Music streaming,cultural consumption and the everyday routines of algorithm management: Exploring how trust and objective setting shape everyday encounters with algorithmic systems

Abstract

Algorithms are often understood to be a powerful presence in shaping everyday life. In particular, the established part that algorithmic recommendations play in cultural consumption is important in shaping what people consume and how tastes are formed. Understanding how users of platforms respond and react to algorithmic processes is crucial for understanding cultural consumption today. This article draws upon semi-structured and detailed interviews with music listeners located in Mexico to explore the way that music streaming recommendations are encountered. The article argues that trust and objective setting are important for understanding how individual listeners seek to manage these recommendation systems. The article shows how everyday routines of algorithmic management are established and how listeners actively shape the algorithmic outcomes and recommendations to which they are exposed.

Introduction

The shifting structures of music consumption have created a fundamental question for listeners: what should I listen to next? The anticipation of the next song to be played is a key part of the temporality and experience of contemporary consumption. The automatic selection of what comes next is fundamental to how culture is experienced. How these vast music archives are navigated has become a central question in everyday cultural consumption. There is a renewed interest in how to locate music within these structures. As Sofia, a postgraduate student, noted during an interview in Mexico City, ‘I get excited with Discover Weekly, every Monday I really want to know what Spotify added to my list’. This brief observation of excitement at what might come next reveals the vitality with which these archives are managed; at the same time, it also suggests something of the way that algorithmic sorting processes themselves captivate cultural consumption. Sofia is keen for the recommendation system to help sort and discover music. As we will explore here, this is far from a one-way process in which the algorithm determines what is consumed. In this article, we look at how individuals specifically interact with those algorithms to get the desired outcome and how, for these participants, these algorithm systems become objects of trust that it is possible to use with specific objectives in mind.

Due to their vast music catalogues, streaming services have inevitably become spaces to find new music (Datta et al., 2018). They are spaces of music discovery that mediate what is listened to as well as notions of genre, scenes and artists. The platform’s recommendations are based upon capturing the listeners’ interactions and a complex recommendation system analysis (Drott, 2018). As a set of structural mechanisms, these feed into various listening practices. The question we want to focus on here concerns how these practices can be understood in the context of algorithmic intervention. Of course, for listeners, discovering new music has always been mediated. Radio, television, magazines or more recently Genius in iTunes or last.fm create particular options for finding what to listen to (Beer, 2008, 2013). Still, functions like ‘radio’ or ‘discover weekly’ on different streaming platforms produce alternative possibilities for the user, in which automation plays a more significant role in listening practices. This article shows how listeners use their knowledge and experience of those algorithmic recommendation systems in streaming platforms to discover new music, highlighting how trust and establishing objectives are important to understanding these listening experiences. Its central contribution is to bring forward these listener objectives and the way listeners enact and realise these objectives while managing their constraints directly into the analysis of algorithmically defined cultural consumption.

In a recent article, Marika Lüders (2021) explains that streaming service users value more than the mobility of listening and the vast catalogue of music available through streaming. Listeners enjoy being able to do things with their music and interact with their platforms as a way of personalising them. Personalisation itself becomes a part of how music is mediated and experienced. So, it is crucial to understand the active part an individual plays in the way that their profiles become personalised – these are not just produced by active algorithms and passive humans. Saving songs for listening to later or ordering them into a playlist are just two instances of how a sense of ownership and intimacy can be created and how personalisation can be actively managed; there are others. The present article suggests that intimacy shapes the experience of using recommendation systems. With its familiarity, we show how listeners know when to relinquish control as they develop trust in the platform for selecting music. As such, the argument follows on the premise that when receiving algorithmic recommendations, the listener is not just a passive receiver (Siles, 2023; Velkova and Kaun, 2019) but engages with these systems in response to the recommendations they are receiving. Using their previous experience and strategies to make sense of algorithmic action, the listener caters for the algorithm in an attempt to manage it (Lüders, 2019). Both listeners and recommendation system enter an iterative and cyclical process in which these ‘users incorporate algorithmic recommendations into everyday life as much as the platform works to colonise users and turn them into ideal consumers through its algorithms’ (Siles et al., 2019a: 500). In short then, this article argues that objectives are used to manage the personalisation processes of algorithms and to ensure that the individual does not feel like their music profile is solely a product of the algorithmic system. Rather, the objectives that listeners set out will anchor the processes and maintain a sense of individual taste within the algorithmically managed spaces of streaming, and when this does not happen, the listener actively follows steps to correct or change the action. As such, trust emerges from objectives accomplished, which makes them feel in a mutual act of learning from each other.

Trust, in our argument, follows the conceptualisation by Karakayali et al. (2017). In their exploration of recommendation systems, they argue that a notion of technologies of the self can be applied to the integration of algorithmic processes in consumption. Taking this concept from Foucault (2008), they show how algorithms are used to shape and transform aspects of the listener’s life, becoming ‘trusted companions’ in listening. In their use of this concept, they noticed that these recommendation systems are not continuously operating as self-publishing tools as in the case made by Foucault, but more as intimate companions that still produce self-reflection and transformation of practices. Their exploration suggested a two-way argument: on one hand, the idea that recommendation algorithms exercise an overpowering control of the users, inducing a continuous state of self-transformation; on the other hand, the possibility of suggesting a symbiotic relation in which the function of these systems is made possible by the users’ willingness to cultivate themselves in new ways. In their argument, one does not exclude the other.

