Putting Trust to the Test: Making Sense of Human–Machine Interactions on TikTok

Abstract

People’s interaction with online content is increasingly facilitated by intelligent user interfaces and artificial agents. In this article, I explore this shift by drawing on ethnographic fieldwork with users of the TikTok app. More specifically, I write on their interactions with the TikTok algorithm as a form of human–machine interaction and through the lens of trust. Along concrete ethnographic data, this article lays out the multifaceted process in which participants negotiated trust in the TikTok algorithm as an interaction partner in their everyday pursuits for relaxation and entertainment. Understanding trust as something deeply relational, mediating the position that one takes to another, this article outlines the constitutive embodied and affective dimensions of trust. It shows how participants dealt with feelings of their trust in the TikTok algorithm being put to the test, as well as how they negotiated their distance and closeness to it accordingly. By doing so, this article will demonstrate how trust functions a key mediator of meaningful human–machine interaction – shaping not just meaningful outcomes but also meaningful processes of interaction. From this angle, this article closes with an argument for research on the foundational role of trust in human–machine interaction, specifically in ways that look beyond the cognitive processes of judging trust and broadening the scope towards the material and cultural contexts in which people trust others.

Introduction

People’s online experiences are increasingly facilitated by machines – algorithmic systems, intelligent user interfaces, artificial agents. Interaction with such machines are characterised by ambivalences in which people oscillate between concerns about machine surveillance, attempts to resist against machine actions, but also co-operation with machine processes (Siles, 2023). The strain of such unresolved ambivalence and limited ability to control one’s online experiences has given rise to what some call a “culture of resignation” (Draper & Turow, 2019) and sense of “mild paranoia” (Ruckenstein, 2023). In this article, I further unpack how these ambivalences materialise – how people negotiate the limits of their agency in relation to machines and the continuous surveillance of their lives through them.

In particular, I address the following question: How are the durability and stability of human–machine relationships accomplished in a cultural context of ambivalence? I follow the argument of Siles, Valerio-Alfaro and Meléndez-Moran (2022) who note that the temporal component of human–machine interactions is understudied. Yet, it is precisely by looking at the temporal durability of tensions in human–machine configurations that we can better understand how power is at play, and how people negotiate it in situ as they enact algorithms as parts of their lifeworlds (Siles, 2023). To formulate an answer to this question, then, I juxtapose ethnographic fieldwork on TikTok with a conceptual notion of trust. I understand trust as a key mediator of the proximity in which humans position themselves in relation to others, like machines, and the consequences of their actions (Baier, 1986).

I draw on TikTok because personalised social media platforms like it embody the tensions touched on above. TikTok exemplifies a shift towards content-centric platform design (Kaye et al., 2022; Su, 2023). Once people open TikTok, they are placed on the “For You” page, a personalised content feed. There, people engage with what is referred to as “the TikTok algorithm” – an imagined singular machine entity. The experience of using TikTok, at its core, thus is a form of human–machine interaction with “the algorithm” guiding people through the seemingly endless amount of content on the platform (see also Schellewald, 2023). As this interaction partner, the TikTok algorithm has been found to be a meaningful companion for self-identity formation (Bhandari & Bimo, 2022) and pursuits of pleasure and relaxation (Şot, 2023). Yet, it has also been described as aggressive (Siles & Meléndez-Moran, 2021), at times evoking eerie feelings in the perceived intimate knowledge it holds about individuals (Kanthawala et al., 2023).

Understanding how people configure meaningful human–machine relationships (Suchman, 2007) despite such ambivalence and tension is the aim of this article. The next section opens with a discussion of this theoretical background on the process of trust and its application in a context of digital media. Then, a brief discussion of the research method is followed by the discussion of the findings. These are presented in a way shaped by the literature on trust, on the one side. On the other, they will discuss how trust around the TikTok algorithm is established, put to the test and maintained. Closing this article, I will reflect on the importance of trust in future research on human–machine interaction.

Theoretical Background

Outlining Trust

In this article, I follow a definition of trust that sees it as something deeply relational and as socially mediating (Levi et al., 1998; Mayer et al., 1995; Warren, 1999). In particular, I follow Baier (1986) and want to begin by echoing a quote that she opens her foundational essay on trust with. Drawing from philosopher Sissel Bok, Baier (1986, p. 231) quotes: “Whatever matters to human beings, trust is the atmosphere in which it thrives.” We require a certain sense of trust, put simply, so that we can meaningfully exist alongside others. Baier’s work developed such an argument by outlining trust as “accepted vulnerability to another’s possible but not expected ill will (or lack of good will) toward one” (Baier, 1986, p. 235).

Baier’s perspective is interesting here in how it allows us to shift our moral evaluation of machines from a binary of good/bad towards an inherently situational and dynamic process of negotiation (Amoore, 2020; Markham, 2020; Siles, 2023). I am speaking, in other words, about the process in which people negotiate accepted vulnerabilities they face when engaging technologies, such as algorithms or artificially intelligent agents – accomplishing through that negotiating engagement under conditions of doubt and vulnerability.

