Abstract
A series of controversies have plunged large social media platforms into a perennial legitimacy crisis, fuelling interest in noncentralised digital spaces. In particular, the Fediverse has come to embody hopes that online speech governance can be more democratic and legitimate. This article examines this possibility by exploring how content moderators on Mastodon, the most prominent Fediverse service, negotiate their political legitimacy. Drawing on political philosophy and in-depth interviews, we demonstrate that this negotiation depends on how these individuals conceptualise their communities (‘instances’) as polities. These perceptions exist on a scale between viewing instances as bottom-up participatory spaces – broadly aligned with a democratic view of legitimacy – and as top-down governable spaces, reminiscent of aspects of an authoritarian understanding of legitimacy. By complicating the assumption that decentralisation necessarily returns social media to ‘the hands of the people’, the article also interrogates the role of technological design in shaping online political culture.
Introduction
Platform content moderation – how these digital intermediaries define and control (un)desirable speech – seems to be in constant turmoil. The decisions companies make to determine what content and behaviour ought to be allowed on their services have never been straightforward (Gillespie, 2018). But in the past 10 years, this inherent difficulty has been greatly compounded by a series of extraordinary crises – the reemergence of authoritarian populism, the COVID-19 pandemic and the wars in Ukraine and Gaza – which reinforced global concerns about platforms’ political role, and called into question their early, relatively lax moderation approaches. Over time, platforms increased the transparency of moderation decisions and rules, deepened their relations with civil society and governments and, above all, ramped up their intervention efforts, building up planetary bureaucracies of human and algorithmic actors (Douek, 2021). While profound, these moves were not able to solve the woes of moderation – if anything, they amplified them. Platform content moderation has come to be perceived as either too strict or not strict enough – a political impossibility that attracted endless controversy and regulatory attention. Taken together, such criticisms contributed to a ‘techlash’ that, as we will argue, is largely underpinned by a crisis of political legitimacy.
Yet, such a crisis has not only shaped how dominant social media works. It has also led to a renewed interest in alternative digital spaces, ranging from ostensibly ‘libertarian’ alt-tech platforms to the so-called noncentralised Web – the object of this article.
The appeal of these technically fragmented services is importantly political. Scholars have suggested that decentralising control over online spaces can help make online governance more legitimate, as seeking users’ participation would make the process more inclusive and meaningful (Zuckerman and Rajendra-Nicolucci, 2023). This view is at the heart of the Fediverse – the ensemble of ‘protocols, servers, applications and communities’ behind decentralised social media (Rozenshtein, 2023: 222). Built as a network of communities (known as ‘instances’) that can link with one another – but without a central authority to directly oversee them – the Fediverse embodies the promise of also decentralising speech governance, as it allows instances’ members to choose their servers’ moderation rules and the individual moderators who will be in charge of enforcing them (Gehl and Zulli, 2023). Such a structure has led its largest software, Mastodon, to brand itself as ‘radically different social media, back in the hands of the people [emphasis added]’ (Mastodon, n.d.).
However, despite the software’s flatter materiality and the opportunities for participation that this design might potentially allow for, instances’ rules are still enforced by administrators and moderators. Not unlike earlier ‘system operators’ (Schneider, 2024), these individuals enjoy, in principle, a more privileged position than regular Mastodon users. They are empowered to decide how other members ought to behave and the content they are exposed to, decisions that are likely to be politically laden and controversial. As any other form of power, this one calls for reasons and justifications. In other words, Mastodon’s design cannot by itself do away with the issue of legitimacy. This raises a critical but hardly studied question: How do Mastodon moderators negotiate their political legitimacy? Addressing it is important so as to probe the extent to which these services’ promise of a more legitimate form of social media, which rely on strong assumptions about the political consequences of technological design, have or have not been fulfilled, and why.
After situating content moderation as a political legitimacy issue, this article conceptualises legitimacy as existing at the entanglement of participation, representation and accountability. Then, drawing on interviews with Mastodon moderators, it argues that how these individuals negotiate their legitimacy depends mainly on how they perceive their instances politically. These perceptions range from viewing instances as bottom-up participatory spaces – broadly aligned with a democratic view of legitimacy – and as top-down governable spaces, reminiscent of aspects of an authoritarian understanding of legitimacy. In so doing, the article nuances common assumptions about the political possibilities of not only Mastodon but also of technological design itself.
Commercial platforms’ legitimacy crisis and the Fediverse
The techlash is, in important ways, a legitimacy crisis resulting from the widespread perception that large for-profit platforms have an undue influence on the lives of billions of people and, in particular, over their speech. Legitimacy is key for these firms, as their participatory nature means that their success is dependent on their users’ ‘voluntary cooperation’ (Abiri and Guidi, 2022: 104). In their early days, platforms’ legitimacy stemmed from their seeming neutrality (de Keulenaar et al., 2023). However, the past decade made clear that when they define what counts as (un)desirable, platforms significantly intervene in public discourse. Acting on speech directly, they essentially set ‘global speech rules’ (Gorwa et al., 2020: 2) – but in ways that often reflect Western norms and legislation and strongly obey commercial, not moral, logics. Without a system of checks and balances, transparency and accountability, however improved, remain limited (Klonick, 2018), all supported by an imposed, rigid form of consent (de Keulenaar et al., 2023).
