Tweeting terrorism: Vernacular conceptions of Muslims and terror in the wake of the Manchester Bombing on Twitter

ABSTRACT Both vernacular security studies and critical terrorism studies (CTS) offer constructivist analyses of security couched in understandings of security speak. However, neither adequately take account of the ways in which social media presents important opportunities for greater insight into how terrorism is constructed. This study analyses tweets posted after the 2017 Manchester bombing, exploring how jihadist terror attacks are constructed on social media. To do this, we combine social network analysis, as a sampling method, with discourse analysis. The study finds that Twitter provides a platform for diverse terrorism discourses to be expressed and contested. This indicates a literate lay audience within post-attack narratives, self-aware of dominant social constructions of “Muslim terrorism”. Indeed, it suggests an audience that, on Twitter, is hardly only audience but seeks to speak security itself. Insights are gleaned with respect to depicting, defending, and critiquing Muslims, constructing what it means to be a terrorist, portrayals of victimhood, and how terror events feed into broader critiques of “political correctness” and “liberal” politics. Therefore, the analysis also provides further insights into the portrayal and (self-)positioning of Muslims in the wake of a jihadist attack and nuances accounts of Muslims’ securitisation qua terror.


Introduction
On the 22 nd of May 2017, a British man, Salman Abedi, detonated a suicide bomb in the foyer of the Manchester Arena as people were leaving a concert by pop singer Ariana Grande. 23 people died, including the attacker, and 139 were wounded. On May 23 rd , ISIS claimed responsibility for the bombing (Dearden 2017). It was the deadliest terror attack in the UK since the 7/7 bombings in 2005. Like 12 years earlier, the attack was grieved, debated, and interpreted in traditional media and, later, by politicians, security services, and academics. However, unlike in the aftermath of past violence, the public response to the Manchester bombing was not left to politicians, "experts", and journalists alone. Lay users on sites such as Twitter immediately discussed the bombing. They interpreted the attack and (de)constructed discursive responses to it, even before security elites had fully and publicly processed the event. 1 This challenges traditional ways of studying and conceptualising the response to terror. It also nuances understandings of supposedly passive lay audiences of security.
We draw on critical terrorism studies (CTS) and vernacular security studies to contribute to ongoing conversations on the role of social media and lay audiences in understanding the construction of (in)security. Additionally, the case of the Manchester bombing offers important insights into the portrayal of Muslims online in the context of jihadist violence. Utilising social network analysis as a sampling method for our discourse analysis, we engage a novel combination of methods to explore how understandings of terror attacks and their perpetrators are constructed on Twitter. To address this broad question, we focus on discourse about "Muslim terrorism" in the aftermath of the Manchester bombing.
Constructions identified within this study are multifaceted and dynamic; they demonstrate that social media is one crucial context to study in order to understand how security speak is conceived of and contested. Insights are gleaned with respect to what it means to be a terrorist, portrayals of victimhood, and how terror events feed into broader critiques of immigration and "political correctness". The discourses analysed also challenge straightforward accounts of Muslims' securitisation in the wake of jihadist violence; besides the continued "suspectisation" of Muslims, users banalise them, defend them and their faith, anticipate and respond to Islamophobic discourses, and question the relationship between religion and terrorism more broadly. Twitter also provides a platform for users to speak as Muslims and denounce terrorism. However, the link between Muslims, Islam, and terrorism is both implicitly and explicitly reified across these discourses.

Conceptualising terror online: vernacular security and critical terrorism studies
Recent work in CTS, vernacular security studies, and security studies more broadly has begun to engage with social media. Yet, much of the current scholarship on terrorism online focuses on the "hard" aspects of the internet as a sphere for extremism and dissent (Aly et al. 2016;Rudner 2017). Critical and constructivist work provides a more nuanced picture. Recent studies have highlighted how social media can contribute to a sense of "togetherness" in the wake of terrorist violence (Merrill et al. 2020), provide a place for memes in resistance to threats (Downing 2020), and host discussions about the motivations of foreign fighters (Da Silva and Crilley 2017), among others. These studies illustrate how social media plays a role in documenting and discussing material, emotional, and socio-political responses to terrorism. More broadly, they underline the richness of online lay discourses and the continued need to expand their study (see Bogain 2020).
We argue that more attention should be paid to the immediate aftermath of a terrorist attack when elite statements are still scarce and coverage in print media is just emerging. While social media use in the aftermath of an attack has attracted attention in work on trauma, communications, social networks, and crisis management (inter alia Kessling et al. 2020;Eriksson 2016Eriksson , 2018Fischer-Preßler, Schwemmer, and Fischbach 2019;Steensen 2018), critical security studies have catching up to do in this regard. Twitter, with its plethora of real-time commentary and high user engagement after a terrorist event (Kessling et al. 2020;Steensen 2018), emerges as the most fitting platform to explore this. Eriksson (2016), in her work on terrorism and trauma, has also found it to be a place for "backchannel communication" that circumvents and criticises mass media discourses in the wake of terrorist violence (building on McNely 2009). This demands further attention from a critical and vernacular lens. Drawing on social network analysis to sample our tweets, we also leverage the networked nature of Twitter discourses, something that so far has received little attention in the critical security studies literature.

Critical terrorism studies
CTS seeks to understand the construction of terrorism and its embeddedness in everyday life (Erickson 2008;Holland 2011;Veloso and Bateman 2013). They have emerged out of a critique of the empirical, conceptual, and ontological weaknesses of conventional terrorism studies (Jackson 2007;Gunning 2007;Jarvis 2009;Breen Smyth et al. 2008), particularly their focus on "problem solving" (Jackson 2007;Gunning 2007). CTS adopts a constructivist stance vis-à-vis terror events (Jackson 2007;Gunning 2007;Jarvis 2009) and foregrounds that constructions are embedded in political and ideological debates (Burnett and Whyte 2005). This enables analysis of terrorism beyond its material realities situating terrorism as socially constructed.