Trusting the recommendation system is a matter of practice, meaning it is also rooted in users perceiving it as a means to specific ends. Listeners learn to use those algorithms for their own objectives. For this to happen, listeners explain algorithms and their possibilities by deploying their own ‘folk theories’ about their operation and action capacities (Siles et al., 2020). Using algorithmic recommendations for specific objectives has been more widely studied from the creators’ perspective. Communities of practice, such as Internet celebrities, use shared ideas and knowledge (Bishop, 2019) to negotiate the rules established by engineers to achieve visibility by adapting their own behaviour (Cotter 2019). This form of setting objectives produces new experts among their users (Bishop, 2020). On the other hand, Cotter et al. (2022) explore perceptions of the recommendations page on TikTok, viewed as a cosmic intervention and embraced as a tool to know more about themselves through its recommendations. However, algorithmic recommendation systems do not solely enhance self-knowledge but relate the individual with the world around them.

Following this framework draws us towards an analysis of how listeners are active users of these algorithms to modify their experience and themselves in the social world. According to Siles et al. (2023) it requires us to think about how this algorithmic companionship operates and develops in different contexts through imaginative explorations and contextual practices without expecting a pure version of either submission or resistance to the capitalist logic of it. As such, while the article presents a form of objectives, it also attempts to shine a light on how listeners challenge the algorithm when it is not achieved and how it also learns to use it for their purposes.

Overall, the article presents a way to understand algorithmic action as something that is embedded in social and individual everyday practices, and as such it can only be as effective as it is inserted in everyday life through a process of accomplishing objectives and gaining trust. This is framed on forms of consumption attached to specific contexts, not only socio-cultural but also spatial, such as the car, the office, or the subway. For that purpose, the following section details the methods for data collection and Mexico as a research site. Afterwards, the article presents a succinct discussion of the recent articulation of music, algorithms, and the self. The following two sections present the main findings and debate the article puts forward, separated by trust and objectives, even when detailing that they are not always separate entities. The article concludes with a commentary on how algorithmic trust can be understood.

Method

The analysis presented in this article is based on 36 interviews conducted with music listeners in Mexico. While the initial idea was to cover a diverse demographic range, finding the diversity of practices and modes of consumption also became important. However, the key selection criteria ensured that participants were active users of music streaming platforms despite the nascent market conditions in Mexico at the time.

The interviews were conducted between 2016 and 2017, with ethical approval obtained from the University of York. Participants ranged from 18 to 60 years old. The interviews included listening to music with participants, not only as an elicitation (Allett, 2010) but also as an object to discuss with the listener (Thompson, 2012) and to share moments and practices with them (Green, 2015; Pink and Mackley, 2013; Woodward and Greasley, 2015). This helped to reveal aspects of everyday engagement with music and with the media through which that music was consumed. The initial analysis was developed through constant comparison guided by Constructivist Grounded Theory (Charmaz, 2014: 1; Clarke, 2005; Star, 2007). In the case of this article, these interviews were drawn upon to see how this type of algorithmic recommendation and companionship develops and becomes established.

At the time of the data collection, streaming platforms had been recently introduced to Latin America. For example, Spotify was introduced to the country in 2013 (BBC News, 2013), while one of the primary sources of consumption was counterfeit copies or compiled mp3 CDs (International Chamber of Commerce (ICC) and Business Action to Stop Counterfeiting Piracy (BASCAP), 2017; Ávila Torres, 2016). In this context, none of the platforms were dominant in the market. However, participants were selected based on their active use of music streaming platforms with capabilities for algorithmic recommendations either as free users or paid subscriptions, ensuring a representative sample despite the nascent market conditions in Mexico at the time. As will be seen, most participants use more than one platform depending on their objectives and access possibilities. The most popular platforms were Spotify, Apple Music, and YouTube. It is useful to understand that most participants decided not only on their own convergent economic and technological situation but also on their own willingness to try new platforms, such as when some used Apple Music via iTunes, maintaining some aspect of their previous libraries and devices. In some other cases, it would involve using multiple platforms depending on the situation and the cost; most of the users of the time would use multiple forms of music consumption in everyday life (Ávila Torres, 2023), instead of committing fully to one; this has to do with technological demands, but also with specific constraints, such as having Internet service on a mobile phone, or having a limited data plan.

In contrast, more recently Mexico has been placed at number 14 on the list of countries generating most income from streaming platforms, according to the reports collated by the digital laboratory Signa_Lab (2022). The same report indicates that Spotify represents the dominant platform in the music streaming sector, with a market share of 84 per cent in Mexico out of 57 million total users, which amounts to less than 50 per cent of the total population. The number reported does not indicate a separation between premium and free users of the platform, as well as those that use multiple platforms or a mix between streaming and other forms of consumption, such as counterfeit CDs. Street piracy or the sale of counterfeit copies adds complexity to the mix. A report by the American Chambers of Commerce covering music and other products indicates that piracy still has a multi-million impact on the Mexican economy (Avila Munoz, 2020). The popularity of this practice is still easy to perceive by a simple walk around any market in which video games, music and films can easily be acquired on the street (Ávila Torres, 2016). Finally, considering that almost 25 per cent of the Mexican population still does not have access to the Internet (Rodriguez, 2023), it helps us to nuance totalising affirmations about the dominance of streaming in economies such as Mexico’s, where users are constantly deciding while navigating infrastructural challenges, costs, and objectives to achieve.

