Abstract
Centering on social media’s public- and profit-oriented nature, this study theorizes how social media users are empowered and constrained when participating in platform governance through user-initiated expressions on platforms. The empirical analysis focuses on user responses before and after Elon Musk’s official acquisition of Twitter, utilizing cluster analysis and topic modeling to examine the volume and content of related discourses among different Twitter user groups. Our results point to user constraint in platform governance. Although a diverse set of users, such as partisans, bots, and cryptocurrency enthusiasts, spoke up, they had diverging and sometimes conflicting objectives; and partisans dominated the conversations. There was an upsurge in user volume and activity level post-acquisition among liberal users, whose critical voices on platform governance might have bolstered platform business. Potential bots also increased in volume and amplified political topics. Our findings shed light on the challenges of user-driven platform governance, underscoring the complex interplay between platform users, economy, and governance.
Keywords
While traditional media outlets were the primary gatekeepers of information in the 20th century, social media platforms have become critical mediators for information in the 21st century (Gillespie, 2018a; van Dijck et al., 2018). Given the significant impact of social media platforms on society, platform governance has received abundant public and scholarly attention. Existing research focuses on the relationship between platforms, states, and non-governmental organizations (NGOs) in shaping platform governance (Duffy & Meisner, 2023; Gillespie, 2018a; Gorwa, 2019). However, platform users can also play a role in platform governance through a bottom-up process (Helberger et al., 2018). In particular, users can collectively exercise agency to shape online discourses within social media and affect platform policies and economics (Gillespie, 2018a; van Dijck & Poell, 2013).
Integrating research on platform governance, platform affordances and user agency, and platform political economy, we introduce a framework for understanding the user-initiated, expression-oriented form of platform governance. Specifically, this framework considers the dual nature of social media platforms—as both public-oriented spaces and profit-driven businesses—and explains how social media users are empowered and constrained by the dual nature when participating in platform governance via expression on platforms. On one hand, active users exert influence on platforms through vocal and visible participation, including expressing their platform-related opinions directly, boycotting to signal disapproval, and buycotting in support of a platform change (van Dijck, 2013). 1 By expressing their voices, they exercise the power of opinion; by modulating their activities, they wield the power of numbers (van Dijck & Poell, 2013). On the other hand, user activities on platforms can be aggregated and monetized as user data (Sadowski, 2019), which can paradoxically make critical voices financially beneficial for platforms. The collective power of users in platform governance can be further weakened because social media discourse in general is seldom unanimous and inherently consists of diverse user groups pushing their competing values and objectives (Gillespie, 2018a; Suk et al., 2023).
It is worth noting that this framework centers around users, emphasizing user agency in the form of expressive participation that leverages the existing space offered by a platform and its affordances, even though these affordances are not purposefully designed for such user-driven platform governance. This contrasts with a more structured, organized way (e.g., voting on a policy) for users to contribute to the platform governance, where platforms provide a customized channel to invite users to participate in governance processes.
One way to study expression-oriented user participation in platform governance is to examine their responses to a specific platform governance shift: ownership change. Social media platforms are often tied to their owners, like Mark Zuckerberg with Facebook, Jack Dorsey with Twitter, and Donald Trump with Truth Social. The acquisition and subsequent rebranding of Twitter as “X” in 2023 by Elon Musk is a recent example that sparked considerable controversy due to the anticipated and actual changes it introduced. Since the ownership structure of a social media platform is an important component of its governance and can produce significant downstream effects on not only its business model but also user experience, users are motivated to react to ownership change. In this process of expressing their opinions and adjusting their levels of activity en masse, they support, oppose, or simply comment on the platform-governance-related change, which can exert an impact on subsequent platform governance. In this process, they are empowered and constrained by platform affordances and economics.
Specifically, we situate our analysis in Musk’s takeover of Twitter, delving into users’ responses (in terms of volume and content) during two critical phases: the proposal stage of Musk’s acquisition (March to April 2022, pre-acquisition) and the actual takeover stage (October to November 2022, during and post-acquisition). We employ cluster analysis to detect different user groups and topic modeling to identify the content of their discussions, followed by statistical tests to observe the volume and content of related discourse from different Twitter user groups. Our findings illuminate participation from various user groups throughout the phases, including partisans, bots, and cryptocurrency enthusiasts. There is a notable surge in user volume among liberal users and a heightened prevalence of political topics among bots. These findings provide important theoretical contributions to the understanding of user dynamics in platform governance, while also offering practical implications for users to understand their power and constraint in platform governance and for social media platforms to improve governance structures.
In the literature review section, we first discuss the dual nature of social media platforms and how they are governed, and then theorize the role of social media users in platform governance while proposing our research questions.
Social Media Platform Governance: User-Initiated and Expression-Oriented
The term “platform” is multifaceted and polysemantic (Flew, 2021), serving as a metaphorical construct that is routinely examined through technological, social, economic, and political lenses (Gillespie, 2010). Digital platforms are technological infrastructures facilitating individuals to establish and utilize networks (Poell et al., 2019), but they vary by service type and business model (Langley & Leyshon, 2017; Srnicek, 2017). In this study, we focus specifically on social media platforms, which, despite their internal variation, share the four stable elements of profile, network, stream, and message (Bayer et al., 2020). The dual nature of social media platforms as both public-oriented spaces and profit-oriented businesses is key to understanding platform governance.
Social media platforms can be seen as public-oriented spaces where users are connected in networks, create content, and share information in networked contexts (Boyd et al., 2010; Zhang et al., 2022). Social media platforms have become a primary venue for networked actors to draw public attention to an issue, advance their arguments, and spark news media amplification and response from other sectors of society (Freelon et al., 2018; Tufekci, 2013). This capacity of social media mirrors the agenda-setting and opinion-shaping power of traditional media, making social media platforms critical mediators of information (Gillespie, 2010, 2018a) and wielders of opinion power in society (Helberger et al., 2020).
