The assemblages of flagging and de-platforming against marginalised content creators

Misused or malicious flagging as online abuse

Online harms are influenced by the social circumstances they are enacted in and by those shaping technologies and governance (Castells, 2001). Schoenebeck and Blackwell (2021) distinguish between ‘platform-perpetrated harms (i.e. those perpetrated by the design of platforms) and platform-enabled harms (i.e. those facilitated by platforms but perpetrated by users or groups)’ (p. 8).

Behaviours identified with online abuse include: sending harassing or threatening messages; posting derogatory comments about someone online, or physically threatening or intimidating them; spreading rumours about someone, stalking someone (Hinduja and Patchin, 2007); ‘threats of rape and other violence, intentionally offensive messages (racist, misogynistic, etc.), hate speech, and libellous personal insults’ (Golbeck et al., 2017: 1).

Different studies define these behaviours differently, from ‘cyberbullying’ (Hinduja and Patchin, 2007; Navarro et al., 2015) to ‘trolling’ (Keats Citron, 2014) and ‘flaming’, which has often been compared to direct offline harassment (Jane, 2014; Mantilla, 2015). Often used by anonymous accounts against women in the public sphere and marginalised communities, flaming consists in ‘issuing graphic rape and death threats’, through a language made of sexually explicit terms such as ‘homophobic and misogynist epithets’, ‘scathing, appearance-related judgments’ and ‘coerced sex acts as all-purpose correctives’ (Jane, 2014: 558, 560). This speech, which Jane defines as e-bile, combines desire and disgust, in a form of ‘lascivious contempt’ that simultaneously hypersexualises and then derogates targets for their sexuality or appearance (Jane, 2014: 560).

Despite the different but overlapping terminologies, previous work on online abuse presents two aspects which are relevant to this study: firstly, the idea of ‘digital mob’ violence against women and marginalised communities; and secondly, the focus on visible abuse, overlooking the misuse of existing harm reduction platform functionalities amongst the behaviours cited as harmful. Misused or failing harm reduction functionalities have meant that anti-discrimination policies often tend to be exploited by dominant groups (e.g. white men) against protected categories (e.g. Black women), but that the automated tools and the human moderators enforcing them would not safeguard those same protected categories in the same way when they were harassed (Diaz and Hecht-Felella, 2021). Yet, apart from work by Crawford and Gillespie, Fiore-Silfvast and Schoenebeck & Blackwell, there is little literature on the misuse of flagging, which is both a platform-enabled harm (because of the misuse of an affordance) and a platform-perpetrated harm (due to platforms’ mistaken uptake of malicious flags).

At this juncture, it becomes relevant to situate the misuse of flagging within the platform infrastructure that enables it, showcasing how different elements of content moderation can assemble to facilitate censorship and online abuse.

The assemblages of flagging and de-platforming

The same existing offline social inequalities and circumstances which influence online abuse also influence the legal and platform frameworks that govern social networks such as Instagram and TikTok, which crack down harder on nudity, sex and self-expression than on violence, and on marginalised communities instead of their abusers (Are, 2022; Paasonen et al., 2019). While platforms are failing to tackle different types of abuse towards women in the public eye, particularly via direct messages (Centre for Countering Digital Hate, 2022), they have notoriously been over-censoring self-expression, nudity, sexuality and content by those same users after FOSTA/SESTA (Are, 2021; Cotter, 2021; Paasonen et al., 2019 etc.).

An exception to the US Telecommunications Act 1996s Section 230 approved in 2018, the United States’ Allow States and Victims to Fight Online Sex Trafficking Act (FOSTA) and the Stop Enabling Sex Traffickers Act (SESTA) made platforms legally liable for hosting content that facilitates sex trafficking and sex work. Aside from lumping sex trafficking (a crime) in with sex work (a job), and therefore facilitating the de-platforming of swathes of sex workers who lost their livelihoods, networks and opportunities to work in safer, vetted spaces (Blunt and Wolf, 2020; Coombes et al., 2022), FOSTA/SESTA resulted in platforms over-censoring anything remotely sex-related to avoid being accused of enabling trafficking, applying US legislation to global content (Bronstein, 2021; Paasonen et al., 2019 etc). Following the exception, creators at the margins – sex workers, models, performers, sex educators, activists – as well as athletes, lingerie and sexual health brands have had their content censored and their profiles disabled (Are, 2021, 2022; Cotter, 2021).