The article then moves away from a focus on Spotify, which currently concentrates most of the scholarly attention. The article focuses on the listener’s relationship to algorithmic recommendations, and while the design and interface of the platform plays an important role there is no room for a detailed analysis of such. However, the article attempts to include the listeners’ differentiation strategies, which contributes to setting objectives for the platform and developing creative strategies to overcome access barriers.

In terms of the contextual settings in which everyday consumption decisions are made, the focus of the data set used for this article is Mexico. Using this data, the article provides only accounts from listeners from outside more dominant accounts in the Global North, contributing to challenge portrayals of Universalism (Milan and Treré, 2019) and expanding on the knowledge of the experience of algorithms in the Global South empirically (Siles et al., 2023). To achieve this goal, it must be noted that the material below focuses on the experiences of Mexican listeners, mainly from urban areas around Mexico City, Cancun, Puebla and Cholula.

Finally, we consider it important to note that while the project is focused on the use of music technology in everyday life, there is no presumption of the importance of music for these listeners. Their use of music as accompaniment for other activities and entertainment seems to be a significant driver for their consumption, but recommending systems seem to play a role in providing them with a sense of control and agency over those practices. Social drivers are also implied, such as being a musician or becoming knowledgeable about certain genres. The article attempts to shed some light on how those appear as a consequence of having experience interacting with such algorithms.

Following a discussion of the algorithm and the self, the article’s findings are split into two sections that reflect two key stages. The first explores how the listener develops a relationship of trust with recommendation systems and how this trust is built in a balance between control and delegation. The first step is about trust in the algorithm, and the second concerns the way that the objectives of listeners become a focal point for a routine type of algorithm management.

Streaming, recommendations, and the self

As has been widely discussed, algorithms now perform a pivotal role in consumption, interaction and even self-making by becoming a regular presence in everyday life (Willson, 2016), by shaping aesthetics tastes (Gaw, 2021) (Beer, & Taylor, 2013), our memories (Jacobsen, 2020), visibilities (Jaramillo-Dent et al., 2022) and keeping us engaged for longer (Seaver, 2019), therefore participating in an overall change of understanding the world (García Ramírez and Valle Jiménez, 2020). Taina Bucher (2017, 2018) has argued that analysing algorithms as social assemblages is necessary by following their outcomes and their effect on the individual and the social world. Bucher (2017) has influentially argued that algorithms produce ‘imaginaries’ – impacting how they are understood and how they shape understandings – that influence social relations in various and sometimes unseen ways. By following this principle, Siles et al. (2020) suggest paying attention to the way individuals explain algorithmic action and the possibilities and limits they get offered as ‘folk theories’. The attention of these ideas to the participation of recommendation algorithms in forming subjectivities underlines the importance of understanding how algorithms are understood in the connections and relations people have with them.

Foucault introduced the concept of technologies of the self to analyse the strategic operations and practices individuals conduct to transform themselves or bring the self into existence (see Foucault, 2008). Seeking to adapt this broad conceptual position for contemporary mediated experiences, Karakayali et al. (2017) point out that these practices require ‘complex assemblages comprising discourses, other people (e.g. disciples and teachers), institutions (e.g. schools or monasteries), meditative techniques and a whole range of tools (e.g. letters and diaries)’ (p. 6). The purpose of drawing upon the notion of technologies of the self is to draw attention to these ways in which self-making occurs through such media and these systems become ‘intimate companions’ in which trust allows the completion of specific objectives to become multiple distributions (Siles et al., 2020), negotiations and tensions between control and passiveness.

If we are to think in terms of these everyday engagements and notions of the self, Tia DeNora (1999) developed a conceptual framework that allows us to follow how reflective activities of everyday music consumption enable individuals to produce themselves with ‘specific feeling forms, attributes and identity characteristics, and as objects of knowledge to themselves and others’ (p. 53). In DeNora’s account, music becomes a resource for everyday life that the listener uses strategically to manage their mood or adjust the rhythm of certain activities. And so here again, the idea of a kind of set of strategic practices in which cultural consumption is a part of self-making is suggested. Eric Drott (2019) brings such a perspective into the realm of streaming services and their promotion of contextual-based services – like ‘music for work’. In Drott’s argument, DeNora’s concept presupposes the existence of a part of the self that needs to be modulated through music. Streaming services promise to suit and fix a mood. Those aims lead to consumption and regulated identities, and they also show the listener’s process of establishing purposes beyond passive consumption in the background.

Inevitably, the production of subjectivities through streaming services needs to consider the spaces for listening and the content circulated to the listener. For Anahid Kassabian (2013), these digital affordances change in what way we make sense of ourselves in the world, producing what she calls a distributed subjectivity. By living surrounded by music embedded in cloud-based streaming services, the listener develops different states of multi-focused attention in constant flow, changing not only through external factors – noise, work, the city – but also our emotions. The subject becomes a node in the streaming network, developing multiple forms of being that, in interaction with algorithmic filtering, phase in and out. Affects and emotions produce those distributed subjectivities, moving through everyday interactions with ‘sourceless’ musics that lead to moments of attention and intense engagements (Kassabian, 2008). The flow of information and the attachments to music cannot always be predicted. As Ferrari and Graham (2021) have argued, there are ‘fissures in algorithmic power’, moments when they do not produce expected or intended outcomes. Hence, ubiquitous music produces identities through technologies in which the listener balances multiple levels of engagement and unexpected encounters (García Quiñones, 2007). These moments of consumption are convergent with the past and the experiences and intentionality of the listener (see Nowak, 2016a). Streaming services and their recommendation algorithms work in tandem with the listener to produce agency and subjectivities. While platforms act as gatekeepers for cultural content, forms of cultural and personal value emerge in the interaction between those technologies, the user and the cultural context (Bonini and Magaudda, 2023). The listener uses them to manage their listening practices and some aspects of the self, like their taste. The listener uses those systems that produce forms of control over them in creative ways, reshaping the algorithm itself and playing with the data.