Despite celebratory narratives about how platforms epitomize “open, neutral, egalitarian and progressive” values, social media platforms are run by platform companies, that is, corporations that create platforms in a relentless pursuit of profits (Gillespie, 2010, p. 352; Gorwa, 2019). In this sense, just like news media organizations that offer mainly free and engaging content to media consumers and sell their attention to advertisers, social media platforms provide free services to their users, whose daily activities on platforms are organized and encoded into standardized data, which in turn are mined and monetized (Alaimo & Kallinikos, 2017). While traditional media organizations rely on audience size for advertising revenue, user volume and engagement level are crucial for social media platforms (Sadowski, 2019), which often leads to the prioritization of popularity and likeability-driven logic to produce more user attention and engagement (van Dijck & Poell, 2013).
Since social media platforms straddle both the public and the business spheres, platform governance has become an increasingly important research subject. One noteworthy model is the “platform governance triangle,” which outlines a platform governance structure through self-regulation by platform corporations, external regulation by governments, and co-regulation by NGOs (Gorwa, 2019). However, the efficacy of self-regulation is called into question, when power is concentrated in the hands of a few individuals with business interests and when platforms prioritize corporate values and interests in determining content and features (Gillespie, 2018b). This situation mirrors the influence of traditional media ownership on editorial decisions—as seen with figures like Rupert Murdoch influencing outlets like the Wall Street Journal (Archer & Clinton, 2018; Wagner & Collins, 2014). On social media, where platforms wield pronounced economic and communicative power (Flew, 2021), ownership has a profound impact on determining platform content, technology, and governance practices (van Dijck, 2013). Tech magnates, including Mark Zuckerberg at Facebook, Donald Trump at Truth Social, and most recently, Elon Musk at Twitter, significantly shape the dynamics and visibility of content on their platforms (Fukuyama et al., 2021).
In contrast to the triangular governance structure, this study theorizes the underexplored user-initiated expression-oriented form of platform governance by focusing on active users and their expressive participation. In theory, social media platforms advertise that they are created for their users, making users a logical component in platform governance (Gillespie, 2018b). In practice, users do not simply passively take platforms as a given: “While platforms structure user activity, users also have power over platforms—maybe less so as mere individuals or groups, but more in the aggregate, the slow, unrelenting shifts in what people seem to want to do” (Gillespie, 2018a, p. 23). Similarly, van Dijck (2013, p. 33) notes “articulated user responses” to “platform changes that affect their online experience” as well as quitting a platform as major forms of user agency. Therefore, active users have an inherent role in platform governance as they contribute to the nature and characteristics of platforms by engaging in all kinds of expressive activities within platforms (Helberger et al., 2018), a process that affords them agency yet is shaped by social media digital architecture.
Importantly, it is worth noting that such less explored user-driven expressive efforts to participate in platform governance through expressions and discussions are distinct from platform-driven means for users to participate in governance. Platforms can actively solicit user expression and input on governance matters by providing formal and tailored mechanisms. For instance, Twitter introduced “Birdwatch” in 2021 (Coleman, 2021) to enable users to add community notes and report actions that breach the platform’s community guidelines (Pröllochs, 2022). Similarly, Facebook, starting in 2009, allowed its users to influence policy decisions directly through a voting mechanism (Johnston, 2012). As users’ opinions or advice on governance can be more directly and clearly communicated to platforms via these means, their voices can likely result in a more immediate, concrete, and tangible impact on platforms.
In contrast, we focus on a user-initiated, expressive approach where users start conversations about platforms on platforms, expressing their opinions and expectations and leveraging affordances and spaces offered by platforms, although not specifically designed for the purpose of governance. Figure 1 summarizes our framework for understanding user-initiated expression-oriented platform governance, underscoring how user participation in platform governance can be both empowered and constrained given the dual nature of social media platforms as both public-oriented spaces and profit-driven businesses. We explicate this framework in the following two sections.
User-initiated, expression-oriented platform governance framework.
User Empowerment in Platform Governance
Social media platforms facilitate the articulation of collective user voices about their platforms by providing users with a networked public sphere to engage in public discussion (van Dijck, 2013). Research has documented the role of social media platforms in facilitating action potential by connecting individuals from various backgrounds through networks and removing the capital barrier for content creation and circulation (Boyd et al., 2010). At the heart of the connective action thesis advanced by Bennett and Segerberg (2013) is the organizing capacity of personalized expression in online networks, which draws societal attention quickly. #BlackLivesMatter and #MeToo movement activists, as well as conspiracy theory believers and extremists, have capitalized on this organizing capacity of social media (Cinelli et al., 2022; Freelon et al., 2018; Suk et al., 2021).
In the context of platform governance, users can actively and collectively articulate their voices related to a platform, such as concerns and opinions about the platform’s openness, the health of the platform environment, and whether the platform’s algorithms accurately represent and protect their voices and identities (Bhandari & Bimo, 2022; Bucher, 2018). Much like user-driven social movements on social media, through what we call the power of opinion, users wield the potential to shape platforms. A platform may pay attention and alter its decision-making related to designs and policies in response to users’ voices, especially when the voices are expressive and unanimous (van Dijck & Poell, 2013). A notable example is the #deleteFacebook movement that emerged in the wake of the 2018 Cambridge Analytica scandal. This user-initiated movement engaged over 90,000 participants, eventually leading to significant policy changes within both government and corporate structures (Mills, 2021). Another example is Facebook’s real-name policy that required users to register under their real names, which faced controversy due to disproportionate discrimination against marginalized communities such as lesbian, gay, bisexual, transgender, queer or questioning, or another diverse gender identity (LGBTQ+) populations and Native Americans, cultural insensitivity, and privacy concerns (Haimson & Hoffmann, 2016). In response to user criticism and protest, Facebook amended its policy by allowing users to provide additional context when their accounts are flagged for using false names (CBC/Radio Canada, 2018). These instances demonstrate how user feedback can influence a platform’s policies and practices.