Vernacular security: from audience to agents
Similarly constructivist, vernacular security studies conceptualises ways security is constructed in everyday terms and responds to the lack of understanding of how laypeople make sense of (in)security, including terrorism (Bubandt 2005;Jarvis and Lister 2013;Vaughan-Williams and Stevens 2015;Jackson and Hall 2016). It has sought to challenge and complement elite-centric understandings of security and instead prioritises the stories of those marginalised in accounts of global politics. Coming without set conceptions of what security is or indeed should be provides vernacular security studies with a theoretical "emptiness" that enables inductive insights into public understanding of, as well as anxiety about, security (Jarvis 2019). This also allows for linking it with related literatures, such as CTS, and new contexts, such as Twitter.
Studies of the vernacular have largely been limited to focus groups and methodologies with explicitly identified research subjects who are addressed and speak as such Lister 2013, 2016;Vaughan-Williams and Stevens 2015;Jackson and Hall 2016). While these methods offer rich opportunities for sustained exchange, Jarvis (2019) notes that these methodologies risk a degree of artificiality and rely on participants' willingness to contribute. While he does not suggest social media as an alternate avenue for researchers, others have identified online platforms to address these concerns and provide a novel view on the vernacular (Bogain 2020;Da Silva and Crilley 2017). This is not to say that social media data is without its own limitations or that one method is more "truthful" than the other, but that online sources can offer a less contrived view into everyday discourses than traditional research settings. Going beyond the unidirectional media, culture, policy, and "expert" outputs long analysed in CTS literature (inter alia Sorenson 2009;Hülsse and Spencer 2008;Jackson 2005;Stampnitzky 2013), this invites novel approaches and insights.
Such elites and expert outputs are featured prominently in security studies more broadly. Meanwhile, their lay audience is often assumed to be passive (Pears 2016, 80) and, besides vernacular work, constructivist approaches such as CTS have largely failed to account for the role of laypeople beyond being "the audience". The terms of security, thus, are understood to be constructed and contestable, yet constructed and contestable mostly by those who are also assumed to have written them. Homolar and Rodríguez-Merino (2019), for example, note that audience comprehension of a violent event depends on its discursive presentation; but their analysis suggests the locus of such presentation at the level of "political agents". Elites speak security, laypeople listena setup also found in the foundational writings of the Copenhagen School (inter alia Buzan, Waever, and de Wilde 1997;Waever 1995). 2 Excluding the vernacular turn, a "'methodological elitism"' (Stanley and Jackson 2016) has characterised much security and terrorism research. Approaching social media from a vernacular security perspective serves to ameliorate this elitism; there, commenting on terrorism is open to all users. Ordinary posters can become dominant actors (Klinger and Svensson 2015), especially during a crisis (Meraz and Papacharissi 2013).
Of course, social media is not an elite-nor expert-free space. Many political commentators, journalists, and academics are popular on Twitter. In the UK, Twitter users in managerial, administrative, and professional occupations are overrepresented compared to the British population (Sloan 2017). Yet, #BlackTwitter (Graham and Smith 2016) and campaigns such as #BlackLivesMatter (Nummi, Jennings, and Feagin 2019) and #IamaRefugee (Estrada, Anderson, and Brown 2021), among many others, have illustrated how the site also allows for bottom-up political contestation and provides a space for marginalised voices. Like Bogain's (2020) analysis of comments on online news articles, "the findings of this article are not intended to offer an exact reflection of public opinion as a whole, but they do provide an important insight into the way security is conceptualised by ordinary citizens online" (597). "Ordinary", in our reading, is therefore better conceptualised as "lay", referencing those not engaged in the elite production of security, such as researchers, policymakers, journalists, and terrorism experts.

Security and the construction of (British) Muslims 3
The Manchester bombing provoked a range of responses, in media, politics, and online as well as offline (Zhao and Zhan 2019). Given its association with jihadist terrorism, the attack particularly implicated (British) Muslims: in the following month, anti-Muslim hate crimes in Manchester increased by 500% compared to the previous year (Halliday 2017) and anti-immigrant as well as anti-refugee sentiments intensified (Mancosu, Cappiali, and Pereira 2018). Generally, anti-Muslim and anti-Arab hate crimes have been found to increase in the wake of terror attacks (Awan and Zempi 2016;Hanes and Machin 2014).
On social media, the most "liked" messages in response to the Manchester bombing were more ambivalent; they contain emotive and human-interest content both positive and negative in nature (Zhao and Zhan 2019). Generally, information-sharing and positive content is more popular on Twitter in the aftermath of a terror attack than negative posts (Burnap et al. 2014;Fischer-Preßler, Schwemmer, and Fischbach 2019). We should, consequently, expect to find mostly positive discourses in our dataset. Meanwhile, established scholarship on Islamophobia online and securitisation in the context of terrorism more broadly, as summarised below, would suggest much more negative posts. These contradictory findings necessitate further analysis of vernacular post-attack discourses on social media.
There are numerous ways in which the category of "Muslim" has been deployed in the context of security and terrorism discussions. In the UK, the label increasingly took on salience after the Iranian revolution and Satanic verses episode (Modood 2006), when both Islam and Muslims became situated as existential threats to the liberal political and social order of the country (Jackson 2005), something that has accelerated post-9/11 in Europe and beyond (Cesari 2009). Muslims and Islam are othered in counter-terrorism and counterradicalisation discourses (Pantazis and Pemberton 2009;Fekete 2009;Awan 2012), with Islamophobia becoming increasingly institutionalised in response to "Islamist" terror (Gilks 2020). These discourses, in turn, build on larger and historical frames of Orientalism and racism (Amin-Khan 2012). These portrayals strip Muslims of their agency; they are reported on, written about, yet rarely given a voice (S. Ahmed 2009). Employing a vernacular lens and turning to social media can ameliorate some of this exclusion, since, there, marginalised groups can oppose mainstream media portrayals (Sobande, Fearfull, and Brownlie 2019).
Conversely, the internet is also a site of sustained Islamophobia (Ekman 2015;Sian 2018;Awan 2016) and scholars have argued that online platforms can amplify anti-Muslim discourses found in traditional media (Törnberg and Petter 2016). In their analysis of 10,000 tweets mentioning #jihad, Aguilera-Carnerero and Azeez (2016), for example, found that users othered Muslims, considered Islam as oppositional to the West, and that discussion was led by Islamophobic voices rather than Muslims themselves. These discourses are not "merely" virtual, however, with research indicating that internet hate can manifest as offline violence (Williams et al. 2020) and that online and analogue victimisation are linked (Awan and Zempi 2016).
Yet, offline media, especially print, have been the primary avenue through which to understand dominant discursive constructions of Muslims from a security lens (for the UK see e.g. Moore, Mason, and Lewis 2008;Nickels et al. 2012;Saeed 2007). The increasing penetration of social media into everyday use to read about and comment on news, but also as a source of data used by "conventional" news sources themselves, necessitates greater consideration of discourses produced about Muslims online.