The algorithmic recommendation system is in constant flux, integrating to their mathematical model the interactions of the users, who are aware of their role in the system, which is why this has been described as a process of mutual ‘domestication’ (Siles et al., 2019a). Consequently, the listener might seek to manage the potential inconsistency in their consumption brought about by the vast flows of content on these platforms (Prey, 2016, 2018). Clearly, as powerful as algorithmic systems might be, the listener cannot be portrayed as a disengaged subject; rather they are developing forms of action that will produce a distributed agency. Karakayali et al. (2017) argue that through interaction processes over time, recommendation systems become intimate experts that gain the listeners’ trust. As part of this trust-making, the recommendation system might even adopt human-like qualities (Siles et al., 2020), negotiating forms of control to achieve specific objectives. In those relations and negotiations, forms of intimacy and subjectivity emerge (Fisher, 2022; Lüders, 2021) beyond the dual framework of control and delegation. Hence, the algorithm’s outcome is the perceived possibility of producing the self as an individual within the vast archives and flows of cultural consumption.

This is a form of access to culture in which the very terms are structured by the listening practises themselves. This means that the listener may still develop a sense of ownership (Lüders, 2019) even when they only access content. At the same time, they also shape the platform and its algorithms through these practices, and so they become a reflection of themselves (Siles et al., 2020). Consequently, the platform and the listener shape each other through mutual interactions. This is embodied in routine acts like selecting what list to play for a specific activity (Eriksson, 2020), creating playlists (Hagen, 2015) or changing music that does not fit with the surroundings or circumstances (Nowak, 2016b). The personalisation of experience requires interaction that raises convergent affordances that the listener negotiates and uses to fit their interests; individuals participate in personalisation.

Maria Eriksson and Anna Johansson (2017) explore this participation in personalisation by analysing the ‘featured content’ in Spotify. They explain that to give a sense of a personalised experience, the platform acts in a prescriptive and controlled manner, setting up possibilities for consumption. This ‘featured content’ is based upon trying to predict what the listener is doing at different times of the day, offering specific materials to listen to for those activities. Here the routines of everyday life are something to be anticipated as part of structuring consumption. For Eric Drott (2018), for instance, this is translated into a focus on the importance given to the ‘next song’. To explore how this engagement with the next song comes about and how it is part of the way individuals negotiate and manage algorithmic streaming, we now turn to two key factors. The rest of the article looks first at trust and then moves to an exploration of the role of objective setting in the routine management of algorithms.

Trust in the everyday routines of algorithm management

It is perhaps obvious that, for many people, music streaming algorithms are deeply embedded in their everyday routines. Within a relatively short space of time, algorithmic recommendations have become routine in everyday consumption. Take the following as an illustrative example of this ordinary embeddedness and prescriptive power:

Most of the time [in the car] I just put on Apple Music, I press play and I like that the iPhone tells me what to listen to. If I ask for a genre or for the newest stuff, it will play it. I am always trying to get new music, or at least music that I don’t know yet. I save all the music I like in a playlist called ‘favourites’, so when the iPhone throws songs I like, I just save them there. Then, when I don’t feel like experimenting, I just go to that list, and I have a very long list of things I really enjoy. (Raúl, 35, Graphic designer and Musician. Mexico City)

Here, a picture emerges of how the algorithm becomes woven into practice. In this case, the algorithm takes over when the individual listener does not wish to experiment for themselves. In Mexico City, a regular commuting journey could take around 2 hours, so for Raúl, this becomes a space that streaming fills. Listening while driving becomes an essential accompaniment for Raúl – the focus on driving creates the opportunity for the algorithm to take on responsibility for music selection and thus move to the inside of the everyday experience. As with other interviewees, Raúl delegates not only the music choice to his device, but he also even describes it as an agent that ‘takes’ decisions. This is an agent that can, for him, be trusted. There is an acceptance here that these decisions are automated. Driving as a contained space offers an important space for consuming music. However, in the data, it appears alongside using the radio, a long-established and trusted medium that helps us understand the trust listeners put in algorithmic recommendations. Listeners perceive algorithms as entities that they have shaped and fedback on, while radio seems to respond to broader trends. Then, the trust in algorithmic recommendations emerges from their interactions. The role of the user in experimenting with their journeys is key, as is interacting with the device and embracing the possibility of modifying the outcome. Raúl indicates that he learns to trust in the algorithmic selections while he also develops a better understanding of himself by knowing in advance when certain music could work better to fine-tune his experience. In this sense, the trust placed in the algorithmic selections reinforces a sense of self and control in his experience. In the interaction with the algorithm, the listener can insert their own feelings and interests (Heye and Lamont, 2010). For instance, by asking their device for a specific genre, the listener sets the boundaries of the algorithmic action, leading to iterative processes. Listeners like Raúl, when listening while doing another activity, are in a balance between control, surprise and distraction.