The perceived inadequacy of mainstream platforms’ ways of governing speech has sparked contrasting attempts to reconceptualise commercial moderation’s legitimacy (see Douek, 2019; Haggart and Keller, 2021; Suzor, 2018; Tworek, 2019) and fuelled the desire for alternatives to their services, materialised in ‘a renewed momentum’ for the Fediverse (Raman et al., 2019: 217). Networking services under this umbrella term – a portmanteau of ‘federation’ and ‘universe’ (Rozenshtein, 2023: 218) – are noncentralised, that is, the servers on which the different software run are owned and managed independently, and can choose to connect with one another (Gehl and Zulli, 2023). In this way, the Fediverse seems to embody the founding principle of the Internet: decentralisation. Its emergence (and renewed popularity) is, however, driven by novel computational protocols and, critically, by novel political conditions – precisely the perception that for-profit social media has eroded that decentralisation principle. Consequently, it bears the promise of a new, politically legitimate approach to not only data privacy but also content governance – including moderation. As anyone can use the ActivityPub protocol that underpins these software to set up their own digital community (or ‘instance’), each instance can decide the types of content and behaviours that will be allowed among its members, as well as define its standards for moderation. In a type of organisation Rozenshtein (2023: 244) calls ‘content moderation subsidiarity’, instances can only moderate content as far as their own server: though the Mastodon Covenant offers suggestions of rules instances ought to follow, no central authority exists that would have a direct, final say in terms of whether a piece of content belongs on the wider network (Roth and Lai, 2024). An instance can also choose to silence or defederate any other instance (defederation), preventing that instance’s posts from reaching its timeline while keeping it visible for individual members or cutting ties altogether. 1 Although this model means that instances have few ways of holding other instances accountable, it is also proof that the Fediverse’s decentralisation is not just a matter of policy but one of architecture (Rozenshtein, 2023). This kind of decentralisation ‘by design’ could help the network resist legal and public pressure, as it may help instances avoid having to comply with moderation decisions imposed upon a superior actor that clash with their values (Raman et al., 2019). Most significantly, the decentralised moderation approach gives the network’s members more agency, allowing them to choose instances that best suit their needs and to use their leverage to influence governance, helping them more readily contribute their input than on traditional, corporate-controlled platforms (Roth and Lai, 2024).
Moderation on the Fediverse is done voluntarily. While, just like commercial content moderators, volunteer moderators support and supervise the discourse in online spaces, they are closer to their communities, of which they tend to act as leaders (Matias, 2019). Moderation labour is often invisible and undervalued, yet it involves acting as a ‘filter’ to sift out undesirable content – an activity that, although it gives users the impression of experiencing the platform in real time, effectively confers moderators the power to shape the content users will come across and interact with. This power is deeply political, as moderators adopt different norms and values and effectively take stances (Roberts, 2016). Content moderation practices, in fact, are pervaded by trade-offs of norms and values, requiring moderators to walk a fine line as they work ‘to facilitate cooperation and prevent abuse’ (Grimmelmann, 2015: 47).
These difficulties and tensions demonstrate that issues of concentration, agency and, ultimately, legitimacy have not been solved on the Fediverse. The ability of ordinary users to change the ActivityPub protocol remains limited; founders can punish instances that do not comply with content policies (such as the Mastodon Covenant) by omitting them from their index of instances. Above all, instances inevitably contain hierarchies among users, with potentially serious implications for how these online communities govern speech. Fediverse instances have administrators and, sometimes, moderators. Given the nature of speech governance, it is expected that these individuals will have to make politically laden and potentially controversial decisions, exerting power over how other members behave. In other words, while in different and potentially less unequal ways when compared to mainstream platforms, these actors’ legitimacy is not necessarily a given. These individuals might have to negotiate issues that are, at least in abstract, not different from those of corporate platforms. Yet how this happens, and the consequences of this process for the political promises of the Fediverse, remain wholly unclear. The budding literature on the Fediverse, and on Mastodon, has largely focused on broad attempts to explain their workings and differences in comparison to mainstream social media (Abbing and Gehl, 2024; Anderlini and Milani, 2022; Rozenshtein, 2023). Little is known about power struggles between members.
Addressing this aspect is important to probe the extent to which these services’ widely celebrated promise of a more legitimate form of social media and content moderation has been – or has not been – fulfilled, and why. To investigate how Mastodon moderators negotiate their legitimacy, we resort below to political theory to define legitimacy and then apply the concept to the issue of moderation in the Fediverse.