Analytical and methodological approaches to understanding the Manchester bombing on Twitter
Analysing social media in relation to security events requires a multi-layered methodological approach. The first challenge is defining and collecting a robust dataset. Once the data is collected, it needs to be mapped into a social network graph to understand connections between users and the influence of posts. Lastly, one can analyse the discourses shared in the most influential tweets.

Twitter data collection
Social media research, like all research, entails unique opportunities as well as limitations. On Twitter, these mainly concern delineation of the sample, data access, and representativeness. However, while these challenges need to be acknowledged, they do not foreclose studies in this burgeoning area of security studies. The data examined in this article is part of a larger dataset focussed on the Manchester bombing. The slice of the dataset used contains a concurrent 27,367 tweets posted between 22:31 on 22 May 2017, when the bombing occurred, and 09:59 on 23 May 2017.
The tweets were obtained through the now decommissioned Texifter service and identified using keywords. Through the service, keywords and timeframe need to be stipulated to determine the tweets one retrieves. Since an exploratory examination of Twitter searching for "Manchester" at the time of the bombing drew in too many irrelevant posts but revealed that a focus of the post-attack discussions were "Muslims", we combined the keywords "Manchester" and "Muslims" to retrieve our dataset. 4 This situates the study within important discussions regarding how Muslims and Islam are enmeshed in security discourses. However, while our choice of keywords builds on the literature's interest in Muslims as focal points of discourses on terrorism and enabled us to get a more focussed dataset, it limits the study to post-attack discussions that mention Muslims and that do so explicitly. With greater financial and human resources, an analysis of the wider post-attack discourse could have been conducted. This is an important avenue for future research.
A likely more significant gap in our data is content removed by Twitter itself, because of its policies of deleting tweets considered hate speech and suspending or deleting offending accounts. The site's guidelines "prohibit targeting individuals with content intended to incite fear or spread fearful stereotypes about a protected category . . . e.g., 'all [religious group] are terrorists.'" (Twitter n. d.). While many tweets in our dataset fit this description and continue to be online, we assume that an unknown proportion of particularly explicit and hateful posts might have been removed before we could retrieve them.
Furthermore, it is important to note that social media users are not a homogenous group nor representative of the general population, particularly those users who participate in political discussions (Barberá and Rivero 2014;Mellon and Prosser 2017). In the USA, for example, Twitter users have also been found to be younger, richer and more left/ liberal politically than the general population, and more likely to express pro-immigration views (Wojcik and Hughes 2019). As noted earlier, anecdotal observations suggest that journalists and political elites might also be overrepresented on the site. Furthermore, our insights are not blindly generalisable to social media per se. We study Twitter as an example of social media and discourse online, not a synonym for it. Further research should explore the (re)conceptualisations of terrorist violence on other sites within a CTS and vernacular security frame, opening the door for cross-site comparisons.
Nonetheless, even with these limitations in mind, the interactive nature of social media and Twitter in particular opens up rich avenues for the analysis of how discursive responses to terrorism are subject to bottom-up contestation and reconstruction. Both in their immediacy, where users can turn to Twitter before the material realities of a terror attack have been officially confirmed, and their multivocality, tweets highlight how terrorism is socially constructed not only by security elites and agents of terror, but by everyday individuals. This renders them a crucial context of investigation.