Similarly, in the case of Elon Musk’s purchase of Twitter, user responses to the significant ownership change, although reactive and expressive in nature, represent a form of user participation in platform governance. By expressing their endorsement, enthusiasm, skepticism, or opposition to the acquisition, users have the potential to influence the platform’s policy and operation directions. Early evidence suggests that Twitter users have indeed played a role in shaping the changes that Musk implemented and in pushing him to adjust his plans accordingly. For example, when Musk proposed charging verified users $20 per month for their blue checkmarks, many users, including influential figures like Stephen King and Nate Silver, expressed their outrage and dissatisfaction. These reactions ultimately led Musk to reduce the proposed fee to $8 (Hibberd, 2022), demonstrating how users are empowered to voice their opinions and how platforms are open to implementing user-driven changes in response.
In addition to directly engaging in discussion to organize and project collective voices on platforms, users exercise agency “by ‘voting with their purse,’ by preferring diverse platforms over others” (Helberger et al., 2018, p. 4), which can be considered users’ “ultimate leverage” (van Dijck, 2013, p. 33). By mirroring traditional lifestyle politics that center around expressing one’s values through consumption choices—such as boycotts and buycotts of consumer brands (Bennett, 1998)—users can wield their power by strategically activating or deactivating their accounts in reaction to platform changes. We view this dynamic as the power of numbers, which stems from the critical importance of user activity and engagement for platform profitability and growth. Activity and engagement metrics are routinely monitored by platforms because user data underpin platform capitalism (Sadowski, 2019). Therefore, a platform’s user base and engagement levels are directly linked to its profitability, shaping strategic decision-making (Srnicek, 2017) and influencing platform governance and future directions. For instance, in 2008, Flickr’s updated design intended to compete with Facebook and Twitter led to user disengagement and vocal opposition to the change: users left the platform or actively campaigned against features they disliked, showcasing a collective effort to preserve the platform’s original community spirit (van Dijck, 2013).
In the case of Musk’s Twitter purchase, Elon Musk paid close attention to the number of active users on Twitter both before and after his acquisition (Reuters, 2023), highlighting the economic significance of such metrics. In response to waning user engagement under Musk’s ownership, as some users sought alternatives like Mastodon and Threads (Perez, 2023; Vallance & Clayton, 2023), Twitter introduced new features and subscription models aimed at boosting user interaction and wooing content creators (Roth, 2023). These examples demonstrate the critical role of user data in shaping the operational and strategic decisions of social media platforms, as they adapt to shifts in user behavior and strive to maintain growth and profitability in a dynamic digital landscape.
User Constraint in Platform Governance
As discussed above, the public-oriented nature of platforms facilitates the expression of users’ platform-related opinions, and the profit-oriented nature of platforms gives users the leverage to influence platform decisions by promoting or resisting their use. However, platforms’ inherently open and contested nature and platforms’ economic model simultaneously counter opportunities for users to engage in governance.
First, social media’s openness and connectedness facilitate the expression of a multitude of voices on a given issue, which embodies the spirit of deliberative democracy (e.g., Mutz, 2006) yet can potentially weaken the crystallization of collective voices on platform governance and undermine collective decision-making capabilities. Within most platforms, users exhibit diverse and at times conflicting values; rules considered legitimate by some are seen as unwarranted impositions by others (Gillespie, 2018b). Research shows that users are embedded and interacting within stable echo chambers, resulting in drastically different opinions on the same issue from various networks (Zhang et al., 2022). This pattern of opinion expression is similarly reflected in social media activism efforts like the #MeToo movement. While social media platforms allowed survivors of sexual violence to share their stories and extend social support, such visibility was met with heightened contention among #MeToo activists and anti-feminists alike (Suk et al., 2023).
The political nature of platforms (Gillespie, 2010), often connected with platform policies, endorsement from political figures, and political perceptions of platforms, likely encourages the participation of opinionated partisan users. Other factors such as technological literacy (Davis, 2020), racial identity (Noble, 2018), and user intentions (Bucher, 2018) also contribute to varying levels of awareness and motivation among users to engage in platform governance. With users’ diverse interests and motivations behind platform use, achieving a coherent voice to influence platform governance may be unattainable. Taken together, the open, contested, and uneven nature of social media suggests that users’ opinions on a platform might not be unanimous but instead feature a plethora of positions from the most vocal users who seek to steer the platform in different directions.
Second, increased voices on platform regulations and governance, even if critical, establish an ironic relationship with the platform by furnishing them with increased user data. As documented by Scharlach and colleagues (2023), “for social platforms, expression is simultaneously social and profitable” (p. 15). From the perspective of platform economics, allowing controversies of any type to simmer and explode can drastically increase user volume and engagement, at least in the short term, thereby enhancing platform profitability. This is because a platform with more active users generates data larger in volume and richer in content. This dynamic explains why companies like Facebook have been slow to contain the spread of misinformation (Flew, 2021). Since false narratives tend to proliferate faster than truthful ones (Vosoughi et al., 2018) and controversial content often attracts more engagement (Rathje et al., 2021), platforms may inadvertently benefit from these dynamics, which can delay corrective actions. Even if the content carries critical voices against a platform, enhanced attention and activity may still translate into platform profitability. As Wells and colleagues (2020, p. 665) note, in the attention economy, “quantity of attention, rather than its quality” often prevails, making “any attention . . . positive attention.”
Moreover, the hidden layers of technology, such as data and algorithms, are the key to business success but often remain opaque to users. However, these coded structures considerably shape the nature of content visibility and user interactions on social media platforms (van Dijck, 2013), leading to the power asymmetry between platforms and users (Delacroix & Lawrence, 2019). This lack of transparency means that certain user activities, like silently leaving a platform, may not be visible or may be underreported. Social media platforms may obscure or misrepresent these facts when reporting user metrics. In addition, platform algorithms can suppress the visibility of voices opposing their company and ownership values, further constraining user voices (Noble, 2018; O’neil, 2017).