Defining (the politics of) legitimacy
The conceptualisation of legitimacy is a centuries-old, multidisciplinary endeavour. Given this article’s empirical focus and space limitations, we take a panoramic approach 2 to argue that legitimacy hinges, centrally, on the production of consent to power and that such consent exists at the entanglement of participation, representation and accountability (see Figure 1).
Legitimacy and its subcomponents (Authors, based on Beetham, 2012; Coicaud, 2002; Haggart and Keller, 2021; Schmidt, 2013).
Let us unpack what we mean by these three terms. First, theorists suggest that, through participation in the processes whereby social governance occurs, subordinates have the chance to express their own values and interests, a practice that, if done fairly, helps people accept these processes regardless of whose views end up prioritised (Beetham, 2012; Coicaud, 2002; Haggart and Keller, 2021). Simultaneously, as legitimate power ought to be accorded to what the community believes to be its proper outcomes, scholars allude to representation, which is grounded in the idea that subjects’ values and interests should be embodied (and acted upon) by the rulers (Coicaud, 2002). Finally, to ensure that the mechanisms of participation and representation can indeed foster people’s values and interests, power must be accountable. This entails not only transparency (about the rights and duties of both those in power and those subordinate to it – see Schmidt, 2013) but also the acceptance that violating and abusing consent can carry consequences, that is, sanctions. Consequently, as Coicaud (2002: 34) suggests, accountability demands from rulers the constant justification of their actions: every leader looking to prove their ‘right to govern’ must ‘satisfy, . . . try to satisfy, or . . . pretend to satisfy the needs’ of the community as failing to do so undermines their ability to govern efficiently, thus diminishing support and heightening the risk of collapse.
In reality, the ways in which legitimacy is sought and negotiated vary, with important political consequences for the normative nature of power (Kailitz, 2013). To understand these consequences, we would like to argue that the relationships between its three aforementioned components create a conceptual spectrum with democratic legitimacy at one end and authoritarian legitimacy at the other (see Von Soest and Grauvogel, 2017).
We propose that a democratic view of legitimacy sees representation as emerging from participation and accountability. The rulers who abide by it view legitimacy as highly dependent on their subordinates’ acceptance of authority (Scherz, 2021), thus constituting legitimacy as a bottom-up phenomenon. Therefore, they seek their subordinates’ participation, as they view it as the key to learning their values and their beliefs as to the proper outcomes of governance, which should, respectively, underpin their community’s laws and rules and the rulers’ actions for power to be legitimate (Beetham, 2012). In fact, in a democratic view of legitimacy, rulers acknowledge the fact that they should stand for the group’s welfare and act accordingly for their subordinates to feel represented (Coicaud, 2002). Together, these actions aim to make their subordinates feel heard and seen, to build trust that the rulers have their best interests at heart – and thus to create meaningful representation from the bottom up.
An authoritarian view of legitimacy, on the other hand, largely dismisses participation and accountability and takes representation for granted, we argue. Rulers who follow this view of legitimacy assume that their values and norms ought to be shared by all those they govern and see themselves as automatically representative of their subordinates (Von Soest and Grauvogel, 2017). As such, these rulers do not seek their subordinates’ participation and do not inquire about their views of what governance should be like. They do not try to be accountable in the democratic sense of transparency and openness to scrutiny, either, as they see these aspects as unnecessary and potentially destabilising (Hollyer et al., 2015). Yet, this does not mean that they do not negotiate legitimacy at all. In fact, authoritarian legitimacy is negotiated, partly, by emulating what democratic legitimacy entails. Though without accounting for their subordinates’ values, authoritarian rulers still attempt to justify their rule, using procedural claims to legitimacy (Von Soest and Grauvogel, 2017). These claims involve the creation of rules in an attempt to dodge accountability by concealing the various arbitrary aspects of their authority (Zhang, 2024). Given that these rules are not directly sourced from users’ participation, however, rulers with an authoritarian view of legitimacy then resort to ‘identity-based legitimacy claims’ through which they attempt to build and reaffirm their community’s shared identity to ensure subordination (Von Soest and Grauvogel, 2017: 287). Representation, thus, is not based on a feeling among subordinates – as in democratic legitimacy – but rather on an assumption of ideological homogeneity that is engineered and tended to, but from the top.
Though consent – the heart of legitimacy – foregrounds those governed as the legitimating body, the article concerns rulers’ legitimacy negotiation tactics and focuses on them instead. Specifically, to understand how Mastodon moderators negotiate their political legitimacy, this article investigates the three constituent variables of consent – namely, participation, representation and accountability – through the following subquestions:
What role does representation play in how moderation happens within Mastodon instances?
What role does participation play in how moderation happens within Mastodon instances?
What role does accountability play in how moderation happens within Mastodon instances?