Scheman (2020) underlines the importance of Baier’s work here in how it opens a view on trust related to positionality. As she writes, since Baier (1986), “trust has been generally understood as deeply relational . . . a more or less critically reflective placing of oneself in a relationship of vulnerability to [others]” (Scheman, 2020, p. 28). My interest in trust as a concept stems from this analytical angle. It equips us with a tool to grasp that which might be problematic about the relationship between humans and machines, yet also the meaningfulness that can result from their interaction – both mediated on the basis of reflexively (dis)trusting the other.

In this article, I hence use trust to understand four things: (1) how humans and machines stand in relation to one another; (2) how trust is necessary for the meaningful interaction between the two; (3) how moments of distrust are dealt with; (4) how a lack of trust might result in the collapse of interaction. I am specifically interested in the acts of judging trust and trustworthiness. I draw on O’Neill (2018) who offers us a useful framework. According to her argument, trustworthiness is judged along dimensions of competence, reliability and honesty. We judge the trustworthiness of another in relation to the honesty that pertains the claims they make. And we negotiate our trust in them based on the competence and reliability in which they deliver on their claims.

Trust in Digital Media

Trust defines not just our interactions with one another, but also the media contents (Arguedas et al., 2024), platforms (Wong et al., 2023) and machines (Ávila Torres & Beer, 2025) we engage with in our daily lives. For example, in the field of human-robot interaction, scholars have encountered similar sets of components along which humans judge the trustworthiness of machines as they do in other humans (compare Ullman & Malle, 2018 to O’Neill, 2018). In this article, what is of my primary interest, however, are not these dimensions themselves but the consequences of judging trustworthiness.

Siles (2023), for example, discusses how people place greater trust in the recommendations for movies that were made by people than those coming from an algorithmic system – which was perceived as lacking embodied and affective experience crucial to make cultural decisions of taste (Siles, 2023, p. 65). In this cultural context, the positionality of algorithmic systems as machines gives rise to a particular dynamic of trustworthiness. Through the limited trust people have in algorithmic systems to make taste-based decisions, they position themselves at a certain critical distance from their outputs which mediates the extent of algorithmic consequences on their lives.

We can see how trust mediates positionality and negotiates distance similarly in relation to surveillant practices upon which such systems rely to carry out their operations. Recent studies like that of Kappeler et al. (2023) underline the negative outcomes that distrust in digital platforms can have on online self-inhibition. They found that participants who expressed a greater sense of distrust in online platforms experienced chilling effects undermining their ability to meaningfully express themselves online and participate in digital spaces – for example, refraining from searching for specific topics due to a lack of trust in how such information will be handled (Kappeler et al., 2023, p. 7).

These empirical findings are contextualised by broader cultural diagnoses. Draper and Turow (2019), for example, frame it as a “culture of resignation” in which people feel increasingly powerless to control their experiences online. That is because, as Schwarz (2021) argues, people’s ability to act in such contexts is one of taking detours and finding loopholes. Siles, Gomez-Cruz and Ricarte (2022) thus describe a “popular culture of algorithms” marked by a state of constant flux and ambivalence. Ruckenstein (2023) captures this more broadly by arguing that the particular structure of feeling that defines this kind of culture is one of “mild paranoia”, emerging from to proliferation of machine tracking and agentic capacities.

In this context, trust, as defined earlier, emerges as a key factor. It captures how people (fail to) negotiate potential vulnerabilities and thus navigate states of ambivalence and feelings of “mild paranoia”. A fall out of trust does here not just exercise potential chilling effects on self-making and meaning-making practices. More so, the technical systems that people navigate rely on trusted engagement. Cen et al. (2023), writing in an engineering context, underline this issue. They outline how distrust in algorithmic systems leads people to engage in strategic behaviours, like concealing parts of their identities (Kappeler et al., 2023). In return, the machine is itself is confronted with challenges in its ability to deliver on promised experiences – further undermining its trustworthiness (Cen et al., 2023, p. 22).

How exactly do people, then, judge the trustworthiness of media technologies? In this article, I follow the work of Steedman et al. (2020). They have developed the influential concept of “complex ecologies of trust”. They underline that trust should not be understood in binary terms, those of a person either having trust in a platform or not, for example. Instead, their research showcases how trust is multifaceted. Along different layers and process, people negotiate a sense of trust in digital technologies. Rather than being a fixed disposition, they argue that trust is a more fluid matter (see also Siles, 2023).

Wong et al. (2023) highlight in this context, for example, that the trust people put into artificially intelligent systems is of conditional nature. Drawing on focus group discussions, they showcase how trust in data-driven systems is tied to expectations of good practices, for example in relation to transparency, and how mistrust may arise when such expectations are not met or violated (see also Lupton & Michael, 2017). Furthermore, they underline the importance of context dependencies, reminding us “how the purpose, i.e. to entertain or to inform, alters [people’s] determination of good practice.” (Wong et al., 2023, p. 10).