Twitter Users’ Response to Musk’s Twitter Purchase
Twitter users’ reactions to Elon Musk’s acquisition of Twitter provide an appropriate window into understanding how users respond to platform change and shape platform governance through their responses. Understanding these reactions offers empirical insights into how users attempt to contribute to platform governance through expressions, debates, and conversations, which illuminates the empowered and constrained nature of their participation.
Extant evidence underscores the diversity of reactions from different user groups, each response being calibrated to align with their respective interests. Musk’s robust advocacy for free speech has resonated with conservative users, translating into increased followers of Musk and Republican politicians (Safak & Sridhar, 2022). Conversely, users who disapprove of Elon Musk’s purchase might have chosen to leave the platform or reduce their activity, as Elon Musk and Democratic politicians lost liberal followers (Safak & Sridhar, 2022). Increases in hate speech and changes in content moderation policy have therefore resulted in a segment of journalists, academics, and celebrities migrating to alternative platforms (Hussain & Healey, 2022).
Focusing on the volume of users and their activity level before and after Elon Musk’s Twitter purchase, we aim to gain insights into how different groups of users potentially promoted or challenged Twitter’s business interests by affecting the amount of user-generated data. Examining the content of platform-related discourse during the Twitter purchase saga, we provide insights into multifaceted user voices in response to Twitter’s ownership change. If user voices are relatively in unison or/and if dissenting voices can impact platform economics, users’ leverage in platform governance can be substantiated. Conversely, if user voices are excessively contentious or/and if contention can be co-opted by platforms, users’ attempts to impact platform governance might be undermined. These theory-driven considerations translate into two research questions:
RQ1. How did different groups of Twitter users respond to Musk’s takeover in terms of user volume and activity level?
RQ2. How did public discourse related to Musk’s takeover shift thematically over time across different user groups on Twitter?
Method
Data Collection
To maximize data coverage, we collected all tweets containing keywords associated with Elon Musk and/or his Twitter purchase (see full list and rationale in Supplemental Appendix A) using Twitter’s academic Application Programing Interface. This was done for two time periods, totaling 92,728,954 tweets. The first phase (T1) from 26 March 2022 to 30 April 2022 represents the proposal stage of Musk’s acquisition, with data collected in May 2022. On 26 March, Elon Musk discussed the future of social media with Jack Dorsey. On 5 April 2022, Elon Musk announced that he bought over 9% of Twitter’s shares, and on 14 April, he offered to buy the platform. These moves sparked heated discussion on Twitter about Musk’s takeover and stimulated speculations and vows to stay, leave, or return to Twitter. The second phase (T2) from 20 October 2022 (a week before Musk’s takeover) to 30 November 2022 covers the formal acquisition, and the data were collected in early December 2022. Musk officially became Twitter’s owner on 27 October 2022, following a series of legal maneuvers that ignited another round of debate about the privatization of the public sphere and content moderation policies. These two phases encapsulate the two pivotal moments surrounding Elon Musk’s takeover of Twitter, when the attention to the event and related discussions was especially pronounced. A descriptive summary of the data set is presented in Table 1.
Summary of Tweets and Users in T1 and T2.
Note. The total tweets were calculated after eliminating all duplicate tweets and non-English tweets.
User Cluster Analysis
To explore the various user categories who participated in Musk’s Twitter acquisition discussions, we applied cluster analysis. Given that retweets often represent endorsement (Flamino et al., 2023; Jiang et al., 2020), we relied on the retweet relationship to identify groups of similar users (Guerrero-Solé, 2017). Specifically, we used the Vintage Sparse Principal (VSP) component analysis, a clustering technique for estimating latent factors in sparse and multivariate data (Rohe & Zeng, 2020), to cluster users based on their retweet information. A recent study using this method to identify collective identities through the retweeting network notes its effectiveness and scalability as compared to other methods like the Louvain methods (Shuster et al., 2024). With a significant user overlap between T1 and T2 (21% of all users), we conducted a cluster analysis of all users across the time points combined. Based on the scree plot that displays eigenvalues of principal components, we set the cluster number to 69 to balance computational efficiency and cluster granularity (Supplemental Appendix B).
Two authors independently analyzed each of the 69 clusters of (retweeting) users by referring to the most frequently retweeted users and the most representative (determined by the cluster loading coefficient) users in each cluster. The authors reached a consensus on cluster interpretation after multiple rounds of discussion. In addition, given that bots (i.e., automated accounts) are likely agents in social media discourse (Bessi & Ferrara, 2016) and that eliminating them was allegedly the primary inspiration behind Musk’s Twitter buyout, we implemented bot detection among the most retweeted users (N = 1,366) and the most representative users (N = 1,161) across 69 clusters. The bot detection was executed using Botometer v4 (Sayyadiharikandeh et al., 2020), a widely used tool implemented in previous studies (e.g., Wang et al., 2024). The program returns values from 0 to 1 for each user, with a higher value indicating greater bot likelihood. However, Botometer was unable to assign scores to 178 of the most retweeted and 183 of the most representative users. This was due to the account availability issues at the time of bot detection, as some of these accounts may have been suspended or deleted, while others may not have been publicly available. Consequently, our analysis focused on users with available Botometer scores, totaling 1,188 most retweeted and 978 most representative users. We established a threshold of 0.6 to identify potential bots, based on their bot-likelihood scores across clusters for both sets of users. Despite Botometer’s widespread usage, its accuracy has been questioned (Gallwitz & Kreil, 2022; Rauchfleisch & Kaiser, 2020). Therefore, two authors further validated our findings by manually coding a random 10% sample of unique users (N = 147), a process that shows an 84.35% accuracy rate and is detailed in Supplemental Appendix B.
The cluster coding and bot detection processes resulted in six larger cluster categories: conservatives, liberals, cryptocurrency enthusiasts, miscellaneous, potential bots, and irrelevant. Irrelevant clusters were removed from the subsequent analysis. One author further inspected the cluster analysis results by manually analyzing a random sample of 200 users (referring to their profile information, the users they retweeted, and the retweeted content), resulting in an overall accuracy of 0.78. The full list of clusters and their descriptions and validation are available in Supplemental Appendix B.