Trust in algorithmic recommendations can be understood through relational trust, where repeated positive interactions build a sense of reliability and predictability (Karakayali et al., 2017). This has been discussed in the literature, arguing that trust in algorithmic systems rests on the user literacy of the technology but also on the content found (Shin, 2021), and the context in which the encounter happens, being mediated by societal norms, values, and accessibility (Ananny and Crawford, 2018). This study adopts a nuanced view of trust, considering it as a dynamic process influenced by users’ experiences and interactions with the platform. Trust is not merely a static attribute but evolves through continuous engagement with the algorithmic system. This type of trust in the algorithm is built through a sense of its prediction accuracy emerging from the listener’s actions. However, in contrast, the algorithm is also trusted where the listener feels there is some grasp of their taste or where it is thought to be able to read taste from prior music consumption. This is often captured in phrases like ‘hit the spot’. The listeners adapt to platforms depending on this sense of trust founded on accuracy. As one interviewee explained,

The best option for me is YouTube. For listening to new music related to my taste randomly. YouTube has better recommendations than Spotify. Spotify never hits the spot when recommending something to me. It is like . . . something is missing . . . I just feel like the algorithm that YouTube uses is better in determining what song to play next to Spotify’s. But maybe it is because of the way I use it. (Gloria, 30, Designer, Cancun)

This position turns back the way in which recommendation systems for different platforms are experimented with through expectations and imaginaries of their role – what it is supposed to do for me (Bucher, 2017). This account suggests experimentation by the user, but also tensions within the boundaries of that experimentation, as free Spotify accounts did not allow the same level of control of other ‘free’ services. This means that the possibilities offered are strained by social factors, such as the possibility of paying for multiple services or not. Therefore, the listener experiences those algorithms through a ‘constrained agency’ (Hesmondhalgh, 2013), as the experience of them comes from a limited range of possibilities dictated by its membership and the way the recommendations system makes its decision. Its ‘algorithmic imaginary’ (Bucher, 2017) leads this listener to the platform they feel most accurately selects an appropriate next track. While the commercial strategies of platforms like Spotify influence user practices, this study focuses on how these strategies intersect with trust in algorithmic recommendations and the setting of listening objectives. For instance, the limitations of free Spotify accounts, such as restricted skipping and advertisements, shape how users interact with the platform and develop trust in its recommendations. This commercial framework is crucial for understanding the broader context in which algorithmic trust and user objectives are formed. A selection of which platform suits this self-making best seems dictated by the context. Also, trust in the algorithm emerges when its track record for that individual suggests its apparent accuracy in predicting or working with their taste.

By finding those states of balance and the achievement of those effects, the listener trusts in the algorithms to serve their purpose of identifying a suitable next track. An interesting feature of this trust is its development through multiple encounters in which both the platform and the listener learn to cater for each other, a process of mutual ‘domestication’ (Siles et al., 2019a). Through those processes the listener seems to value their own possibility of acting through the platform:

The way it works is that it tells you the songs that you have been listening to, it can tell you more about the artist, and so on. So I just play the artists, one song, and I let it keep going with the reproduction. When something that I do not enjoy comes up, I just skip it. Bit by bit, it goes on making mixes of your favourite songs, so sometimes I just put that list to not let anything else in. (Israel, 20, Student, Cuautla)

An expression of the constrained agency was explained before through the very interface of the platform. The listener here responds where the algorithm gets things wrong by skipping while accepting that predictions don’t always need to be accurate to be trusted. The affordances and constraints that guide the interaction between algorithms and listeners go beyond the interfaces and outcomes – skip buttons or lists – but they are also conditioned by broader forms of access. Maria, who did not have access to the paid version of Spotify at the time of the interview, explains,

There are some things that make me enjoy YouTube more, because I can skip songs or go back, I cannot do that in [the free version of] Spotify. Still, those two are my most used apps. If I want to listen to a complete album I just write the name, but regularly I would only hear two or three songs of that artist and I would let it go by itself. I also created a list with what I listen to everyday that I can trust if the recommendations are not working very well or if the songs are just out of fashion. I just do that by saving when I am listening. (Maria, 18, Student, Cholula)

Among the participants, algorithmic recommendations and platforms are also described with human-like qualities (Siles et al., 2020). Maria is one of the youngest participants in the project, and she comes from a rural area and is doing a degree at a public university. Still, she seems very interested in what she gets from the algorithm and not in its limitations. The algorithm, trusted for its human-like qualities, is a companion that is expected to know what to recommend. It is expected to follow and listen to the listener’s interaction (Kibby, 2006). Through that process, the recommendation system is sometimes understood to be like a trusted friend shaping how the listener positions themselves and the cultural landscape they are then engaged within:

I usually start with a playlist in Spotify, then I let it start the ‘radio’ function from it. I listen to what it recommends to decide which ones I like more. There are also certain friends who recommend lots of music to me . . . but to be honest . . . I do not always like them, I enjoy music that is not very popular. I have struggled to find friends with the same tastes as me, they might like similar things, but not exactly the same. I think 2 or 3 friends can recommend something to me, something that makes me say ‘I really like this’. [. . .] To be honest, my source of new music is Spotify and YouTube, I really enjoy the automated mixes that they have for me. (Gloria, 30, Graphic Designer, Cancun, Mexico)

Here, again, the listener still exercises preferences to which a future response is expected. There is a sense that such actions lead to the algorithm sharpening its future predictions. At this point the convergence of the discovery of music gets expanded. If as explained, the listener prefers social and intimate recommendations for new music (Johansson, 2018; Rivera Magos and CarriÇo Reis, 2015), the recommendation system becomes that intimate social connection. The algorithm is treated as a guided friend that will provide those affective experiences through the balance and flow through different levels of attention. The algorithm though is also framed as an expert who knows the individual’s listening practices. The element of trust in the everyday engagement with algorithms emerges through a sense of its responsiveness to past choices and the guiding of future music consumption.