Trust as a Process

In this article, my approach to trust is aimed at empirically uncovering mechanisms of judging trust in human–machine interactions. Juxtaposing empirical data on TikTok with an existing notions of trust (Baier, 1986) and trustworthiness (O’Neill, 2018), I expand understandings of how trust functions practically to enable the stability of such interaction over time.

For example, recent work by Ávila Torres and Beer (2025) on algorithmic recommender systems underlines the importance of context dependencies (Wong et al., 2023) in shaping the evaluation of trust. They showcase how people judge trust of algorithms in relation to their ability to deliver on expected outcomes and ability to act on provided inputs – creating a sense of trust that opening up to the algorithm today will increase its quality as a companion in the future (Ávila Torres & Beer, 2025, p. 12). This connects to work by Pink et al. (2018), which I follow here. They underline the constitutive role that trust places in enabling meaningful experience with data and machines over time.

In particular, Pink et al. (2018, p. 3) write that “humans trust when we feel confident enough that any improvisatory action is sufficiently cushioned by the familiarity of process or place.” Moreover, they go on to argue that this confidence does not result from a cognitive evaluation of the trustworthiness of the system. Instead, they say that it is based in a practical familiarity with the technological system (Pink et al., 2018, p. 7). It relies on what Giddens (1984) has theorised as a form of practical consciousness – the practical ability to know how to carry on in everyday life, without the need to be consciously aware about such practice.

People’s interaction with algorithms more generally resonate with notions of practical ability. Bucher (2017) has foundationally outlined how people assume a position in relation to algorithms by sensing their presence – for example, through recommended content and the emotional impact these recommendations carry for individuals. From that felt presence, people imagine their interactive possibilities which determine future interaction paths in the human–algorithm relationship (Bucher, 2017, p. 42).

Put simply, people primarily make sense of algorithms in a mode of “learning by doing” (Cotter & Reisdorf, 2020). This gives rise to a relationship that is fluidly evolving in every interaction (Siles, Valerio-Alfaro, & Meléndez-Moran, 2022) and marked by various emotional states of both negative yet also positive form (Ruckenstein & Granroth, 2020). Scholars have here in particular argued that vital moments in the relationship are ones of irritation – both in providing opportunities for people to learn from their unexpected experiences, yet also potentially undermining senses of agency (Swart, 2021).

As we can see, then, trust emerges as a key component of human–algorithm relationships. It is in confrontations with algorithms and their power, as I will outline in this article, that people are forced to reflect on the ambivalent foundation that their interaction with these machines rests on. I articulate trust, then, as an affective-reflexive component through which people negotiate how they want to and/or can stand in relation to machines that they interact with habitually in their daily lives. It is within this reflective process that algorithms are enacted (Siles, 2023) as practically manageable and durable configurations, ones in which persistent structural tensions are situationally and momentarily resolved.

Research Method

This article draws on data from a larger digital ethnography (Pink et al., 2016) of TikTok consumption (Schellewald, 2024). Fieldwork was conducted in 2020 and 2021. First, over a period of 6 months, I immersed myself in the context of the TikTok platform through daily usage and walkthroughs of the TikTok app (Light et al., 2018). Second, over the course of 1 year, I worked with a group of 30 young adults based in the United Kingdom. Participants were engaged with through a variety of methods and techniques. Primarily a series of semi-structured interviews were conducted at 3-month intervals. In addition, participant observations, media mapping techniques and informal conversations were used to gain a deeper understanding for participants use of TikTok.

When fieldwork was conducted, TikTok was primarily seen as a “kids app”. However, young adults already constituted the core of the platform’s audience in the United Kingdom (Loose et al., 2020). Resultingly, the research focussed on this group to engage with typical TikTok users in the cultural context. Participants were recruited through a promoted tweet targeting Twitter (now X) users aged 18 to 24 in the Greater London area who showed an interest in TikTok. Additional participants were gathered using snowball sampling. London was chosen as an area for sampling due its nature as a global city, providing access to people from a diversity of cultural backgrounds. The participants represented a group of mostly well-educated and middle-class individuals, with most either having completed or being currently enrolled in higher education programmes. 22 identified as female, 7 as male and 1 as other. 19 were of White British and 2 of White American backgrounds, the remaining participants were a mix of Black British (4), British Indian (1), Indian (1), Middle Eastern (2) and South East Asian individuals (1).