Social network analysis as a sampling method for large Twitter data sets
Social media presents researchers with a near overwhelming amount of data, even more so if one seeks to do qualitative analysis "by hand". Twitter is no exception. One might manage this wealth of text by randomly sampling tweets or narrowing down the sampling frame, but these approaches ignore a key feature of social media posts: their influence. Therefore, they fail to account for one of the central elements that makes these data both different and analytically rich. Mindful of these shortcomings, we leverage social network analysis (SNA) to generate a sample of the most influential tweets, underline the role of lay actors online, and understand the connections between them.
SNA uses a mathematical calculation to quantify influence with all actors within a network. To do this, the dataset was imported into the network analysis and visualisation software Gephi (Bastian, Heymann, and Jacomy 2009), which was used to generate a social network graph used here that demonstrates: • How individuals share messages.
• How their behaviour forms clusters.
• The relationships between clusters of different messages.
Gephi enables the rendering of large datasets into "sociograms". Within these graphical representations, individuals are represented by points (nodes), and their interactions with each other by lines (edges). Thus, SNA examines structural relationships between socially connected actors (Davies 2009) and offers insights into social media behaviour including establishing patterns of situational awareness, alongside the salience of particular messages (W. Ahmed and Lugovic 2019). With this in mind, it is possible to examine conversations on Twitter not just in terms of their content, but in their patterns of information flow (Himelboim et al. 2017). SNA enables us to map the various constituents of this network space and their relations to each other (Smith et al. 2014).
This relational approach enables researchers to identify influence within a network, including the most frequently interacted with posts. It can do this through a measure called betweenness centrality which determines how connected an individual is to the rest of the network. In broad terms, the more connected an individual is, the more engagement their messages are receiving and thus the more influential they are. Graphically, betweenness centrality is a measure of centrality based on shortest paths. Within outputted sociograms, with every pair of vertices in a connected graph, there will always be at least one shortest path and the betweenness centrality for each vertex is the number of these shortest paths. This means that in terms of network theory, a node with higher betweenness centrality would, in effect, naturally have more control over the network, because more information will always pass through that node. As such, nodes with the highest betweenness centrality demonstrate the capacity to influence through these messages. This is an important and undervalued aspect of SNA that can be applied to studies even when the aim of the study is not to interrogate the networks themselves. We used betweenness centrality to rank the tweets by influence and identify the top 10% most influential tweets, reducing the initial sample of 27,367 posts to 2,736.
In Figure 1, one can see the disparity of influence by tweet and actor; a few actors, identifiable by their different colours, are connected to a much larger number of actors by edges within the graph demonstrating connectedness and engagement in the forms of retweets and replies. In the appendix, the individual four network graphs are shown, extracted in their entirety from Figure 1. Notably, none of the central actors are members of security elites; rather, they are football and pop star fan accounts, as well as "regular", non-expert social media users.
It is important to note that influence and engagement do not equal agreement with a message; actually, they can mean the opposite, as engagements with tweets that make them "influential" through betweenness centrality can be antagonistic. For example, the most influential user within the entire network, a young, white, British woman with fewer than 500 followers, posted an initial message that sought to banalise Muslims, as displayed in Figure 2.
Although this positive message has been shared widely, many connected users respond to it negatively, specifically in relation to Islam and terrorism: This example demonstrates the importance, methodologically and conceptually, of not conflating influence on social media with agreement or endorsement. It underlines how social media is a space where dominant discourses are de-constructed and challenged alongside their reification, and, methodologically, highlights the need to combine SNA with an analysis of posts' content.

Discourse analysis of the sampled tweets
Discourse analysis is an interdisciplinary methodology (Brown and George 1983) seeking to understand how language enacts social and cultural perspectives and identities (Gee 2004). Within this, critical discourse analysis (CDA) seeks to bring power relations and ideology into the analysis to examine how these are established and re-produced through language (Fairclough 2013). Following Amin-Khan's (2012) critique that security studies and research on Muslim securitisation need to take power and socio-political contexts into account, such attention to the social dimension of language renders CDA a fitting methodology for understanding constructions of jihadist terror and its perpetrators online.
Regardless of theoretical underpinnings, discourse analysis requires operationalisation in how it is robustly and systemically applied to any given dataset. This study chooses to operationalise discourse analysis via critical reflexive thematic analysis, using a 6-step coding process described by Braun and Clarke (2006): (1) Familiarisation with data.
The umbrella of discourse analysis is wide, and, in our view, risks being methodologically fuzzy; since reflexive thematic analysis overlaps significantly with constructionist, patternbased discourse analysis (Braun and Clarke 2021), we borrow the analytical strategy of the former to concretise the implementation of the latter.  Once betweenness centrality had been used as a metric of influence to generate the set of tweets to be analysed, unreadable, non-English language tweets, and those which could not be clarified because linked media or tweets were unavailable were removed, leaving 2,677 posts for coding. The data was sorted and coded in Excel, with the unit of analysis being an individual tweet. When relevant, posts were assigned to more than one theme. Retweets are included in the sample as they indicate both message saliency and reception.

The audience speaks back: (De-) constructing "Muslim" terrorism on Twitter
The tweets analysed engage with a variety of discourses; they reflect on the role of religion in general and Islam in particular, they assign blame, they discuss the figures of the terrorist and his victims and consider the appropriate response to such violence. In other words, they provide the insights into "the specific language and text produced by lay members of the public about the subject of 'terrorism' in terms of its specific nature, causes, and solutions" that Jackson and Hall (2016, 3) have called for. Taken together, these categories shed light on how terrorism is constructed in the wake of an attack and emphasise the active role lay individuals play in the construction and contestation of security speak. Indeed, there are no significant security elite voices, such as national politicians or security experts, in the sample.
What is notable are the high proportion of tweets in defence of Muslims and Islam and the range of discourses on interpretations of the incident. While there is no discussion on whether the attack can be understood as terrorism -this descriptor is adopted across the data, users differ in their response to it. They are aware of both the supposedly pro-and anti-Muslim/Islam narratives emerging in the aftermath of jihadist terrorism and actively respond to them. Twitter functions as a site of metadiscourse and, following Eriksson (2016), backchannel discourse on how to respond to and interpret terrorist violence. Like other studies have indicated, positive messages, meaning those defending Muslims, were prominent, but over 10% also shared negative, Islamophobic sentiments, and we argue that even the "Islamophilic" posts reify the link between Muslims and terrorism. It is essential to take such (meta-) discourses into account, since they indicate that users are not only aware of the dominant narratives produced in the wake of an attack, but that they also shape and challenge them.
The data was divided into eight main themes (Table 1), including an "Other" category that is not analysed here, and covers neutral posts, news, and tweets that did not fit within the other themes:

Defence of Muslims/Islam
A majority of the tweets are not concerned with the material realities of the attack per se, but the response to it; they reference perceived or predicted anti-Muslim and anti-Islam narratives emerging in the wake of the incident. Some directly respond to Islamophobic tweets, while others refer to Islamophobia more broadly (see Figure 4). They function as active "backchannels" to discourses that blame Muslims and Islam at large for terrorism (Eriksson 2016). To do this, users banalise Muslims, evoke a good/bad Muslim dichotomy, or construct Islam as a positive, peaceful faith and render terrorists inherently un-Islamic. However, close analysis indicates that even the defence of Muslims and Islam risks reifying their discursive link to terrorism.
In high prevalence across the data, Muslims are banalised (Downing 2019); they become "taxi drivers offering free lifts", "Muslim doctors", fellow Mancunians, or young concert attendees instead of terrorists. The incomprehensibility and violence of the attack is juxtaposed with the commonplaceness of Muslims. Furthermore, their portrayal  highlights not only their ordinariness, but also their helpfulness in the wake of the attack. In some tweets, Muslims do not only belong to the Manchester community, but contribute to it. The salient part of their identification is meant to be their jobs or their Manchester residence rather than their religion (also see Downing 2019). In other tweets, Muslims and Islam are defended from a religious vantage point; their faith is described as a faith of peace, and violence is understood as antithetical to it. Rather than banal, Muslims become innocent. Following this discourse, some users want to "remind" others that it was not a Muslim who did the bombing; that it simply could not have been one, since the brutality of the act renders it inherently non-Islamic. Muslims, therefore, cannot be blamed, since the attack was committed by a "monster" not a Muslim. Instead of the good/bad Muslim dichotomy, bad Muslims are no longer Muslims at all. Rather than a suspect community (Pantazis and Pemberton 2009;Breen-Smyth 2014;Awan 2012), Muslims in their plurality are imagined as an innocent one. Neutrality, however, is foreclosed.
Other users draw on exactly this good/bad Muslim dichotomy (Mamdani 2002;Maira 2009) to reinforce it, rescuing the majority of Muslims from the charge of terrorism by blaming a few "extremists" that take the scripture too far or misinterpret it. Not "all Muslims", this line of argument goes, are terrorists. A wholesale condemnation of the community is dismissed. Nonetheless, the notion of Muslims as potentially suspect and dangerous is implicitly maintained.
Across the variations of the defence discourse, the imperative of "don't blame Muslims", a phrase popular in the data, anticipates an audience to be addressed; at times, these are specific users as indicated by the @ function or retweets, but often the audience is Twitter and society at large. The users exhibit familiarity with the anti-Muslim discourses expressed in discussions about terrorism and the increase of Islamophobia and hate crimes in the wake of terror attacks (Halliday 2017; Awan and Zempi 2017; Hanes and Machin 2014). Manchester becomes embedded in a series of terror-Islamophobia cycles that the users want to break. This is particularly notable since it adds nuance to the picture painted by previous studies that have focused on online Islamophobia (inter alia Awan 2016; Aguilera-Carnerero and Azeez 2016; Horsti 2017).

Terrorism has no religion
The second most common theme within the tweets is the declaration, often verbatim, that "terrorism has no religion" (see Figure 5). This allows us to analyse not only users' understanding of terrorism, but of "religious terrorism" in particular. The discourse can be read from two angles, as outlined below, one that seeks to construct terrorism as a-religious while the other portrays it as pan-religious. Neither approach, however, fully discounts religion as a valid lens through which to interpret terrorism, even if it is as its antithesis.
The "no religion" argument often overlaps with a defence of Muslims or Islam, and it is never brought forward by those blaming Muslims for the attack. In many tweets, it is an extension of the more specific "the terrorists cannot have been Muslim because Muslims would not do something like this" reasoning. The act was not only un-Islamic, but irreligious, users argue. This entails that to maintain the dichotomy between religion and terrorism, terrorists are denied their faith. There are some "moderate" or "good" Muslims who can speak and stand for Islam; but jihadist terrorists with their explicit invocation of the faith cannot. This underlines the different ways "Muslim" can be taken up as a self-descriptor (Adamson 2011) and raises questions about agency and authority regarding who and what gets to be Muslim in the context of (the war on) global jihadist terror that are beyond the scope of this article.
Markedly, "religious terrorism" is not called into question due to the ambiguities that separate it from so-called "secular terrorism" (cf. Gunning and Jackson 2011); the "no religion" discourse is not based on material or empirical concerns about what constitutes terrorism, but interpretations of what constitutes "religious". This defence of faith exceeds a narrow desecuritisation of Islam, it makes a normative claim about religion as a whole, as inherently non-violent and opposed to terror. However, this discourse is also highly contested, with unique tweets as well as retweets and responses emphasising the role of Islam in the Manchester attack and terrorism in general, as outlined below.
Other claims about the irreligiosity of terrorism appear more concerned with its panreligiosity, referencing that both perpetrators and victims of it are not limited to one creed. While terrorism becomes an absolute evil external to faith(s) in the first iteration of the "no religion" discourse, terrorism is a danger possibly inherent to all belief systems in the second one. It is constructed as the problem of "extremists" and "radicals" within a religion. Islam, then, implicitly and at times explicitly appears particularly vulnerable to such extreme readings and the good Muslim/bad Muslim dichotomy is reified. However, the overarching drive of "terrorism has no religion" is to bring together rather than divide.
Notably, whilst users emphasise that terrorism has no religion, the response to it apparently does: the hashtag #prayforManchester is used in 17% of all analysed tweets, often alongside the "terrorism has no religion" claim. This call to prayer appears to be used by Muslims as well as Christians and, presumably, other faiths represented in the data. Echoing similar sentiments expressed on Twitter after other recent terror attacks in Europe (Eriksson Krutrök and Lindgren 2018), prayer functions as a signal to indicate both grief, solace, and solidarity. Its hashtagged nature further highlights the call's function to unite the users and constructs prayer as a shared, universal response to the incident. In paradoxical tandem with the argument that terrorism has no religion, #prayforManchester helps to disassociate religion from the violence of the terror attack and locates it on the side of the victims and the grieving public instead.