There are, of course, subtle distinctions made within music consumption. These make it hard for the algorithm to maintain trust in the predictions they will offer. For instance, during these research interviews listeners often separated the music that they ‘like’ from the music they ‘love’. The latter is entangled with memories and emotional affects – giving loved music a greater power to shape everyday lived experience (see Bennett, 2016). Recommendation algorithms might be trusted and intimate, but to produce long-lasting musical engagements the mediation of affective dimensions of social life become essential (as discussed by Nowak, 2016c). Through such nuances the algorithm can adapt the flow of everyday listening, as Pamela explains,

Most of the time I would listen with headphones, and I have to choose what to listen to, because now that I am in love with someone again, if something romantic comes up I start singing. (Pamela, 28, Publicist, Mexico City)

In this sense, the lived experiences of the music and the attachments created as a result impact on how algorithmic systems are used. In this case Pamela takes control of the music listening, anticipating that the algorithm might not be able to adapt the recommendations to a changing situation, the balance between passive and active, control and delegation, changes through these mediations. This is suggestive of the limits of the trust placed in these streaming algorithmic selections and how control is sometimes reasserted by listeners where they anticipate algorithmic knowing will fall short.

To work in everyday life as intimate companions algorithms require some element of trust. This is a kind of trust in their predictions. In the everyday routines of algorithmic management, it is trust in prediction that accumulates over time. This is based partly on the way that the actions of the listener are understood to shape future recommendations and their constraints. In addition, this trust is built in tacit engagements with these devices. The algorithm is expected to produce an experience that the listener is not always able to express or even recognise. Of course, this sense of what the appropriate next song to play is, also forged by the song that plays as well as a sense of what fits with the moment and with the listener’s sense of their own taste. The algorithm remains performative even though the listener seeks to actively build trust with these systems. The specific affordances that emerge in the encounter of each platform become important for allowing that sense of control, but also for building up trust. Through those recommendations and the development of a form of trust, for these listeners algorithms do not seem to relegate the importance of the social experience and the meanings attached to music. The role of emotions, intimacy and memories remains but is in part opened to algorithmic intervention.

Objectives in the everyday routines of algorithm management

As well as trust, algorithms as a type of technology of the self rely on producing a sense of control and delegation. Control and delegation require some sense of a desired outcome. As we have explored, this relies on forms of mutual domestication, trust and constrained agency, that are mediated by affordances and affects allowing the listener strategic decisions in everyday life. As we have also suggested, that building of trust in the algorithm is in part a product of the way that the algorithm is guided by the listener while determining their consumption; the listener uses their own imaginaries to make sense of the work of the recommendation system. Central to this type of algorithmic management are the listener’s objectives when confronted with the algorithmic structures of music streaming. Recommendation algorithms are indeed complex assemblages allowing strategic operations. To elaborate, this section explores some of those strategic operations for discovering music and how this is aligned with gaining knowledge and shaping different dimensions of the listener’s identity. The relations between algorithms and culture are far-reaching. As it has already been established, it seems now impossible to understand cultural taste and consumption outside of the context of algorithmic systems. Algorithms are a defining presence in cultural formations and experiences.

While listeners may have specific goals, their interactions with algorithms are also shaped by habitual practices and affective responses. This study recognises that musical experience emerges from a blend of rational objectives and experiential engagement. For instance, listeners might set goals to discover new music, but their choices are also influenced by mood, context, and social interactions. This dual approach acknowledges the complexity of user behaviour in algorithmic environments discussed in recent scholarship. For instance, Gal and Elkin-Koren (2017) discuss how algorithms collect and analyse consumer preferences, highlighting that consumer behaviour is influenced by both rational decision-making and affective responses. Therefore, when personalised by AI, consumer experiences are shaped by rational objectives and emotional responses (Gera and Kumar, 2021). In music consumption and streaming, it has been argued that when consuming music, users value the ability to interact with their platforms, personalise their experience, and involve goal setting and emotions (Lüders, 2021). The way that listeners engage with objectives in their listening practices feeds into the algorithmic context in which they are consuming music that is particularly important. The objectives themselves form in interaction with these recommendation systems. These objectives can often be implicit and potentially highly routinised in their form. Such is the case with Elena, who found out that the music she was listening to while working or relaxing fits into a genre that has been labelled trip-hop, through the algorithm she wanted to try to follow that genre further:

Yes, I find a lot of things [in Spotify]. Anything. I really like it to suggest things to me. When I discovered the radio function, I expanded my music catalogue a lot. It identifies that I listen to a lot of nineties music, particularly Portishead, Hooverphonic and others. It keeps going through the same genre with more songs. (Elena, 30, Journalist, State of Mexico, Mexico)

Here a filtered music genre has worked its way into the individual’s routines, resulting in them setting an objective to explore that particular category of music further. During the interview, Elena defines herself as a Trip-hop fan, the objective of exploring the genre is made through the industrial definitions of the genre and algorithmic sorting, constraining what she can get. Of course, genre definitions can be very flexible in the flux of music listening. Some algorithms work more through song features (Prey, 2018), of which the listener is not necessarily conscious (Gran et al., 2020). However, they still work in terms of mood and context (Siles et al., 2019b) and provide a sense of control to the listener. This can create complexity for genre-focused listening.