Topic Modeling of Tweets
We employed Latent Dirichlet Allocation (LDA) topic modeling, which is one of the widely used unsupervised machine learning methods for analyzing texts to discover the latent topics present in user discussions regarding Musk’s acquisition of Twitter (Blei, 2003). Given the substantial volume of tweets, we randomly sampled 10% of tweets from each user cluster category, which yielded 7,573,895 total tweets and 2,665,342 unique tweets. For topic modeling, we focused exclusively on the unique tweets. Following the standard textual analysis procedure (Maier et al., 2018), we pre-processed the unique tweets by removing stop words, punctuations, numbers, and symbols and converting all texts to lowercase. In addition, we removed words appearing in more than 90% or less than 0.002% of the corpus of unique tweets (Grinberg et al., 2019). Next, to find the optimal number of topics, we further sampled 10% of the unique tweets in our sample, and then iterated over a range of potential topic numbers (K, from 50 to 200, in the skip of 10, see Barberá et al., 2019). Based on both perplexity and log-likelihood scores, we determined 130 to be the most optimal number of topics.
Two authors independently examined the top words and the associated top tweets to interpret each topic, and then any disagreements were addressed through discussion. An initial set of 130 topics was assigned to a topic label, and those that fell outside our research scope (mixed/others) were excluded, which led to a selection of 97 topics for further analysis. These 97 topics were subsequently grouped into 12 distinct topic categories, which fell under five thematic domains (following the practice in Kligler-Vilenchik et al., 2020): platform-centered, Musk-centered, politics, cryptocurrency, and mixed/other. Mixed/other was excluded from further analysis. Supplemental Appendix C presents the full list of topics and top terms, as well as detailed explanations about the topic, category, and thematic domain generation process.
Results
Shift in the Level of Activity Overtime Across Different User Cluster Categories
Table 2 describes the four categories of users. Based on the VSP result, 92.8% of users were identified as partisan–either liberals or conservatives, who actively retweeted politicians, partisan media outlets, and political pundits, and often explicated endorsed or incorporated political terms and hashtags in their tweets. Among these partisan users, conservatives (N = 4,829,209, 59.78%) substantially outnumbered liberals (N = 2,667,354, 33.02%). Cryptocurrency enthusiasts comprised 6.6% of the user base. In contrast, miscellaneous users and potential bots constituted a mere 0.18% of the total user population.
Summary of User Cluster Categories.
Note. The irrelevant category has been excluded from subsequent analyses.
RQ1 explores user volume and activity level changes across different user cluster categories. For user volume, we examined the daily total user count per category. For activity level, we focused on daily tweet count per user per category. Together, these two measures provide a comprehensive understanding of the amount of user data generated. We compared them between two time periods—T1 (pre-acquisition) and T2 (during and post-acquisition). Analytically, recognizing that 21% of the users engaged in discussions during both T1 and T2, we applied a paired t-test to this subset of recurring users. For the remaining non-recurring users, who appeared only in T1 or T2, we employed Welch’s two-sample t-test. Recurring users serve as panelists that allow us to understand changes in potentially more committed users, whereas non-recurring users allow us to assess shifts in user base across T1 and T2. We focus on the significant relationships (p < .05) with a non-trivial effect size (Cohen’s d > .2 or <−.2), given the large sample size of our data set. Supplemental Appendix D provides the description of all significant relationships.
We first compared daily total user counts in each cluster category between T1 and T2, for non-recurring users. Note that this analysis does not apply to the recurring users because they appeared in both periods. The Welch two-sample t-test results revealed a significant increase in liberal users during T2 compared to T1, t(66.43) = 3.05, p < .01, Cohen’s d = .69. Similarly, the daily number of potential bots witnessed a significant increase, t(78.78) = 2.06, p < .05, Cohen’s d = .46, whereas the daily number of cryptocurrency enthusiasts decreased, t(79) = −3.26, p < .01, Cohen’s d = −.72. There were no significant user volume changes among conservative and miscellaneous users.
Next, we compared daily tweet counts per user in each cluster category between T1 and T2. For recurring users, paired t-test results demonstrated that only liberals, t(656,732) = 179.17, p < .001, Cohen’s d = .22, posted significantly more per day in T2 than T1, suggesting a noteworthy impact of the takeover on tweet counts within this category. For daily tweet counts per user among non-recurring users between T1 and T2, the Welch two-sample t-test revealed that miscellaneous users, t(8,372.4) = 18.903, p < .001, Cohen’s d = .41, and potential bots, t(6,164.4) = 9.07, p < .001, Cohen’s d = .22, exhibited a significant increase in daily tweets. The detailed t-test results are available in Supplemental Appendix D.
Shift in Thematic Domains Over Time Across User Cluster Categories
RQ2 asks about the shift in the content of tweets over time among various user groups. We first describe the four core thematic domains from the topic modeling results: platform-centered, politics, Musk-centered, and cryptocurrency (see Table 3).
Description of Thematic Domains.
The platform-centered thematic domain, which comprised approximately 45% of total tweets, focused on platform governance and expected changes following Musk’s acquisition of Twitter, expressing opinions over content moderation, regulation, and data privacy, while also anticipating or opposing new features. The top hashtags widely circulated in this domain further corroborate this, such as #twitterstillopen, #leavingtwitter, #freespeech, and #riptwitter (Table 4). This thematic domain most directly represents users’ unambiguous attempt to contribute to the governance of Twitter. Some called for returning freedom of speech to the platform and criticized past platform governance, which was often linked to reinstating Trump’s account as well as calls to maintain freedom of speech at all costs. Some tweets talked about the censorship of the political right and so-called “cry babies” on the political left, while others took a critical approach towards Musk’s influence and Twitter takeover, such as issues with absolute freedom of speech. For example, one user writes, “@elonmusk look at the manipulation divisive tactics. You sure are a toxic cretin, bring back Trump!!” Others highlighted the harmful aspects of unrestricted freedom of speech, including “@laurenboebert @elonmusk re-phrase that for the Republican party in your case it’s:—free lies—free racism—free hate speech—free anti-women speech—free anti-gay speech—free encouragement of criminals speech—free gun speech that are killing kids daily. Your free speech is filth.” Other discussions included appeals to add an edit button and change profile page features as well as general discussions on bots on Twitter. For example, one user writes, “@elonmusk he has to determine how many real people, how many bots, how many fake accounts, how many people with multiple accounts. A Twitter poll is a dumb idea.” One of Musk’s primary claims during his acquisition is that too many bots are on Twitter, aiming to remove them post-takeover.
