studies. Third, however, our model is not merely a working model, but able to integrate a wide range of findings from different areas of application, as we have demonstrated using the examples of news feeds and websites, music streaming, and online dating.

Fourth, based on the distinction between autonomous and controlled motivation, our model explicitly allows for individual differences. That is, we take into account the fact that not all individuals react to personalization processes in the same way. Of course, there may be other sources of interindividual variance beyond motivational differences that could be used for expanding the present model (e.g., power usage, trust or prior experiences with personalization; Glikson & Woolley, 2020^; Sundar, 2020). As the motivational differences that we have discussed are deeply rooted within the individual and closely connected to need satisfaction and ultimately well-being, however, it seemed justified and important to use them as the key building block of our model. In other words, we assume that the motivational differences lie at the heart of many of the differences that can be observed between individuals. Fifth, and related to the previous point, the model offers a nuanced view of the potential risks and benefits of the personalization of digital environments. Simply put, we neither claim that the personalization of digital environments is good or bad per se. Rather, our model provides a framework that can help to determine when and why personalization may have which positive or negative effects. So which conclusions can be drawn based on the proposed model and the studies that we have interpreted through its lens?

Conclusion

Well-being in the age of digital media is matter of concern for scientists and the public alike. Our model and evidence from three fields of application suggest that motivational predispositions predict the extent to which personalization of digital environments leads to need satisfaction and well-being: Whenever individuals are driven by controlled motivation, efficient personalization is the key determinant for user satisfaction. However, this kind of user satisfaction can only guarantee hedonic but not eudaimonic well-being. In addition, autonomously motivated individuals will only achieve need satisfaction and ultimately eudaimonic well-being when they experience the personalization process as transparent and when they have the opportunity to play an active role in it. As navigating through personalized digital environments is daily business for most individuals living in modern societies, considering these psychological effects when designing recommender systems and personalization algorithms seems highly relevant—not only from an individual but also from a societal perspective.