In many cases, the definition of a genre is not the lead concern for a listener. Still, the objective for the listener is focused more directly upon a sense of being updated. As is the case for María,

[in YouTube] I used to play One Direction, and it suggests songs from other artists, sometimes artists that I would know already, but I would follow the links because I wanted to hear the new stuff. That is how I started getting more knowledge. At the end of secondary school, I started following a list called ‘Exitos [Hits] Mexico’. I listened to all the new songs that I would like later. To give an example, I did not like reggaeton, the early one, but now I like the new one. I used to like mostly pop, particularly in Spanish, even a bit of the Country music that I would find there. (María, 18, Student, Cholula, Mexico)

The listener uses the algorithm to shape their taste through exploration, with the algorithm assisting in their aim for diversity. The logic of discovery is established with the aim of a wider consumption range across genre boundaries. As in the last section, there is a notion of control and delegation, but the clear objective is to expand music consumption and open the limits of discovery. Above, María explained that she was not a paid member of Spotify, which limits some of her actions, so sometimes she feels more control in other apps, and sometimes she wants less control and goes for defined playlists or ‘recommendations lists’. The role of the algorithm in these accounts offers listeners different forms of subjectivities adopted through music without having to fabricate them themselves (Drott, 2018), but also constrains it to the platform’s design (Hanrahan, 2018).

The trust placed in the algorithm, as explained in the last section, becomes essential in achieving the listener’s objectives. The algorithm is trusted to respond to the listener’s aims and to offer guidance on what is perceived as a journey or a chain:

I think YouTube helped me a lot, because I could listen to a song, let’s say, by Caifanes, then when it’s over, some suggestions start and I would decide what sounds good. I would start a chain of music in which I would start somewhere with something and would end up with something totally different. Then I have the chance to discover some new music. (Ernesto, 19, Student, Cholula)

The move from the current song to the next is perceived to be part of an established and ongoing chain of songs. In this type of vision, the algorithm connects the link into a chain. Ernesto explains that the journey is not led by charts or genres but by the simple objective of getting new music and not getting bored. Here, the algorithm is understood in terms of the ability to find something new and enliven cultural consumption patterns. That does not mean that listening is approached without any limits, rather listeners’ awareness of the work of recommendation systems is at play and in response to this they develop their own strategies to shape the algorithm to their objectives. A key technique here is to start ‘the journey’ or begin the ‘chain’ of songs with a specific band that the listener anticipates will elicit a particular and fitting response from the algorithm.

However, algorithms are associated with broader notions of popularity or genre elsewhere. This ‘journey’ is in part formed by a sense of wider scenes and movements, and the way the individual attaches or distances themselves from it. The listener can sometimes actively alter the direction of those recommendations to stand apart from that journey. Some participants use recommendation systems to stand against a trend they dislike, relying on the algorithm:

YouTube sends you genres that it guesses you would enjoy according to your search history. That is where I found Carla Bruno and similar. I have been discovering more music from that period; I am in love with it and I want more. I dislike what people listen to today. (Luz, 23, Student, Cuautla)

This listener seeks to look backwards rather than simply follow the perceived forward motion of the algorithmic selection. The algorithm is understood to be uncovering the ‘new’ and so action is taken here when the older music is the objective. Rather, the listener is actively engaging in algorithmic prediction and taking action to tweak or adjust the outcomes of that prediction. Alongside this, in another instance, it was noted that the algorithm systems themselves facilitate a break with established listening practices and to avoid getting stuck in repetition. For others then there is also a sort of understanding of the algorithm as providing a means of disruption. But this narrative seems to assume the algorithm as a technological force outside others’ consumption, Carla Bruno appears in relation to others that might not be in Luz’s direct context outside urban Mexico. The features of streaming themselves are understood to offer the means by which listeners can achieve the objective of not getting bored and stuck in their music:

What I mostly do now in Spotify is following suggested artists or similar artists to make fresh trails and avoid getting stuck in the same music. Previously that range of possibilities was not possible, so anything that you would get was fantastic. (Chispa, 40, Arts Teacher, Puebla)

Chispa presents an interesting contrast from the rest of the participants in this piece; they actively participated in forms of music consumption outside digital platforms. Their accounts are framed by their radio consumption, records and more genre-oriented labels. They recognise the limitations of those modes, and they use recommendation systems as a way out of those limitations or to serve those limitations. In this case, the algorithm is trusted not to lead the listener to become trapped in a particular type of listening patterns.