Top 30 Hashtags from Each Thematic Domain.
Moreover, broader conversations in the platform-centered domain discuss cancel culture, diverse opinions, and the importance of access to information from both ends of the political spectrum, with specific calls for Musk to implement changes. Conservatives often mentioned liberals and their reactions to Musk’s acquisition. For example, one user wrote, “Liberals around the world ranted that if Elon Musk buys Twitter, they will quit Twitter. Not one has quit Twitter.” More frequently, conservatives discussed claims of limited freedom of speech and advocated for free-speech absolutism. Another user stated, “To prove my point that Twitter is still a problematic app with zero positive changes since all the hype about @ElonMusk buying Twitter, scroll through replies on newer AOC posts. Conservative replies are deleted and/or disallowed. I posted a reply yesterday and it has already been deleted.” On the contrary, liberals most often tweeted about misinformation and the issues with absolute freedom of speech. For instance, one user wrote, “@TedCruz @ElonMusk @Twitter Breaking news: Facts matter. Eight of the top ten states receiving free socialist federal aid are red states. Eight of the top ten states with the highest homicide rates are red states. Disclaimer: Facts, not bots. #VoteBlueIn2022 Had enough of supporting red? Go get a job!”
The second thematic domain focused on politics (21.01% of total tweets), branching into various political and politicized topics, including elections, global politics, the economy, social movements, climate change, the Ukraine–Russia War, lesbian, gay, bisexual, and transgender (LGBTQ) issues, cancel culture, racial tensions, and vaccine efficacy. Relatedly, discussions included Musk’s excessive wealth and political leanings and his ability to control social media platforms as a result. In another example, there were intense discussions challenging Musk’s outspoken anti-transgender views.
The third thematic domain pertained specifically to Elon Musk himself (15.7% of total tweets), encompassing both positive and negative sentiments toward him (including well wishes and words of encouragement from supporters as well as sarcasm among his critics), evaluations of his personality, and discussions related to his businesses (such as critiquing Musk for being a fraud and incompetent in the business world). In addition, some topics within this domain addressed his business initiatives to acquire Twitter. Given our data collection keywords during the two critical time periods of Musk’s Twitter acquisition, we argue that the Musk-centered domain, while not as closely related as the platform-centered domain, is still relevant to platform governance in this specific context. As shown in Table 4, the top hashtags in this domain also included #twittertakeover, #freespeech, and #elonmuskbuytwitter, suggesting the relevance of platform governance discourses. Our qualitative reading of tweets in this domain further supports our argument. For example, in the Musk-centered domain, a user criticized, “@elonmusk you’re an idiot. You don’t know what you’re talking about but are pretending you do.” Although this comment appears unrelated to platform governance, it may reflect criticisms of Musk’s policies on Twitter. Another tweet stated, “@elonmusk don’t act stupid, you’re probably not. If that’s your workday, you’re not contributing to the business. We know this to be true, they’re fired and Twitter is still working just fine.” This tweet criticizes Musk but also relates to the mass layoffs at Twitter following his acquisition of the company.
Finally, the final thematic domain concerned cryptocurrencies (8.39%), reflecting Musk’s known interest and substantial influence in the cryptocurrency market (Ante, 2023).
Combining user analysis and computational content analysis results, Table 5 illustrates the diverging thematic prevalence across different user cluster categories. Both conservatives and liberals primarily participated in discourses focused on the platform, Elon Musk, and politics. Other user groups exhibited expected patterns of interest. For example, cryptocurrency enthusiasts primarily participated in cryptocurrency-related discussions while showing limited engagement with other themes. The miscellaneous category, consisting of a diverse group of general Twitter users, manifested a relatively balanced distribution across all domains. Meanwhile, discourse within the potential bots category predominantly fell into the mixed/other domain, aligning with the understanding that automated accounts tend to generate topically mixed or inconsistent messages (Wang et al., 2024).
Distribution of Thematic Domains Across Different User Cluster Categories.
Note. The mixed/other thematic domain has been incorporated to illustrate the distribution among diverse user clusters. However, this domain was excluded from the actual analyses.
To examine the thematic emphasis that distinct user cluster categories placed on various domains following the change in Twitter’s ownership (RQ2), we employed the same analytic approach with our user-level analyses to compare average daily tweet counts from each user cluster category under each thematic domain.
Paired t-test results for recurring users indicated that liberal users, t(155,063) = 124.20, p < .001, Cohen’s d = .32, exhibited a significant increase in platform-centered discussions in T2 compared to T1, whereas conservative users shifted attention away from Musk-centered conversations, t(99,342) = −63.91, p < .001, Cohen’s d = −.20. Regarding potential bots, a notable transition was a shift to political, t(86) = 6.57, p < .001, Cohen’s d = .70, and platform-centered, t(54) = 2.14, p = .037, Cohen’s d = .29, topics in T2 compared to T1. Miscellaneous users displayed a greater interest in political topics, t(63) = 2.93, p = .005, Cohen’s d = .36, and a reduced interest in Musk, t(119) = −3.00, p = .003, Cohen’s d = −.27, and crypto-centered, t(78) = −2.18, p = .032, Cohen’s d = −.24, topics during T2 compared to T1.