The collected data were analysed thematically in a threefold process of reflexive ethnographic writing (Abu-Lughod, 1991). The first two stages were comprised of iterative coding of the data. The first coding round was descriptive and coded the data in terms of types of uses, context of use, interaction partners related to TikTok, apps mentioned in relation to TikTok, the emotions experienced using TikTok, and so on. In the second round of coding, overarching themes were developed. A theme was broadly defined as a shared pattern of meaning, practice, or experience. Theme development generalised the data through the principle of “juxtaposition – contradistinction, comparison, sequentially, referentiality, resonance and other ways of patterning across multiple observations” (Dourish, 2014, p. 11).

The results of this thematic analysis were then used in the process of ethnographic writing, which was aimed at providing an intelligible account of participants’ experience (Geertz, 1988). For this purpose, themes were treated as discursive objects that bring “a kind of order and coherence onto the otherwise chaotic outlook of the empirical” (Ang, 1996, p. 65). I draw on this overall ethnography that was written (Schellewald, 2024) and have juxtaposed it with the literature on trust outlined earlier. By doing so, I accentuate the dimension of trust within the data and generalise broader principles of human–machine interaction that we can derive from the accounts of my participants’ experience of using TikTok in their daily lives.

Like any ethnographic work, the findings are limited in the sense of being positioned truths (Abu-Lughod, 1991). This article’s findings represent the experiences of typical middle-class individuals and experienced media consumers in the United Kingdom. I position the analytical relevance of this group following Miller’s (2024, p. 5) argument that looking at middle-class people provides an opportunity to capture a variety of both positive and negative experiences in anthropological descriptions of media consumption. Understanding trust as a process of negotiating vulnerabilities, engaging with young individuals for whom TikTok might appear relatively unproblematic to manage – due to societal privileges that other groups do not enjoy (Karizat et al., 2021), or algorithmic literacy that other demographic cohorts lack (Gran et al., 2020) – offers thus diverse material to ground a theorisation of trust as a mechanism to integrate algorithms in everyday lifeworlds (Siles, 2023).

Findings

The Relevance of Trust

My participants’ engagement with TikTok is exemplified in experiences like Tanja’s. At the time of the research, she was a university student in her early twenties. Next to her studies, she worked a part-time job and had occasional family obligations, like taking care of her younger sibling. In short, she lived a fairly busy life. Her daily schedule was comprised of small, fragmented tasks – an experience common to the socio-cultural context of the United Kingdom (Shove, 2003). In this setting, TikTok emerged to her as a convenient tool to take short “breather breaks”.

To Tanja, and other young adults I met, TikTok appealed as an app that was distinctly entertaining and affording escapist qualities in that sense (see also Kang & Lou, 2022; Siles & Meléndez-Moran, 2021). These distinct entertainment qualities emerged specifically from the fact that the content engaged with was from “ordinary strangers”. This made it feel easily relatable, yet also distanced at the same time, allowing for an engagement freed of social obligations and pressures. In the lifeworld of my participants, TikTok was primarily engaged with by oneself and for oneself, as a means to momentarily take a step back from everyday social life and enter a pleasurable “me space” (see Schellewald, 2023).

The interactive dynamic of stepping into this “me space” is captured by what Siles and Valerio-Alfaro (2025) theorise as “uncommitted attention”. They define it as an interactive mode in which people engage with immersive flows of content that require active participation in the form of scrolling and responding to videos placed on feeds by an algorithmic system. And it is within this context of an escapist desire to relax, unwind and disconnect that algorithmic trust emerges as a key factor.

Trust was a factor not in the evaluation of the TikTok algorithm’s cultural expertise (Siles, 2023, p. 65) but, instead, its ability to create “interactive pathways” (McKelvey & Hunt, 2019) that evoke desired feelings of relaxation and escapism – sparking in that process tensions through forms of intimate surveillance (Ruckenstein & Granroth, 2020), attention capture (Seaver, 2022), or epistemic uncertainty about one’s sense of self (Kant, 2020). In what follows, I show how trust materialised as a reflection of my participants’ confidence in the TikTok algorithm’s ability (Ávila Torres & Beer, 2025) and familiarity with its processes and dynamics (Pink et al., 2018) in this context.

Establishing Trust

Joyce was one of the recent university graduates living in London that I met during fieldwork. She had just started her first job and used TikTok a lot to unwind after work. For her, like other participants, TikTok was just a convenient tool to engage with. At the same time, the cultural context of young adult TikTok use back then was still clouded by an image of TikTok being for kids to post dance videos and “stupid” memes. As such, participants like Joyce – having been able to engage with other kinds of content on her “For You” page – did at times turn into advocates. As she told me,

When I do get my friends to join TikTok, they sometimes be like, ‘oh . . . I don’t really like it. I’m not getting any good videos’. And I always say, ‘just like the ones that you like and press not interested on the ones you don’t because you will end up having the perfect For You page.’ (Joyce)

What we can see from Joyce’s account is how trust emerges as a necessary element to stabilise interactions in the present for a reward of personalised experiences in the future (Ávila Torres & Beer, 2025). People need to trust the machine process, so to speak. They need to participate in the construction of their “For You” page for the algorithm to deliver content that, at some point, will feel adjusted to personal preferences. And this process, as the hesitation of Joyce’s friends shows, is a matter of trust. It showcases a vulnerability and threshold that needs to be crossed affectively – an observation that resonates with other research which found that the initial process of “training” the TikTok algorithm is accompanied by feelings of annoyance and insecurity (Siles, Valerio-Alfaro, & Meléndez-Moran, 2022, p. 8).