Methodology
We conducted semistructured, in-depth interviews with moderators from Mastodon, currently the largest and most popular software of the Fediverse (for other works using this approach to study Mastodon, see Zulli et al., 2020, and Caelin, 2022). In-depth semistructured interviews were chosen as they help explore social actors’ practices and perceptions, allowing us to gain insights into how Mastodon moderators view and negotiate their legitimacy (Charmaz and Belgrave, 2012). Notwithstanding interviews’ inherent limitations (selective recall, social desirability bias and limited generalisability), our choice generated insightful findings, which we explore in the next section.
Participants were selected using purposive sampling – they had to be moderators of a Mastodon instance – and were chosen with a view to diversity. The aim was to include moderators of both general interest and topic-specific instances and of instances with open, closed or approval-based registration. The participants were recruited through a toot – Mastodon’s equivalent of a tweet – on the Mastodon profile of one of the researchers and by using the Fediverse Observer, an online list of Fediverse instances – to find Mastodon instances and contact their moderators directly. The characteristics of the participants and their instances are outlined in Table 1.
List of interviewees and the characteristics of the instances they moderate and/or administer.
Small = 1–499 monthly active users (MAU); Medium = 500–1999 MAU; Large = 2000–4999 MAU; Very large = 5000 + MAU.
Mastodon’s historical focus on privacy meant that many users were sceptical about being researched, and interview requests were often met with silence and refusals. Consequently, our data was eventually sourced from 15 in-depth interviews, which were conducted in May 2024. We prepared an informed consent form which detailed the ways participants’ data would be handled and emphasised their right to stop the interview, for which we received approval from the University of Groningen’s ethics committee. Moreover, we gave interviewees the option to choose the videoconferencing software we would use for their interview and pseudonymised them. Acknowledging their scepticism towards researchers and concerns about privacy, we decided not to collect demographic data about interviewees – a type of information that, we judged, was not essential to address our research question.
The interviews lasted between 50 and 115 minutes and were conducted and recorded using online conferencing software. They began with a few opening questions about the interviewees’ instances and what the moderation thereof entails. To help facilitate an understanding between interviewer and interviewee, questions did not use the terms ‘legitimate’ or ‘legitimacy’ but referred instead to participation, representation and accountability. We used software to transcribe the conversations and checked transcriptions manually for accuracy.
A combination of inductive and deductive thematic analysis was used to examine the data (Braun and Clarke, 2006). This approach combines using data-driven and theory-driven codes (Fereday and Muir-Cochrane, 2006), allowing us to use the subconcepts of participation, representation and accountability as general themes to inform the analysis. These theory-driven general themes, in turn, guided the development of data-driven initial and second-order codes. To increase transparency and the quality of our conclusions, the findings were sent via email to interviewees, who were offered the chance to comment on them.
Findings
Although the findings are organised following the three dimensions of legitimacy, a finding that cuts across these themes is the existence of different broad types of instance governance. While many Mastodon communities align with an ultimately hierarchical organisation where moderation is mainly led by a few individuals, others do not. These instances have opted to be cooperatively governed, making it a point to be steered by all their members’ ideas and values – not only moderators and administrators. The differences in their approach are outlined in the following subsections.
Participation
Instance members’ participation in moderation is conceptualised as a prerequisite for consent – which, this article argues, is at the heart of moderators’ legitimacy. As moderation on Mastodon is done according to rules, which each instance can define for itself, we investigated both the opportunities moderators give members to participate in moderation and the opportunities to partake in the establishment of the rules that guide it.
The interviewees revealed that their instances’ members can participate in moderation mainly indirectly, by reporting other members or posts, contacting moderators or volunteering to become moderators themselves. This is partly due to the instances’ approaches to moderation, which, many reported, are not directly participatory. Peter, for instance, suggested his instance being the administrators’ ‘pet project’ as a reason, and Chris admitted that their approach is simply ‘more authoritarian’ than democratic. Only the moderators of a cooperatively governed instance we interviewed (Lisa and Sacha) encouraged their members to participate in moderation more actively by joining working groups through which they could shape its functions. Another reason is the way Mastodon moderation tools are structured, whereby regular users have no access to moderation tools beyond reporting or blocking (Adam), whereas the interviewed moderators reported feeling constrained or even ‘steered . . . towards different actions’ (James) by the tools’ design, highlighting Mastodon’s founder Eugen Rochko’s and Mastodon’s computational makers’ influence on the tools available. Given such limitations, interviewees such as Sarah and Lily reported asking instance members for opinions on instance-wide decisions (such as defederation) and encouraging them to use user-level moderation tools, like muting or blocking fellow users or instances. However, participants explained that even when given such opportunities, instance members did not appear interested in partaking in moderation nor in volunteering to become moderators. Instead, the interviewees’ instances’ members preferred to participate by appealing the moderation decisions with which they did not agree – a tactic which only appeared to change a decision’s outcome on instances where moderators did not discuss every case, according to Cassie.