Constructing terrorists
11% of the tweets provide insight into who terrorists are constructed to be, with nearly all emphasising their exceptionalism. Echoing Roach, Cartwright, and Pease (2020) findings, people also generally express anger at the perpetrator. Markedly, none of the users mention Salman Abedi, whose identity was leaked shortly after the bombing, by name. Instead, users draw on broader tropes about Muslim terrorists and appear to make a point about terrorism and, some, "Muslim terrorism" more generally instead. This highlights how individual terror events are embedded in larger (in)security discourses, on which users draw, as well as how they contribute to these. Others have noted that such communitarisation of terrorist violence, shifting away from its individual perpetrators, is also more common in the reception of attacks like the Manchester bombing: while white and right-wing perpetrators are "lone wolfs", jihadist terrorists stand in for their faith community (Breen-Smyth 2020).
In many of the tweets and at times overlapping with the "Don't blame Muslims" discourse, someone who commits a terrorist attack is an "extremist", rather than a "real", regular Muslim. While, according to some tweets, such "radicals" also exist in other faiths, this discourse echoes policy and academic concerns about the relationship between radicalisation, terrorism, and Islam (e.g. Ranstorp 2010; critical e.g. Heath-Kelly 2013). Terrorism is constructed as the extreme and the terrorist as far removed from the mainstream, as "sick" or "insane". According to Crenshaw (2014), coverage of domestic terrorism emphasises perpetrators' pathological and personal abnormality, while foreign attackers are portrayed as transnational threats with little personal context. 5 Tweets in the dataset mostly fall into the latter category, but with a third narrative, that of foreignness qua religious radicalism, predominating. While not mentioned in the data, Abedi's own status as a British citizen further complicates the domestic-foreign distinction proposed by Crenshaw (2014). Instead, it points to the subjective and constructivist nature of the discourse. "Muslim terrorists", thus, are othered and rendered extreme through their religion, even if due to their extremism they fall outside the boundaries of "real" Muslim identity. Independent of their citizenship status, their extremism is Islamic and, therefore, implicitly foreign. Personal or secular motivations, meanwhile, are not considered in the tweets (also see Gunning and Jackson 2011). Neither do users draw on the "terrorists as vulnerable or brainwashed victims" narrative featured prominently in Jackson and Hall's (2016) focus groups.
Beyond being portrayed as extreme, the language used to refer to terrorists indicates a degree of depersonalisation. Users write of "monsters", "barbaric animals" and "DEMONS". Many express their disbelief about how someone could commit such a violent act. These terrorist monsters also stand in implicit juxtaposition to Muslims as a banal community, they "help to reinforce boundaries between self and other, civilisation and barbarism, good and evil" (Devetak 2005, 642; also see Rai 2004). Terrorists are the incomprehensible and abject other; to religion and Islam, as in the "Don't blame Muslims" and "terrorism has no religion" discourses, but also vis-à-vis society at large. As indicated elsewhere in CTS and securitisation literature, such othering of terrorists can have severe policy consequences, including justifying extra-legal measures in the name of counterterrorism (Jackson 2005).

Criticising Muslims/Islam
11% of tweets in the sample are critical of Muslims and/or Islam, with many assigning blame for the attack to the Muslim community or Islamic faith as a whole. Drawing on broader Islamophobic and xenophobic narratives, the tweets provide insight into the construction of "Muslim" violence and its supposed perpetrators. Indeed, users present an inverse of the "no religion" argument -they argue that exactly Islam as a religo-cultural Other is to blame for the violence. Albeit making up a comparatively small part of the sample, these posts represent an important discourse that other users are both anticipating and directly responding to, as outlined previously. Furthermore, the relative lack of anti-Muslim sentiments in our dataset might be read more as an illustration of Twitter's liberal and pro-immigration user base (Wojcik and Hughes 2019), as well as evidence of the deletion of xenophobic and Islamophobic posts than sign of the infrequency of postattack Islamophobia in general.
Many users characterise terror as inherent to Islam, and Muslims as suspect outsiders. Some differentiate between the "radicals" described previously, the "religious Muslims with long hair [who] support Jihad" and non-terrorist Muslims but still maintain that a susceptibility to violence is inherent to the faith. Several refer to Islam as a violent or sexist religion, echoing longstanding Orientalist tropes (Said [1978(Said [ ] 1979. While some tweets express their anger at "fucking Muslims" as a whole, others highlight that while not all Muslims are terrorists, (nearly) all terrorists are Muslim. This suggests that non-Muslim violence may not be interpreted through a terrorism framework. Terrorism, understood as something extreme and incomprehensibly violent -extreme and incomprehensible to a white, Christian, Western sensibility, that is -becomes the perpetual problem of the Other.
Additionally, some posts evoke a terrorism-Muslims-migration nexus and securitise migration qua terrorism (Nail 2016), stating, for example, that "[t]his is the result of mass Muslim immigration. THIS MUST STOP!" Tweets like this highlight how post-attack discourses transcend a focus on terrorism and engage with other salient issues and policy demands, in this case, border control and restrictive migration regimes. They indicate a partial convergence of the figure of the '"Muslim immigrant" and that of the terrorist, portraying those who allowed Muslim immigration as complicit in the bombing. This serves to illustrate Mancosu et al.'s (2018) finding that the attack intensified antiimmigrant sentiments. However, amalgamations of terror, Islam, and migration are hardly new nor limited to the Manchester bombing. Bogain (2020), for example, found similar narratives in his analysis of online comments on articles about the French state of emergency.
Other posts link to "Muslim terrorism" more generally and reference non-UK contexts, such as the "travel ban" instated in the US the same year or the conflicts in Syria and Iraq. Notably, more than half of the tweets in this theme were posted by users listing an identifiable non-UK location. The event, therefore, is understood beyond its national context and Islam is constructed not as a local foe, but a global one -echoing a clash of civilisations logic. "Manchester" changes from a specific locality or event into shorthand for "what happens" when Muslims "buy into the ideology they profess to believe [in]". This transnational dimension of the discourse underlines how studying cyberspace might allow researchers to transcend methodological nationalism in security studies and raises new questions for both vernacular security studies and CTS in terms of the construction of an attack in a particular national locale that becomes embedded into a larger international public sphere. However, given that the project focuses on tweets in English and that many users turn off geo-localisation, a more in-depth analysis of the location of individual messages is not possible and beyond the scope of this project, presenting a task for future research.
Across the tweets, Muslims and Islam are essentialised and, if at all, differ mostly in their degree of radicalism (see Semati 2011 on the essentialisation of Muslims). This homogenisation can be read as the counterpart to efforts at individualising Muslims in other themes. While "Muslim defence" tweets banalise Muslims as individuals in the midst of Manchester's community, the "blame Muslims" discourse securitises them as a collective external to (British/Western) society.