Finding something new can also lead to long-term attachments – with the algorithm intervening in tastes and preferences over time. Nonetheless, several participants expressed that the actions taken after they receive recommendations are an important aspect of their objectives and the role the algorithm is playing:

What we do here is sometimes we buy digital music. I would usually search on YouTube for a band that I would feel like listening to. That leads me to the lists available there, I always do this while working, so when I listen to something interesting I come back to the browser to check who is it, and then I copy the name of the band or the album or search it in there to find the complete works beyond the ‘flashy’ song. (Raúl, 39, Graphic Designer, Mexico City)

The objective is to pursue, in this case, deciding what record to buy. The algorithmic listening leads to other connections and consumption of the music and the artist’s other works. For this action to be completed, the recommendation does not act by itself, it results from the flow of attention and the convergent nature of devices and platforms. As other participants explain, the discovery process is commonly related to routine actions that are convenient to complete.

Another interviewee explains this balance and how the algorithm leads them to explore new musical experiences. The individual track can open an album or an artist, causing a switch in the listener’s objectives. Pedro is a fan of progressive rock; during the interview, he explains that he considers that genre as based on oeuvres, focusing on the album rather than individual tracks. Listening to what he gets recommended in a streaming platform becomes a matter of making smarter decisions to achieve his goal to discover bands for expanding his record collection:

I held back [from Spotify] for a while until I understood I could listen to the albums and then decide if I wanted to listen or not. Now I am a big Spotify fan, this feature of the ‘Weekly discovery’ is amazing because it is so personalised. It has made me discover lots of music, and then when a song attracts my attention, I can save the whole album. When I listen to the album, I can decide if I want to buy it or not. I am still very attached to the ‘physical album’, but I am much more selective. It is also a matter of resources. (Pedro, 48, Chemical Engineer, Puebla)

Here, we see how different materialities of consumption can interact, combining streaming and physical formats. That list of recommendations offers only individual songs, but Pedro identifies how sometimes the way those ‘randomised’ songs work together is ‘exciting and interesting to discover new connections’. A single song can actively open up different paths or chains. For Pedro, it is the state of flow between attention states that allows him to discover music while he works or commutes in public transport: ‘when I am reading or working and the song distracts me, it means that is a good song’. The streaming platform through its algorithmic recommendation plays some tracks, but then Pedro intervenes when a song catches his attention; that artist then becomes a focal point for retraining the algorithm. Setting objectives within a limited range of options is a key feature of those listening experiences. Those actions can sometimes be aimed not only at modulating situations or emotions but also at managing and performing identities and tastes.

Conclusion

Across different platforms, cultural consumption is defined by algorithmic recommendations. This article has argued that trust and objectives are crucial for understanding everyday engagements with algorithmic systems such as music streaming platforms. The listener is not passive when confronted with algorithmic recommendations. Rather, listeners build trust in their selections and intervene or let the algorithm align with established listening objectives. The listener does not only ‘cater’ (Lüders, 2021) for the algorithm to produce better recommendations; they also make algorithmic systems work to achieve goals or preferred outcomes. Users balance control and delegation, relying on algorithms for music consumption while retaining the ability to intervene. More specifically, it shows that the process of mutual ‘domestication’ (Siles et al., 2019a) of the algorithm allows these goals to be successful through learning to manage balanced states between attention and inattention, control and delegation, and new music and boundaries. Overall, the argument also notes the importance of considering literacy and contextual factors to develop such trust and objectives. It is important to note that almost 25 per cent of the Mexican population did not have Internet access at the time of the study (Rodriguez, 2023), which may limit the generalisability of the findings.

The adaptation of the concept of ‘technologies of the self’ to recommendation systems, as suggested by Karakayali et al. (2017), alongside the concept of Mutual Domestication (Siles et al., 2019a), proves useful in understanding how individuals enter into an iterative set of relations with algorithmic outcomes and their platforms. The participants interviewed use those algorithms to find the next track, delegating that task while often engaging in something else. The algorithm then becomes a means to manage the vast archives of culture found within such streaming services. As this article has explored, this requires both a sense of trust and a sense of ideal outcomes, both of which feed back into the relations listeners have with these algorithmic systems. The objectives vary and include avoiding listening to the same music all the time, avoiding music that is too familiar or too unfamiliar, which could be distracting, or aiming to achieve affective states that allow them to manage their own experience in the world. The everyday routines of algorithm management vary in line with the trust that is built and also the way that a notion of a future outcome becomes part of how the algorithm is dealt with in the present. Listeners also use algorithms to manage themselves and the way they interact with the world, as when trying to expand knowledge on specific genres or bands, or to broaden their listening experiences.

This brings us back to a discussion of the importance of music in everyday life and its relative value for individuals (as discussed in Hanrahan, 2018; Hesmondhalgh, 2013; Marshall, 2019). Those individual values still remain, but they are now forged in relation to both algorithmic systems. Within these relations, the way in which those algorithmic systems are understood and responded to is a part of how personalisation and recommendation occur. Music streaming is described in quite intimate terms. In addition, identities may be assembled from vast archives of content and cultural forms (Cheney-Lippold, 2011a, 2011b; Fernández Vicente, 2020; Prey, 2018). In this sense, music streaming becomes a source of ‘distributed subjectivities’ (Kassabian, 2013). In the logic of the music streaming site, this translates into a focus on the next song. We suggest that a place to start is to understand the role of trust and objective setting in everyday and routine engagements with algorithms that occur in cultural consumption. Future research could expand on detailing the role of other recommendation systems, as well as socio-economic factors in building such trust and objectives, to expand the possibilities of understanding agency on algorithmic mediated cultural consumption.