Complicating simple assumptions of decentralisation, most participants said that their instances’ rules were established upfront by administrators – sometimes without consulting the moderators. Once more, only the cooperative instance sought members’ active participation by inviting them to vote on rule drafts and proposed changes. In other instances, although the administration and moderation teams were open to members’ participation in the form of feedback, they did not seek their input in the creation of the rules. Various reasons – oftentimes linked to the instances’ administrators’ viewpoints – underpin this approach, such as the idea that this kind of participatory approach would be messy due to vastly different perspectives or the belief that an instance’s rules ought to, first and foremost, reflect the beliefs of its administrator – who often owns the physical server on which it runs, said Lola. The rules, however, reportedly could evolve based on moderators’ observations, which, in turn, depend on users’ participation in submitting reports for moderation. Together, these insights suggest that although members’ participation is mostly indirect, it is valuable to the process of moderation as it contributes reports upon which the latter relies and influences the instances’ development. However, given the importance of administrators’ perspectives about how the instances should be moderated and how their rules should be defined, these findings also highlight that participation is not contingent upon moderators’ actions and beliefs alone.
Representation
Representation – grounded in a community’s feeling that the rulers embody the interests and values of subordinates – is key to moderators’ legitimacy. The fact that all interviewees reported wanting their moderation actions to keep their users feeling somehow ‘safe’ and ‘happy’ (as Adam summarised) appears to suggest a common desire by participants to act on behalf of their communities. Yet the ways in which they deal with this wish vary considerably.
This can be seen in the definition of a key governance mechanism – instances’ moderation rules. In most of our interviewees’ instances, members’ values and interests appear to be largely assumed beforehand: rules were mostly based on what administrators believed would be the views of current members or the type of user they sought to attract. Except for the cooperatively governed instance – of which the members ‘collectively define the vibe’ (Lisa) through a publicly available list of norms – there was simply no attempt to understand what members wanted. Some justified this saying an instance’ rules were ‘just baseline ethics’ (Chris); others feared discussions would result in controversy (Theo). In this way, the values and interests that underpin moderation rules actually precede members’ signing up to an instance.
At the same time, in some instances, representation does appear to be a concern from a perspective of demographic diversity. Interviewees like Sarah and Lisa reported making efforts to diversify their moderation teams’ backgrounds, gender identities and perspectives to account for the variety of viewpoints and backgrounds of their instances’ members, in a bid to help them adjudicate on moderation matters in an informed way.
However, others, like Theo, Chris and Peter, found it difficult to evaluate the extent to which they are actually representative of their instances’ members – partly due to some instances’ large sizes, general interest nature and the lack of statistics about users and partly due to the previously mentioned selective participation of their members, who reportedly rarely respond to feedback requests. This proved to be difficult to gauge also on the cooperative instance, as – due to a technical limitation of Mastodon – its members need to use a separate platform to vote on the rules, which could hinder their engagement. Therefore, the moderators acknowledged that voting would only be representative of the members interested, echoing Peter’s claim that members’ lack of engagement is a preference in itself. Together, these insights show that although moderators make efforts to represent their users, these are contingent on their instances’ administrators’ approaches to seeking members’ input, as well as on members actually contributing it.
Accountability
As explained above, we approach accountability – whereby subjects can ensure that rulers abide by the consent given to their power and is thus key to the resilience of legitimacy – through three aspects: (i) transparency, (ii) justifications and (iii) the possibility of sanctions.
Most interviewees reported that their instance’s moderation rules are publicly visible, both to help potential users assess their fit with the instance – though some participants doubted whether users actually read them – and to help moderators avoid subjectivity in their decisions. Meanwhile, on the cooperative instance, transparency’s key aim was to build trust in moderators and their actions. Despite being visible, however, some instances’ rules were not exhaustive. Lola, for example, explained that certain members’ behaviours may create the need for a new rule. Transparency appeared to be most important in decisions with instance-wide impact – such as defederation, spam waves or rule changes – which interviewees reported publicly sharing and explaining. Individual-level decisions’ outcomes, on the other hand, were only shared with the instance member concerned, with degrees of explanations ranging from none to a detailed reference to the broken rule. Some participants also reported only sharing explanations about such decisions upon request, as Peter explained that publishing them risked making them ‘a bigger deal than [they] really need to be’. Many moderators shared the feeling that making all decisions public would be problematic. While some mentioned privacy reasons, others suggested that it would bring them more work and unnecessary attention, while also potentially backfiring against the moderation team. Interviewees like Peter also explained that they saw little use in sharing all decisions publicly, as their instance’s members already hardly engaged with information about instance-wide decisions. Such insights suggest that moderators might think their instance members appear largely uninterested in their instances’ decision-making processes.