Resultingly, it is crucial for us to articulate what this form of trust entails to understand how a meaningful form of human–machine interaction is made possible and becomes durable. Yet, it is also crucial to understand how power plays out in the search of a meaningful entertainment experiences online where people like Joyce and her friends are increasingly placed in a position of dependency to algorithmic systems. Guided by the literature outlined earlier, specifically Baier’s (1986) notion of trust, we can formulate how people position themselves in this relation by looking at two vulnerabilities.

The first vulnerability is that of boredom or wasting one’s time. Participants, such as Adna, often told me: “if I’m on TikTok, I don’t want to see anything annoying or unfunny”. This might not be a surprising statement. Nobody is likely to spend their time with things that they find annoying. However, the gravity of this vulnerability is underlined by the sense of “time pressure” (Wajcman, 2015) that is common to the experience of individuals in societies like the United Kingdom’s. Adna was also a recent university graduate and struggled finding employment at first. The job she eventually found in the education sector was not necessarily providing her with much fulfilment. As such, coming home from work, she had high expectations for her scarce leisure time. Here, trust is thus key to enable the continued engagement with TikTok and specifically its algorithm.

Committing to interacting with it, people engage in a risk that, potentially, the algorithm will show them content that is just boring and uninteresting. As a service, TikTok (2022) promises to “make your day” – a claim that, meta-communicatively, the TikTok algorithm, tasked with constructing the “For You”, risks not living up to as primary interaction partner. The TikTok algorithm does, of course, not directly communicate with people, and neither do they with it. Nonetheless, my participants imagined algorithms as interactable entities, and their navigation of content feeds as a quasi-dialogical interaction (see also Siles, Valerio-Alfaro, & Meléndez-Moran, 2022, p. 9). From such an angle, people’s experiences of algorithms are grounded in a judgment of truthfulness (O’Neill, 2018) – a sense of honesty that the algorithm lives up to the claims that were made meta-communicatively about its performance.

Put differently, for people like my participants the TikTok algorithm emerged as that element of the TikTok app which is responsible for managing the vulnerability of wasted time by providing personalised content recommendations. The central reliance on this system, however, gives rise to a second vulnerability around which trust needs to be dealt with, that of data privacy – information being leaked, used for surveillance purposes, or negatively affecting information choice and mental health issues.

Some participants, like Joyce, using the app in their scarce leisure time, simply said: “I think those are criticism of the app and probably a lot of them are correct. But I do actually just ignore them because I’m having way too much fun on TikTok at the moment”. In relation to this risk, participants interacted with the TikTok algorithm while being aware of two things. First, that the TikTok algorithm collects information about them. And second, that such information might be used in ways they did not anticipate and potentially leaked in some possible scenarios.

Participants managed this vulnerability of data privacy in forms of “strategic ignorance” (McGoey, 2012), consciously not thinking more deeply about certain problematic aspects. This was to assume a position from which they could lean back and go with the machine process for its entertainment benefits without constantly being sceptical about underlying commercial objectives and associated data practices. And, as noted above, the desire to take this position emerged specifically from the contextual constraints of their daily lives – being busy and at times not necessarily fulfilled, providing a need to finding escapes and “me spaces” like the TikTok algorithm opens them.

Linked to the literature on complex ecologies of trust, my participants established their initial trust along different layers. Some did so on a more macro-level. Benjamin is one of these people. He had just completed an undergraduate degree in Psychology and was about to start a Master’s. When we spoke about TikTok’s data practices, he said to me that “I have enough awareness to know that if I’m going onto an app where it has a ‘For You’ page that I’m going to expect that it tracks certain things”. For people like him, there was no need to establish trust in the TikTok algorithm specifically. Rather, it was positioned as part of a larger system of data collection practices, like those of personalisation (Kant, 2020), for which Benjmain had both an awareness and prior experience.

Such ways of positioning the self in relation to the TikTok algorithm were common. Not all displayed similar depths of algorithm awareness or technical understanding. Yet, participants all felt that being observed by the TikTok algorithm was mostly non-threatening – echoing related research (see Lupton & Michael, 2017). We can see this on the most concrete level in the case of Bea. She was a freelance journalist, having recently moved from Canada to the United Kingdom, being in her mid-twenties. She mentioned that she sometimes finds the TikTok algorithm quite specific – for example, how it manages to keep showing her content from Canada even though she has been living in London for a while already. Yet, when I asked her about privacy concerns, she told me the following:

Sometimes I’m like: ‘how much can they really learn about me?’ They know that I like dogs, they know where I live, they know I like cooking videos . . . I don’t feel threatened by that, you know, in the grand scheme of things. I think that I’m more afraid of Google than of TikTok. (Bea).