Muslim response
10% of the posts are concerned with Muslims' response to the attack (see Figure 6). The vast majority of them are messages by self-identified Muslims themselves that condemn and grieve the violence. However, a smaller proportion of the content, about 2%, are posts sharing reports of Muslims allegedly celebrating the incident. The tweets illustrate the importance and self-aware performance of "correct" responses to terrorist violence that adheres to collective "feeling rules" (Eroukhmanoff 2019a) -failure to comply provokes outrage and proves Muslim complicity.
Both discourses concerned with Muslims' response to the attack assume that they are supposed to respond. A few tweets even call on them to do so, with one asking "[I] [w]onder if Muslims in Manchester & everywhere will trend #MuslimsDenounceJihad." They remain potentially suspect unless they actively declare their opposition to violence. Neutrality, meanwhile, is foreclosed. In its denial qua condemnation as well as when it is "proven" in the Muslim celebration discourse, the association between terrorism and Islam is reified.
The condemnation discourse also places Muslims into a position where they are speaking to jihadist terrorists and non-Muslims simultaneously, having to communicate to both that violence and its perpetrators are un-Islamic. They "condemn this evil & barbaric act" and "extend [their] deepest condolences" to the victims. It is impossible to know if their message is heard, but it is re-tweeted, with a range of Muslim condemnation tweets being shared and commented on by other users. While Abdel-Fattah (2017) argues that these efforts to condemn terrorism can be interpreted as internalised Islamophobia, the "Muslim condemnation" tweets also indicate Muslims' agency to respond to and deconstruct discourses about their religion and community.
Notably, this diverges from Aguilera-Carnerero and Azeez's (2016) finding that most people commenting on jihad online are non-Muslims and calls for more research into mobilisation of social media to renegotiate (in)security by Muslims in the context of jihadist violence. Using identifiers such as "we" and declaring "I'm a Muslim too", ostensibly Muslim users seek to represent their faith towards the outside and position themselves with the grieving public rather than the Muslim attacker. Not aiming to deny internal differences, the shared identifier "Muslim" is leveraged to construct Muslim belonging, indicating how it can be employed strategically for collective claims-making and political articulation (Adamson 2011;Kahani-Hopkins and Hopkins 2002). The hashtag #WeStandTogether, used in 26% of the "Muslim condemnation" tweets, underlines this further.

Using Manchester to critique "politically correctness"
7% of the tweets are critical commentary on the defence of Muslims and Islam described above (see Figure 7). Often overlapping with anti-Muslim narratives, these posts decry that the left and "liberals" are supposedly preoccupied with Islamophobia and "Muslims' feelings" rather than worried about the real issue, "Islamic terrorism". Like the "Muslim response" theme, this discourse highlights the contested nature of postattack reactions. It underlines the significance of studying not only constructions of what terrorism is but how lay responses to it are discursively governed. Notably, in this discourse, users are less concerned with elite responses or policy, their criticism primarily focuses on other Twitter users. Underlying this is a stance critical of both Muslims and a "liberal", politically correct culture that is supposedly more concerned with minority rights than the lives lost in the attack. Echoing state of emergency logic, some users indicate a willingness to sacrifice civil liberties and minority rights in order to appropriately, read forcefully, respond to the attack. What is constructed as empathy and solidarity in the "Muslim defence" discourse is here understood as competitive victimisation and politicisation of the tragedy -never mind the inherently politicised nature of the "anti-liberal response" discourse itself. Indeed, many of the tweets reference the attack's immediate victims as those who people should actually be concerned about. Users emphasise that "children" died and that the "BLOOD ON THE BODIES IN MANCHESTER" had not even dried before people pivoted their attention to Islamophobia. Understanding terrorism, thus, transcends the immediate circumstances of an attack; it includes the myriad of ways such an attack is interpreted and integrated into existing political discourses. Whether that is one that is concerned with minority rights and hate speech, or one that is opposed to supposed political correctness.

Constructing Manchester through victimhood
Although rarely a focal point of CTS, victimhood is not neutral (McGowan 2016). 6% of the tweets give insight into which attributes of the bombing's victims are assigned post-facto meaning and how they contribute to the interpretation of the attack. In the sample, victimhood is gendered and aged, with users highlighting that many of the concert attendees were young women and girls, given the demographics of Ariana Grande's fanbase. They are ideal victims (on this, see Lawther 2014; Christie 1986): "innocent little girls", "literal children." Their suffering is evoked to underline the barbarity of the attack and, often, that of its Muslim perpetrator. The focus on victims' gendered and ageassociated innocence and fragility converges with Jarvis and Lister (2016) and Jackson and Hall's (2016) findings that vernacular understandings of terrorism, similar to those found in elite discourse, generally define it as the killing of "innocent" civilians. However, in line with McGowan (2016), who such a victim is and how far both innocence and victimisation extend is subjective and contested.
Tellingly, there is a second discourse about the victims of terrorism present in the data: one about non-white and Muslim victims. Some users argue that Muslims were also present at the concert and were equally impacted by the attack. A few posters make more general statements criticising perceived worthy/unworthy victim differentiation (Camarillo 2018), with one tweet stating that "Muslims and Muslim countries get bombed, no one ask [s]." Conceptualising Muslims as (fellow) victims can be read as a de-securitising move, however, it maintains an essentialising victim-perpetrator binary. In this discourse, the victimisation of Muslims is what allows them to avoid suspicion; mirroring the "defence of Muslims" and "Muslim condemnation" themes, Muslims cannot be neutral bystanders.