Expectedly, most participants had never had to justify why they are a moderator or the ways they moderate – which Adam thought might be because members are supposed to have accepted the rules he enforces, while James said most of his instance’s members were simply ‘inherently friendly’ to moderators. Nevertheless, interviewees wanted their decisions to be considered appropriate according to their instances’ rules and by their instances’ members – particularly as some saw ‘no good reason why [they] are moderators and not others’ (Peter), which urged them to be able to justify their role to themselves. Others, including Sarah, believed that being ‘willing to put in the emotional and physical effort’ into moderation justified her position, helping alleviate her self-doubt. Sacha, one of the cooperative instance’s moderators, felt he had an ‘obligation’ to explain as his decisions impacted both members and other instances. Overall, participants reported only ever needing to justify their reasoning to other instances’ moderators whose rules are often incompatible – showing the difficult politics of moderation in an online space as large and varied as the Fediverse.
While our interviewees reported there had been no need to come up with sanction procedures for moderators to date, Mastodon’s decentralised, open-source nature also means that software-wide consequences for misconduct are technically infeasible. Though Adam described this as an ‘intrinsic power imbalance’ between users and moderators, other interviewees suggested that moderators could nonetheless face consequences in three ways. First, their moderation rights can be revoked by their instances’ administrators, demonstrating not only internal accountability but also the administrators’ power. Second, other instances may choose to federate with their instance, which can tarnish its reputation. And third, members may change instances following disagreements. While the last measure may appear as a key accountability mechanism, many interviewees were unalarmed by members leaving their instance, and some even encouraged it: Cassie, for example, stated that discontent members are ‘welcome to leave and . . . everybody will be happier for it’. Such stances may be explained by the Fediverse’s lack of commercial incentive, as – unlike commercial social media platforms – instances do not necessarily seek growth.
Discussion
The findings suggest that all moderators negotiate their legitimacy, but to different extents and in different ways. Some of the interviewees appeared more interested and invested in seeking legitimacy than others, which indicates that Mastodon’s legitimacy mechanisms and regimes are more complex than what its decentralised, instance-based nature might suggest. More specifically, moderators’ legitimacy negotiation tactics appear to depend mainly on how these individuals politically perceive their instances – understandings that are associated with particular legitimacy-seeking practices.
Building on our conceptualisation of the politics of legitimacy seeking, we contend that these perceptions materialise as two broad types: the view of instances as bottom-up participatory spaces, which broadly aligns with a democratic approach to legitimacy, and the view of instances as top-down governable spaces, closer to an authoritarian take on legitimacy and which appears to be enabled by Mastodon’s technological design. Importantly, these are just explanatory models. Only in the case of cooperative instances do these views apparently overlap with an actual server. Rather than a fully democratic versus a fully authoritarian instance dichotomy, multiple shades of legitimate and illegitimate speech governance seem to coexist in most participants’ descriptions – and imaginations – of their instances.
Instances as bottom-up participatory spaces
Some of the moderators demonstrated, in many ways, a clear willingness to put their instances’ members first and run their instances for them and their benefit. They perceived their instances as spaces designed for users and highly valued their perspectives. The interviewees who saw their instances as bottom-up participatory spaces sought participation from their users in various ways and at different stages of the moderation and governance processes. For example, they asked members for their opinions on instance rules and changes therein, organised votes or remained open to feedback – though without necessarily outwardly seeking it. Such practices show a desire to moderate legitimately.
Participants who viewed their instances as participatory were also open and willing to be held accountable, as they displayed their rules transparently and referred to them to decrease the role of subjectivity. In such instances, representation appeared to emerge from participation and accountability. Moderators sought to learn their members’ values and built their instances’ rules upon them, aligning with the conditions that subordinates should be able to express their will and recognise themselves in the outcomes of power for it to be wielded legitimately, as suggested by Coicaud (2002) and Beetham (2012). They encouraged members to vote on new rules and instance-wide decisions, favouring electoral choice in their decision-making and simultaneously learning what instance members believe to be the proper outcomes of power (Beetham, 2012). Through visibility and transparency – of rules, moderation outcomes and explanations – the moderators attempted to show that they try to act upon the interests of their communities – thus helping ensure that their members feel represented (Coicaud, 2002).
Such behaviours appeared to be intrinsically motivated: unlike on commercial social media, participation, accountability and the resulting representation of users were not a consequence of fear of backlash, but rather an embodiment of the moderators’ view of what social media moderation and governance should look like. This is not to say that perceptions of instances as participatory ‘automatically’ meant that moderators held a fully democratic view of – or approach to – legitimacy. In fact, such a view is not a strict description of any instance’s reality, though the cooperative instance’s commitment to participation in moderation, representation of members’ values and accountability through transparency and explanations comes close. Instead, such perceptions created specific tendencies in moderators’ practices – particularly in terms of participation and accountability – without precluding the possibility of behaviours that do not align with democratic practices. In other words, their perceptions broadly aligned with a democratic approach to legitimacy.