Put simply, by locating the TikTok algorithm within complex ecologies of media trust (Steedman et al., 2020), people negotiate their own positionality and sense of comfort. Some participants, like Benjamin, comfortably navigated around all types of personalisation technologies, regardless of platform and context. On the other hand, people like Bea more carefully negotiated their specific position in relation to the TikTok algorithm, contrasting such from other technologies like Google Search. In either circumstance, however, trust functioned as relational mediator (Baier, 1986) that defines the proximity at which human and machine meet and can come to interact with.

In my research, I observed some differences mediated by educational background that shape the depth and detail at which these negotiations unfolded. However, in line with previous scholarship (Cotter & Reisdorf, 2020; Swart, 2021), I mainly observed that, like in cases of Benjamin or Bea, it was prior, embodied experience with data-driven technologies that substantially shaped how a position in relation to the TikTok algorithm was assumed.

Testing Trust

Trust is not a binary and fixed quality (Steedman et al., 2020). The findings of my fieldwork speak towards this. I describe these trust dynamics as a form of tests – transgression from expected experiences, putting trust in the machine to the test. Furthermore, these trust tests emerged in concrete situations of use, rather than being prompted by larger events, like reports about corporate data leakage (Ruckenstein, 2023). In the accounts of my participants, two types of trust tests emerged that we could structure along the vulnerabilities I have outlined earlier.

The first trust test is one in which the TikTok algorithm is perceived to show too many unspecific videos. When recommending too much boring content and thus wasting people’s time, trust is put to the test – and people start to questions the machine’s competence. Beyond competence, however, people’s trust in the TikTok algorithm was also tested on the dimension of reliability more concretely – lowering the confidence and expectations for future rewards from the relationship (Ávila Torres & Beer, 2025).

Some participants, like Hannes, mentioned that when they noticed their content becoming very unspecific, they made use of the “Not interested” feature. They drew on the platform-provided means to communicate with the TikTok algorithm. Working as a journalist, Hannes was navigating TikTok for personal leisure but also used it to find relevant content for his work. Moreover, though not uncommon, Hannes mostly used the app occasionally and could by no means be described as an outspoken “fan” of the app.

Considering this context, using the “Not interested feature” had the opposite of its intended effect. Namely, using the feature had, in his view, no effect on changing the quality of recommended content, further undermining his trust in the algorithm’s reliability: “No, I don’t find it does work. The continuous effort probably shows that it doesn’t at all [laughs]”. Hannes’ account showcases how expectancy violations can create a sense of undermined agency (Swart, 2021). He unsuccessfully communicated with the TikTok algorithm, telling it through the “Not interested” feature a clear message about disliked content that was not picked up by the machine.

The importance of trust here, however, is further underlined in stories like that of Gretta. Gretta was an accountant working in London, partly from home and partly from the office. She told me that sometimes, usually on a Monday, she finds the TikTok algorithm kind of forgot her preferences – an experience of instability other TikTok research too has observed (Siles, Valerio-Alfaro, & Meléndez-Moran, 2022, p. 11). Gretta’s explanation to this was that she doesn’t use the app on the weekend, as she mostly scrolls during her lunch breaks. In response, she said that she usually needs to give the algorithm some time. In other words, and in contrast to Hannes, her trust in the machine process enabled the algorithm to continue gather more data and adjust its content offering again (Cen et al., 2023) – underlining the importance of trust as a stabilising mechanisms that mends doubts in moments of transgression.

The second trust test is diametrically opposed to the first, emerging from moments where videos are too specific – an experience that other research has identified as key to the formation of algorithm awareness (Bucher, 2017). Seeing videos that felt too specific often created affective responses that made people stop scrolling and reflect. Specifically, they reflected on the second vulnerability, that of privacy related risks. The common scenario participants experienced here were videos that felt “scarily accurate”, touching on aspects of their personality they characterised as being very intimate and in many cases not consciously revealed in prior engagements (see also Kanthawala et al., 2023).

One example is the case of Jade, a politics undergraduate student in her early twenties. She was using TikTok quite frequently, every day and often in various occasions throughout a day. As this frequent use suggests, TikTok provided her with at least some form of meaningful experience. Yet, she talked to me about coming across videos every now and then that felt “too close to home”. When she sees these videos, she told me that: “I always cover my camera [laughs]”. What we see here is how this irritation has an affective and embodied dimension – the video evoking a sense of “mild paranoia” (Ruckenstein, 2023) in which Jade adjusted her material relation to her phone, which is perceived as embodying the TikTok algorithms surveillance capacities.