Welch’s two-sample t-test for the differences in thematic focus among non-recurring users indicated liberals engaged in cryptocurrency less, t(12,936.90) = −15.90, p < .001, Cohen’s d = −.26. A marked shift was observed among potential bots; similar to the trend noted in recurring users, their thematic attention moved from cryptocurrency, t(144.87) = −2.24, p = .03, Cohen’s d = −.26, to political, t(786.12) = 13.54, p < .001, Cohen’s d = .79, platform-centered, t(348.02) = 2.54, p = .011, Cohen’s d = .24, and Musk-centered, t(189.90) = 4.15, p < .001, Cohen’s d = .43, discussions. The full results of t-tests can be found in Supplemental Appendix D, and the summary of the results is presented in Table 6.
Summary of Significant Relationships (p < .5) With Small to Large Effect Sizes (Cohen’s d >= .2).
Discussion
Extending upon existing platform governance research that often undertheorizes the role of users, this study introduces a framework for understanding the user-initiated, expression-oriented form of platform governance, that is, how users participate in platform governance by leveraging platform affordances and spaces in a bottom-up, leaderless, and less organized fashion. This framework, which is less studied compared to the platform-initiated, more organized ways for users to participate in governance processes, integrates user agency with platform affordances and platform political economy. We conceptualize that active users’ opinion expression and level of activity constitute a form of user-initiated participation in shaping platform governance. Our theorizing of the dual nature of social media platforms as both public-oriented spaces and profit-oriented businesses suggests that platforms can both empower and constrain social media users in their contribution to platform governance.
Situating our analysis within the Twitter acquisition by Elon Musk, we investigate how Twitter users responded to this ownership change pre- and post-acquisition and whether their response could constitute as an effective form of user-driven platform governance. Combining user cluster analysis and topic modeling to conduct both user-level and content-level analysis, we demonstrate how the dual nature of platforms leads to the inherent asymmetric power dynamics between platforms and users (Delacroix & Lawrence, 2019; Gillespie, 2018b). Through this way of participating in platform governance, users are more constrained than empowered.
First, our findings show that social media platforms indeed provide open spaces for users to participate in platform governance by allowing them to express their voices and modulate their activities. However, this open space, characterized by a diverse user base, can dilute the power of opinion due to the variety of values and perspectives presented. As shown in our data, although the majority of conversations (about 45%) were directly about Elon Musk’s plan to buy Twitter and his actual acquisition (the platform-centered thematic domain), they were also accompanied by divergent topics that may be less relevant, such as cryptocurrency or politics. Even within the platform-centered thematic domain, we show that it included a range of user groups from those with strong political preferences to cryptocurrency enthusiasts and automated agents, and the levels of participation among these users were uneven. Not all users were equally concerned about platform governance due to their varied interests and reasons for using the platform (Davis, 2020; van Dijck, 2013). This finding indicates that users who were highly attentive and responsive to Musk’s activities and decisions during these periods were predominantly partisan, which is aligned with previous research demonstrating the politically contentious nature of platform governance (Gillespie, 2010; Helberger et al., 2018). The results also suggest that even the same group of users discuss different topics, making a collective voice even more challenging to achieve. For example, partisan users, although mostly focused on platform-centric topics, occasionally dabbled in politics and cryptocurrencies as well.
The impacts of user-driven platform governance can be further weakened due to the profit-oriented nature of platforms, which renders any voices, including critical ones, into profitability (Scharlach et al., 2023). In this user-initiated and expressive form of platform governance, users must project their opinions about a platform via the platform (van Dijck et al., 2018), and for platform owners, content is synonymous with data, and the sheer volume of data streams is of paramount interest (Srnicek, 2017). This is particularly evident among liberal users, as observed in our study—expressions of discontent on Twitter can inadvertently contribute to the platform’s profits. Specifically, there was a significant increase in daily total users among non-recurring liberals, suggesting an influx of unique users chiming in on related topics during and after the acquisition. Similarly, recurring liberals exhibited a marked increase in their daily posting. This surge in recurring liberal user activity mirrors the failure of the #deleteFacebook movement, where widespread vocal boycott did not translate to a mass exodus (Perrin, 2018), instead presenting an alternative model of platform capitalism (Mills, 2021). This can be attributed to Twitter’s significant network effects (Flew, 2021; van Dijck, 2013), and the fact that outrage is more likely to drive content generation than affinity (Brady et al., 2021). Liberals most often criticized Musk’s takeover and post-takeover decisions, which we observed in the thematic shift among recurring liberals toward platform-centered topics during and after the acquisition. This shift primarily concentrated on challenging the free-speech policy and advocating for more robust content moderation by both Twitter and governmental agencies. Nonetheless, regardless of user intentions and the content, the increase in user volume and activity translates into valuable data for the platform (Langley & Leyshon, 2017), inadvertently contributing to Twitter’s business success, at least in the short term.
Furthermore, the profit-driven nature of the platform determines that the platform, including its owners, ultimately retains control over the platform’s digital architecture, algorithms, and policies. This constrains users’ ability to effect meaningful change through their expressive activities alone, especially in the short term. As shown in our findings for partisan users, despite engagement from both sides, liberals and conservatives expressed competing values. Conservatives endorsed the changes on Twitter following Musk’s takeover, whereas liberals expressed concerns, dissatisfaction, and even outrage toward the changes under new ownership and leadership. This is further supported by the findings that conservative users demonstrated a constant interest in Musk’s acquisition over both periods, with their numbers nearly twice those of liberals. This divergence can be attributed to the potential for a new platform ownership to favor specific political values and ideologies through platform design, policies, and algorithms (Gillespie, 2010). Elon Musk’s persona as a free-speech absolutist ignites hope among conservatives that Twitter under his ownership could favor Republicans and conservatives who believe that the platform has been suppressing their views (Anderson, 2023; Rebecca & McGraw, 2022). As a result, conservative users have shown a heightened interest in the acquisition, actively participating in discussions and expressing their expectations for the platform’s future. The divergence in opinions might provide platforms with opportunities to selectively represent one side’s voice over time (Davis, 2020), thereby making user-driven governance more challenging. As users become increasingly divided along ideological lines, the potential for meaningful, consensus-driven change diminishes leaving the platform with greater control over its governance decisions.