Instances as top-down governable spaces
The perception of instances as bottom-up participatory spaces, however, was largely overshadowed by a view of servers as top-down governable spaces. In such cases, the perceptions moderators had of their instances were often shaped by the visions of instances’ creators, who appeared to see their instances as their property. While it is known that administrators hold almost complete control of their instances – after all, they are the owners of the hardware on which those run – such importance being given to administrators’ views goes against their portrayal in existing literature as those merely ‘responsible for carrying out the day-to-day administrative tasks’ (Anaobi et al., 2023: 3110). Instead, it points to a political hierarchy wherein the administrators’ position makes their voices more significant than those of other members, thus defying the instances’ promise to embody ‘democratically-run platforms for civil discourse’ (Struett et al., 2024: 1). In fact, the interviewees who saw their instances as top-down governable spaces had a tendency to negotiate their legitimacy in ways closer to an authoritarian approach, as we defined above.
Specifically, they did not seek their members’ participation in defining rules, and only allowed members to participate in moderation indirectly, by reporting undesirable content or other users – an equivalent to commercial social media platforms’ mechanisms to flag objectionable content. The moderators also wanted their moderation decisions to be seen as appropriate according to their instance’s rules, which were often unilaterally established by their administrators without members’ participation. In this view, instances are like political fiefdoms: users are free to join and leave but should not expect to have a substantive say in how the community is governed.
Paired with the tendency to rely on administrators’ rules but also to moderate beyond them, such an understanding suggests – as mentioned by Chris – an authoritarian enactment of legitimacy, namely, procedural legitimacy. This approach entails the use ‘rule-based mechanisms’ to justify one’s rule despite its arbitrary aspects (Von Soest and Grauvogel, 2017: 5). Rather than reflecting the principles of the rule of law, this kind of practice can be termed ‘rule by law’, as it effectively engages instances’ laws ‘as an instrument of rule’ (Solomon, 2015: 427).
The findings suggest that such an approach is made possible by the fact that moderators do not have to fear backlash from their instances’ members, which also means that they do not necessarily look to be accountable. These interviewees did not appear to worry about other instances defederating from theirs or members leaving their instance for another. In fact, some moderators even encouraged dissatisfied members to change instances. This sets them apart from commercial social media platforms – which depend on their users for their profitability and continued existence – and simultaneously appears to eliminate a significant reason to actively seek legitimacy, as members’ ‘voluntary cooperation’ appears inessential (Abiri and Guidi, 2022: 104).
This kind of authoritarian view of legitimacy takes representation for granted. As the rules – and, thus, the values and interests that underpin them – are often devised by administrators prior to members joining the instance, they become selection criteria through which representation is designed to be automatic rather than collectively built. Amid such manufactured homogeneity, moderators arguably negotiate their legitimacy by, mainly, creating and enforcing the boundaries of their community, in an approach reminiscent of that of regimes that centre their legitimacy claims around identity. Specifically, this approach to representation resembles authoritarian regimes’ legitimacy claims based on ideology, whereby a particular ‘belief system [is] intended to create a collective identity and, in some cases, a specific societal order’ (Von Soest and Grauvogel, 2017: 4).
This notwithstanding, it is important to note that while there are parallels between this view of moderation legitimacy in Mastodon instances and views of authoritarian legitimacy in states – they differ considerably. Moving instances is obviously easier than changing states of residence, which also prevents those in power from imposing their views on their subordinates. As instance members are sometimes encouraged to change instances, moderators and administrators are free to choose the communities they would like to foster, which points to selectivity. As such, Mastodon moderators arguably take a novel approach to representation, as they seek to have their own values represented while reserving members the freedom to leave – a practice enabled by the aforementioned materiality of the software.
The ambivalence of Mastodon’s decentralised architecture
Beyond moderators’ perceptions, the materiality of Mastodon and the tools at their disposal appear to influence the extent to which moderators negotiate their legitimacy. Findings suggest that Mastodon’s decentralised control, ironically, can create hierarchies of power on at least three levels.
First, between users and moderators, as instance members do not have access to the same tools as moderators, and no public moderation log exists. This establishes what Adam called an ‘intrinsic power imbalance’ as it not only impacts the extent to which members can participate in moderation, but also the ways in – and the extent to – which moderators can be transparent about their decisions, and held accountable therefor.
Second, between moderators and administrators, as the possibility for anyone to create their own instance – intended to decentralise power – simultaneously decentralises power and recentralises it at a different scale. Specifically, while it indeed distributes power at the scale of the Fediverse – which is fragmented into instances – at the level of the instance itself, our findings echo those of Ermoshina and Musiani (2025) as they suggest that Mastodon’s technical infrastructure appears to concentrate a significant amount of power into administrators’ hands.