What is interesting about Jade’s story, however, is how she resolved this trust test. She did so by drawing on a meme around online surveillance, the so-called “FBI agent watching you”. This meme suggests that behind the machine image of the algorithm sits a person, such as an FBI agent, that is observing “you”. Although this meme is by no means accurate to the situation that Jade experienced – considering she is based in the United Kingdom, TikTok is owned by a Chinese company, and the FBI is a US law enforcement agency – the meme helped her resolve the trust test. She was able to generalise her particular experience as something common, as an experience that others likely go through as well. As she told me: “usually it’s just like a passing thought and I’m sure it’s just like . . . um . . . ads and you know, like the stuff you interacted with before”. Being normal in that sense, she restored a situational sense of trust that there are no material risks in engaging with the TikTok algorithm and, therefore, she can continue scrolling.

In sum, trust tests are a roadblock on the interactive pathway of the “For You” page. These tests disrupt the experience and interaction with the TikTok algorithm, bringing that interaction itself to the foreground. Creating a transgression, they force people to stop and make a decision, and are embodied in this sense. Trust tests require that people need to re-evaluate where they stand in relation to the machine, if they want to continue interacting with it from their current position, or, instead, move back and lower their exposure to the vulnerabilities of boredom and data privacy.

What these tests underline then, generally speaking, is the affective component of trust and judging trust – and the wider structure of feeling that these are part of. As Ruckenstein (2023) argues, what instances of irritation and anxiety reveal is an experience of unease that is defining of the cultural context of algorithms. And what we can see in accounts like that of Jade above is how the shared nature of these feelings of “mild paranoia” (Ruckenstein, 2023) do not cause a condition of impasse but instead can be mobilised as a means to make do with media technologies as practicable resources (Martin-Barbero, 1984).

Maintaining Trust

The previous sections discussed how trust was established and challenged. In particular, I outlined how trust was put to the test in moments of transgression. These trust tests were, for my participants, a constant part of their experience of consuming content on TikTok. What emerges thus as a final question is how, despite frequent challenges to their trust, did they manage to keep engaging with the TikTok algorithm? As I have discussed previously, in situ participants drew on tactics to resolve challenges of their trust, for example, on cultural resources like memes helping them to cope with feelings of anxiety. On a larger scale, however, forms of strategic ignorance (McGoey, 2012) were the main mechanism that maintained my participants’ conditional trust in the TikTok algorithm.

As noted earlier, the idea of strategic ignorance refers to the deliberate choice not to acquire information which might bear disadvantages to one’s interests (McGoey, 2012). It resonates with the account of my participants, for example expressions like that of Joyce, mentioned earlier, who told me that they are aware of the privacy concerns around TikTok but chose to not further reflect on them. In other words, my participants maintained trust in the TikTok algorithm by means of actively looking beyond their worries – be those related to privacy, but also competence, such as the case of Gretta, also mentioned earlier, who reassured herself that the algorithm will deliver eventually. In other words, trust was maintained by moving worries and concerns about vulnerabilities towards the back of the mind, so to speak.

The notion of ignorance, however, might be too strong a term for forms of human–machine interaction. And that appears to me for two reasons. First, in an entertainment context, trust is key to the experience. As Ang (1985, p. 19) has argued for the case of television viewing, entertainment has to appear as a form of simple and uncomplicated pleasure, as a “mere entertainment”, for us to be able to enjoy it. If one were to constantly question the basis of that pleasure, the whole point of its experience would be null. Trust is thus a necessity to experience entertainment as entertainment. This, second, then underlines some form of dependency on algorithmic systems today, and thus also forms of coerced or uncomfortable trust (Barassi, 2019).

A common them across conversations about interactions with the TikTok algorithm ended with comments like Gil, an undergraduate Biology student in his early twenties, made them: “I mean, that’s about as much as I can do”. What Gil was referring to here was practices such as consciously liking videos or quickly skipping past and not watching those he senses might not be of his interests – all to engage in a sort of limited dialogical interaction with the TikTok algorithm (see Siles, Valerio-Alfaro, & Meléndez-Moran, 2022). Talking to Gil revealed not so much a sense of resignation but almost acceptance for the limits of his ability, concentrated around finding creative tactics and workarounds to communicate with algorithms (Siles, Gomez-Cruz, & Ricarte, 2022). This was a common theme, particularly because like Gil, many had grown up navigating personalisation algorithms on sites like YouTube, on which they encountered similar experiences and struggles for more direct control over the content recommended to them.

This resonates, then, with the theme of “digital resignation” that Draper and Turow (2019) have outlined, speaking towards a normalisation of a sense of powerlessness as inhabitant of today’s digital media landscape. As such, for my participants, ignorance was by no means a form of ignorance of the problems themselves. Instead, it was a strategic mechanism utilised to make do in an imperfect media landscape that they don’t own or control – just like people always had to make do with such imperfect media resources (Martin-Barbero, 1984).