Conclusions on terrorism and vernacular security in the aftermath of the Manchester bombing
The study finds that Twitter provides a platform for a plethora of competing, contradicting, and self-aware vernacular discourses on terrorism in the aftermath of terrorist violence. This is likely to be a generalisable conclusion, with some important caveats. First, social media platforms are ever-evolving; their shifts in terms of popularity and their variation in output formats will dictate how and where constructions about terrorism occur in the future. Twitter, while likely to remain important, is only one in an increasingly diverse social media ecosystem, and future studies should consider how other platforms offer insights into terrorism constructions. Second, this article has demonstrated the unpredictability of the discursive landscape around terrorism. While some of the tweets demonstrate the salience of marginalising discourses, users also defended Muslims and attempted to nuance their position in British society. This paradoxical and dynamic discursive repertoire is likely to be a persistent feature of bottom-up, user-generated content about terrorism. It suggests that lay social media audiences are not only increasingly literate in post-attack discourses on "Muslim terrorism" but are confident enough to deploy their own narratives within the discursive landscape. Further research might seek to compare such discourses found on social media with those uttered "from above" in response to the same event.
Like Jarvis and Lister (2016), we did not encounter disengaged nor silenced security audiences (cf. Roe 2012). Rather than the passive receptors of security speak, awaiting its interpretation by security elites, we found laypeople who speak security themselves and even pre-empt common tropes on terrorism, seeking to engage with, manipulate, and critique them. This affords discursive agency both to the lay audiences and the Muslim subjects of security -agency for which traditional "accounts of security discourse have left depressingly little room" (Pears 2016, 80). Social media, with its decentralised and networked structure of influential speech, therefore, presents a fundamental challenge to theorisations of security and securitisation that presumes the audience as passive, and underlines the importance of the vernacular turn. Especially in the direct aftermath of an attack, when expert analyses are still being conducted, political speeches written, and media reports put together, Twitter offers a view into immediate lay responses and emergent discourses. Indeed, our findings suggest an audience that, on Twitter, is hardly only audience but seeks to speak security itself and challenge securitisation. The "Muslim response" tweets, additionally, indicate that social media may serve as a site for securitised groups to contest their portrayal (also see Bahfen 2018).
Furthermore, our findings challenge the conclusion that social media, such as Twitter, functions as mere amplifier of anti-Muslim narratives spread by traditional media and add nuance to scholarship that primarily focuses on the internet as a place for sharing Islamophobic messages. Many users in our dataset move beyond securitising Muslims or Islam and construct them as banal or even a force for good instead. The religious and political meaning of "Muslims" and "Islam" themselves is subject to ambiguous and competing interpretations. Being attentive to such re-constructions and deconstructions of common discourses about Muslims and terrorism heeds Jarvis's (2019, 120) call for vernacular security studies that are "responsive to precisely these [experienced and lived] understandings and imaginaries rather than having potential threats or fears mapped out in advance". However, continuing links between terrorism, Islam, and Muslims even in the "Muslim defence" discourse also illustrate that even when securitising speech acts are criticised or deconstructed, the underlying logic established by them can endure. These insights render social media an ever more important arena for both the study of vernacular security but also for critical terrorism scholars who seek to understand the variety of ways that contemporary terror events are constructed "from below". Notes 1. Hours after the detonation, American media identified Abedi as the suspected perpetrator and journalists began to construct a narrative about him in the following days (inter alia Evans et al. 2017). Meanwhile, Theresa May gave an official statement at Downing Street over 12 hours after the attack (Walker and Elgot 2017) and the UK's Joint Terrorism Analysis Centre, after expert deliberation, raised the national threat level from severe to critical the evening of the 23 rd (MI5 2017). 2. Insufficient conceptualisation of the "audience" has also drawn criticism from within the securitisation framework (Jarvis and Legrand 2017; Léonard and Kaunert 2010; also see Huysmans 2011, on security beyond the exceptional). 3. Given the location of the attack, we focus on studies conducted in the UK, but similar observations regarding securitisation and Islamophobia have been made in other contexts (inter alia Eroukhmanoff 2019b; Cesari 2009; Cherney and Murphy 2016). 4. While other studies on Twitter delineate their sample with the help of hashtags (e.g. Eriksson 2016;Merrill et al. 2020), we searched beyond hashtags to include a wider range of posts and in recognition that neither #Muslims nor #Manchester are sufficiently established and specific to warrant a more targeted search. 5. Crenshaw, notably, did not control for attackers' faith.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Notes on contributors
Dr Joseph Downing works at the European Institute, LSE. He is a Fellow Nationalism at the LSE and has been a Marie Curie Fellow at the CNRS in Marseille, France and SOAS, University of London. Sarah Gerwens is a PhD candidate at the European Institute, LSE. She draws on critical race theory and critical whiteness studies to research institutional discrimination and discourses on race and "diversity" in German schooling.
Richard Dron works at the Salford Business School, University of Salford. He is a lecturer in digital business and has published extensively on social network analysis, digital marketing and social media. Figure A1. Network graph created by engagement with the most central influential user's content, who criticised those "blaming muslims [sic]" and emphasised Muslims' positive contributions in the wake of the attack. Figure A2. Network graph created by engagement with the second most central user's content; a user whose other posts and username focuses on the pop band One Direction and who asked people not to blame Muslims, declaring that "terrorism has no religion". Like the users at the centre of Figures 1, 3, and 4, the poster themselves does not appear as an influential actor, but their content is highly engaged with, illustrating how an influencer in a terror debate on Twitter may rise from obscurity and return to this post-event. Figure A3. Network graph created by engagement with the third most central influential user's content who highlighted hat "terrorism has no religion" in their initial tweet. Figure A4. Network graph created by engagement with the fourth most central influential user's contenta highly critical "message to those abusing Muslims" posted by a Manchester United fan account.