And third, between moderators and Mastodon’s computational makers – including Eugen Rochko – who, interviewees reported, retained, at the time of the interviews, significant control over the software’s code despite it being open-source. Such a dependency on Mastodon’s computational makers’ visions of moderation – which inform the interface available and are infused with particular values and beliefs – suggests that the same code which fragments power also contributes to its centralisation.
Though the latter two levels of perceived hierarchies are effectively outside of moderators’ direct control, such inherent hierarchies challenge Mastodon instances’ potential as ‘democratically-run platforms for civil discourse’ (Struett et al., 2024: 1), bringing the software closer to its commercial counterparts – none of which were endowed with moderation tools for democratic decisions.
Yet, it is important not to fall into an inverted form of techno-determinism and assume that decentralised software will necessarily create the same issues as commercial platforms. In theory, it is possible to envision a version of the ActivityPub protocol that more transparently allows moderation actions to be seen and contested, which makes it easier for instances to be collectively governed and funded, and where code design and management are made technically easier and truly democratised. Democratically legitimate governance, however, cannot be magically brought about by design alone.
Conclusion
Altogether, the ways moderators on Mastodon negotiate their legitimacy do not necessarily demonstrate how to moderate in a more legitimate way than on commercial social media platforms. Instead, they show that moderation on Mastodon consists of various combinations of more or less legitimate approaches in terms of participation, representation and accountability, which appear to depend on the ways in which moderators perceive their instances as political communities.
As previously explained, most participants’ approaches appeared to lean towards a nondemocratic view of legitimacy. This is further exemplified by the fact that all of our interviewees reported wanting their moderation actions to contribute to their instances’ ‘safety’, without elaborating on what it means. While such an approach appears to suggest their desire to act in the interest of their instance members, the fact that this interest is assumed to be safety is reminiscent of authoritarian regimes’ legitimacy claims based on their ability to provide security to their citizens.
Providing ‘safety’ by engineering homogeneous instances – as the moderators showed little desire to engage in discussion with users, or attempt to navigate and reconcile their different perspectives – also shows a desire to consolidate control by avoiding (and perhaps suppressing) conflict. Homogeneity in an instance is likely to limit controversy and thus the need for politically laden decisions, giving the impression that power is absent and limiting moderators’ need to negotiate legitimacy in the first place.
Therefore, while it seems intuitive to say that ‘decentralizing control over online spaces can bring legitimacy’ as it offers users greater input and more choices (Zuckerman and Rajendra-Nicolucci, 2023: 7), it does not by itself return social media to ‘the hands of the people’ (Mastodon, n.d.). On the scale of the Fediverse or of Mastodon, an approach based on giving users the choice between instances with different rules – which appears to help address the impossible conundrum of moderation – only seems to do so because it devolves highly charged political issues to the more granular level of individual communities. If that is the case, rather than helping solve the challenges of content moderation or necessarily demonstrating how to moderate in a more legitimate way, our research shows that Mastodon moderators’ approach might demonstrate the endurance of the difficulties early online spaces faced when dealing with speech governance (Schneider, 2024).
However, it is essential to say that both Mastodon and the Fediverse are very diverse, and any attempt to put a simplistic, all-encompassing conceptual blanket over them would be inaccurate. The above conclusion may differ depending on the sample under observation, and more research is needed to corroborate such findings. For instance, future studies could investigate the perceptions of legitimacy of Mastodon users to supplement and nuance the moderators’ insights this article focused on. Moreover, they could research legitimacy in other spaces within the Fediverse – such as Pleroma or Pixelfed – or investigate legitimacy more broadly, focusing on the actual disputes around how to design and tweak the affordances allowed by the ActivityPub protocol, which underlies them all. This study’s findings, in turn, aim to contribute to further dialogue about what moderation should entail, who should be in charge of it, and how it could be used in a way that is mindful of our differences, so as to make social media a place that truly serves its communities – rather than mainly those who control the spaces in which they form.
Footnotes
Acknowledgements
We would like to thank the interviewees for their time, trust and insightful contributions. Thank you also to the editors and three anonymous reviewers for their extensive, thoughtful, and highly constructive comments.
Author contributions
Klara K. Matusewicz led the research design and the writing of the article and conducted and analysed the interviews. João C. Magalhães participated in the research design, findings interpretation, and article writing.
Declaration of conflicting interest
The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship and/or publication of this article: The writing of this article was partially supported by a Veni grant awarded to João C. Magalhães by the Dutch Research Council (NWO).
Ethical approval and informed consent statements
The interview guide and methodology for this study were reviewed and approved by the Research Ethics Committee of the University of Groningen on 14 October 2024 (Ethics approval number: ID 98895845). Informed consent was obtained from all study participants in writing, after they were clearly informed about the purpose of the project, the duration of the study and the voluntary nature of participation, the possible risks and discomfort they could experience, as well as the way their data would be treated and stored confidentially.
Data availability statement
The original interview dataset cannot be made available due to participants’ privacy. A fully anonymised version of this dataset might be made available, depending on the nature of the request.