The pervasiveness of these feelings emerged across my fieldwork period in cases like that of Rhea. In her mid-twenties, she had been working in the film and tv industry and recently moved out of London due to a new job offer. At the time of our final conversation, Rhea had been using TikTok for well over a couple of years, since she picked it up quite early. During this time, TikTok was a constant part of her daily routines, scrolling at times for an hour or more on the app every day, helping her in various moments, such as when being bored or feeling lonely. Talking about the TikTok algorithm, she told me about the following in our final conversation:

I try not to be too surprised but sometimes, yeah, it is like content seems very driven towards you or a situation you are going through. But yeah, I guess less so now because I’ve used it more and, I mean, I’ve liked so many videos. Whoever is in control can see exactly what I’m into. It’s less surprising . . . but still weird. I’m still feeling like there is a spy in my phone because videos can get very specific” (Rhea).

In relation to trust, such is a strong statement. Rhea had been interacting with the TikTok algorithm for years. Such interactions would need to have been at least somewhat meaningful, otherwise she, more or less obviously, would not have continued interacting with the TikTok algorithm and the “me space” it curates for her on the app. At the same time, over these years, her trust has remained conditional, constantly challenged and renegotiated over and over again.

In that sense, accounts like that of Rhea underline the argument of Ruckenstein (2023) touched on before, that feelings of “mild paranoia” are not an exceptional but defining cultural feature. Relationships with machines, like algorithmic systems, are not harmonious (Lupinacci, 2024) but rather marked by constant dissonance that needs to be dealt with. They require a constant form of maintenance, the work of trust, through which people affirm or alter their position and that of the TikTok algorithm in their lives. And these processes, as I have found in my research, appear to be largely personal, unfolding in the direct relation between human and machine (Siles, 2023), through a continuous negotiation of trust and trustworthiness.

Conclusion

This article has discussed findings from an ethnography of TikTok consumption by juxtaposing them with the notion of trust. At the core, I argued that trust is a key component to the interaction between humans and machines on TikTok. More specifically, I evidenced how trust was vital in enabling meaningful entertainment experiences on TikTok, and that dealing with moments of trust being challenged was key to enable the durability of the relationship between human and machine on the app. Outlining the process of establishing, testing and maintaining trust, this article has demonstrated how human–machine interactions become durable configurations. It has shown, in other words, how trust is mobilised as a mechanism to enact algorithms as practically manageable entities in everyday lifeworlds (Siles, 2023) despite the potential vulnerabilities they introduce in that course.

From this angle, this article contributes further empirical evidence to the related literature. It provides further examples on the potential chilling effects of distrust (Kappeler et al., 2023), the complex and multifaceted nature of trust (Steedman et al., 2020), and the conditional nature in which it is negotiated and established (Wong et al., 2023). More closely connecting these debates to research from algorithm studies (Siles, 2023), I have theorised trust as a reflexive component, and one which has a strong embodied and affective component. Building on work foregrounding the affective component of human–algorithm interaction (for example, Bucher, 2017), I have looked beyond a typically cognitive evaluation of trustworthiness and paid attention to how trust is judged in concrete interactions.

The mediating role of trust in this configuration emerged from my data thus as a means to structure the relational positionality of human and machine to one another. When establishing trust, people assume an initial position and distance at which they want to stand in relation to a machine like the TikTok algorithm. This position needs to be continuously maintained when challenged, for instance in the form of trust tests. As such I defined moments of transgression from expected experiences that force people to stop their current interaction and reflect on the positionalities upon which the form of human–machine interaction they engage in is currently configured.

From here, my research echoes the diagnosis of Ruckenstein (2023) that “mild paranoia” is not a glitch but rather defining feature of algorithmic culture. Trust, I argue, allow us to capture how this latent sense of paranoia is dealt with in situ. Trust allows us to capture the continuous reflexive process in which people deal with these senses of unease (e.g. through forms of strategic ignorance, cultural resources that contextualise moments of irritation, learning about algorithmic transgressions, etc.). As such, trust consolidates the fluid and complex interactional dynamics that Siles, Gomez-Cruz and Ricarte (2022) put at the heart of a “popular culture of algorithms” – a cultural form in which people address their meaning-making needs with imperfect means (Martin-Barbero, 1984).

By studying trust in the foreground of this popular culture of algorithms, I argue, we can come to even better understand how exactly, in practical terms, people construct durable configurations of human–machine interaction in cultural contexts of ambivalence. Looking ahead, trust is thus an interesting concept for future research on human–machine interaction.

More systematically exploring how dimensions of education, prior experience and cultural context shape trust practices will be vital to further enhance our understanding of how meaningful and durable human–machine configurations can be formed (Suchman, 2007). Such appears crucial specifically as these interactions take place with ever more complex entities, visible chat bots like ChatGPT, but also integrated agents like platform algorithms. Trust is in that sense an integrative aspect to the construction and maintenance of these relationships – mediating how people enact machines as part of their everyday lifeworlds (Siles, 2023).