Taken together, these findings underscore the power asymmetry between platforms and users in the context of user-driven expression-based platform governance. Given the expressive nature of user-initiated efforts to participate in platform governance as well as the dual nature of platforms—both public and profit-oriented—the consequences may be inherently slow, or even, weak (Jang et al., 2023), especially when such efforts occur within platforms. This weakness is evident not only in the less organized, leaderless, user-initiated attempts for platform governance but also in platform-initiated and structured governance. For example, to combat misinformation, Twitter introduced “Birdwatch,” inviting users to identify misleading tweets by writing notes. However, research shows that despite Twitter’s initiation and support of community-based fact-checking efforts, these initiatives encounter difficulties due to diverse opinions and polarization within the user community (Pröllochs, 2022). An important reason for somewhat ineffective user input (through both platform-initiated and user-initiated ways), as we theorized, is that social media platforms are open spaces for diverse opinions and voices, making collective and strong voices difficult to achieve. Efforts such as those proposed by Bennett and Segerberg (2013) can have observable effects only by sustaining and building strength over time due to their personalized rather than organized nature. Therefore, while users are empowered to project their voices by leveraging the existing platform affordances, immediate and tangible effects may be hard to observe. This does not imply that such efforts are meaningless in the long term; rather, it highlights that users are more constrained than empowered in a bottom-up governance process. Nevertheless, active users can continue to raise awareness about platform governance issues through sustained and collective efforts over time (Bennett & Segerberg, 2013). On the contrary, users’ active engagement and vocalization of their demands through the platform ultimately become data, generating revenue for the platforms (Srnicek, 2017). Moreover, platform architectures are engineered to control user agency, which often remains opaque to the users (Gillespie, 2018a; van Dijck, 2013), making critical voices risk oppression by algorithms (Noble, 2018) and potentially become the primary source of revenue in the platform economy (van Dijck et al., 2018).
One additional finding, while not directly related to user-driven platform governance, highlights the limitations and challenges of self-governance as driven by platform owners. Specifically, our results reveal notable participation from potential bots in terms of both user activity levels and thematic focus. During the acquisition phase, there was a significantly higher number of accounts and the level of activity in terms of daily tweet counts than pre-acquisition. In addition, bots exhibited an increased engagement with politics-related topics during the acquisition. These patterns are consistent with previous research on the behavior of automated accounts, where a small number of such entities can generate a disproportionately large volume of posts (Woolley & Howard, 2018) as well as the weaponizing potential of bots to politicize the public sphere and divert online discussions (Broniatowski et al., 2018). One plausible explanation is that the acquisition phase in our data set coincides with the 2022 US midterm election, suggesting that bots may have been deployed to disseminate political content aimed at influencing public opinion (Bessi & Ferrara, 2016). Paradoxically, although the eradication of bots was one of Musk’s most emphatically stated commitments during the Twitter acquisition, our findings suggest the contrary, at least in the short term.
We acknowledge the limitations of our research. First, our findings are based on one case study specific to the US context and English content only. Given the global influence of the platform, Twitter’s ownership and policy change likely elicited responses from non-US regions and non-English speakers. We invite future researchers to expand upon these inquiries to encompass multiple platforms and diverse settings in subsequent studies. Second, our data exhibit pronounced temporal trends in response to Musk’s acquisition activities (Supplemental Appendix Figure S7); yet, we did not explore the time series relationship between Musk’s actions and user responses. Future research should investigate the varied reactions of distinct user groups to Musk’s interventions. Third, it is important to note that as platforms typically do not disclose subtle participation data, such as user exposure and screen time, our analysis is limited to examining the more active, expressive, and visible aspects of participation in platform governance. Future research should employ more advanced approaches (e.g., Araujo et al., 2017; Christner et al., 2022) to capture latent forms of participation and offer a deeper understanding of user participation in platform governance. Last but not least, we did not investigate the direct impacts of this type of platform governance attempts. On one hand, since this study aims to call attention to the form of user-initiated expression-oriented platform governance, examining its effects are beyond our scope, due to the nature of the impact that is hard to quantify. Nonetheless, early evidence suggests that Musk introduced the subscription feature in response to decreased user engagement (Roth, 2023). On the other hand, these effects might not be immediately observable or could be ineffective unless users persistently push their efforts (Bennett & Segerberg, 2013). We encourage future research to use our study as a starting point, focusing on the role of user agency, to conduct further investigation into the effects from a long-term perspective.
Despite the limitations, our study contributes to existing literature on platform governance by theorizing the user-initiated expression-oriented form of platform governance. Previous research has primarily examined the users’ role through more organized and top-down platform-driven approaches, this study broadens the understanding of bottom-up efforts initiated by users in platform governance. Grounded in the dual nature of platforms, our framework theorizes the paradoxical situation users face when expressing their opinions on platform governance via platforms. Our empirical findings show that users are more constrained than empowered in this context. Nonetheless, we note that the collective power of users should not be underestimated, given its potential to shape the nature, characteristics, and operational directions of the platform if sustained over the long term. As users continue to engage critically with platforms and advocate for their rights and interests, they can play an increasingly important role in driving positive changes and holding platforms accountable. These insights further offer practical implications for social media users and platforms. As summarized in Figure 1, user participation in platform governance might be more effective if users can approach it from an ethics perspective rather than an identity perspective. And social media platforms need to provide more transparent and conducive avenues to facilitate this process.
Supplemental Material
sj-docx-1-sms-10.1177_20563051241277606 – Supplemental material for Empowered or Constrained in Platform Governance? An Analysis of Twitter Users’ Responses to Elon Musk’s Takeover
Supplemental material, sj-docx-1-sms-10.1177_20563051241277606 for Empowered or Constrained in Platform Governance? An Analysis of Twitter Users’ Responses to Elon Musk’s Takeover by Rui Wang, Yini Zhang, Jiyoun Suk and Sara Holland Levin in Social Media + Society
Footnotes
Data Availability
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental Material
Supplemental material for this article is available online.
Notes
Author Biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