Since content moderation takes place in a complex ecosystem of policies, algorithmic recommender and demotion systems, automated detection tools and human enforcement (Bishop, 2019; Diaz and Hecht-Felella, 2021; Goanta and Ortolani, 2021), I frame flagging and de-platforming as moderation assemblages, or ‘the logics, processes and outcomes of social media content moderation’ (Gerrard and Thornham, 2020: 1279–1280), particularly when discussing how they come together to perpetrate online abuse and reinforce offline inequalities against marginalised groups. Used by Gerrard and Thornham (2020) to conceptualise how different elements in platform governance come together to perpetrate sexism, assemblages are the human and mechanical elements joining forces to generate specific results. As such, interrogating how different elements of content moderation – and in this case flagging and de-platforming – assemble means asking questions about content, interfaces, platform policies and machine-learning to understand what the joining of different elements of platform governance does to the social world (ibid).

In the case of misused or malicious flagging, the content is nude, sexual and queer posts disproportionately targeted by platforms post-FOSTA/SESTA (Are and Paasonen, 2021). The platform policies are rules, which are set subjectively, informed by the rule-setters’ biases, or, as Roberts (2019) argues, made ‘in the specific and rarefied sociocultural context of educated, economically elite, politically libertarian, and racially monochromatic Silicon Valley, USA’ or in socially conservative China (pp. 93–94). As a result, platforms’ machine-learning processes are informed by and targeted towards enacting this worldview, disproportionately regulating content by marginalised communities while falling short of protecting them and letting harms by dominant groups go unpunished (Diaz and Hecht-Felella, 2021).

Using assemblages theory to understand the misuse of flagging is relevant because ‘assemblage theory helps to direct us to silences’ (Gerrard and Thornham, 2020: 1280), and flagging is characterised by silence: platforms share very little information about its use or uptake (Crawford and Gillespie, 2016), and users only see the action’s result, such as de-platformed content or profiles, often after different forms of online abuse or of virality (Are, 2022, 2023; Perrett, 2021; Stokel-Walker, 2022).

Given the pressures social media platforms find themselves under when governing content (Are, 2021) and Instagram and TikTok’s known lack of transparency in communicating details about their content moderation decisions or the take-up of user reports (Are, 2022; Crawford and Gillespie, 2016), the misuse of flagging on Instagram and TikTok and its relationship with de-platforming demands more scrutiny to interrogate the effectiveness of platform affordances, highlight user experience and educate social media account owners about a still under-researched form of online abuse. Towards these aims, this paper answers the following questions:

RQ1

How can misused or malicious flagging towards users’ content and/or accounts be identified?

RQ2

How does de-platforming after suspected misused or malicious flagging affect targeted users?

Methods

For this study, I carried out 12 semi-structured, virtual, ethnographic interviews via the videoconferencing software Zoom. In qualitative research interviews, the researcher asks open-ended questions to elicit rich responses from participants (Roulston, 2022). Ethnographic interviews rely on participants’ description of spaces, actions or events and on researchers’ ongoing analysis of data through field notes, observation and participation in research settings (ibid). The researcher begins by asking participants open-ended questions to make sense of their space, time, events, people and activities, to then qualitatively analyse the data (ibid). Researchers carrying out ethnographic interviews often participate in the settings they study (ibid), which I did by observing and noting down my own ongoing experiences of censorship following perceived flagging on Instagram and TikTok – experiences often similar to participants’ and which informed this research’s design, its inclusion criteria and interview questions (Are, 2022).

The study’s aim, sample specificity, theoretical background and data determined the sample, since particularly when the study’s aim is narrow and the combination of participants is specific to its objectives, Malterud et al. (2016) argue it does not require large participant numbers, as the data produced will be relevant and rich.

This study’s aim and sample specificity were indeed narrow, prioritising similar experiences, that is, content or account take-downs after perceived malicious flagging. I therefore capped participants at 12 to diversify the represented experiences to include both users posting content that has been notoriously over-censored (e.g. nudity, sex work, LGBTQIA+ expression, as shown by Are, 2021 and Haimson, 2021) and journalists and activists (Stokel-Walker, 2022).

To take part, participants had to have experienced content or account deletions on Instagram and TikTok and have received negative comments on their posts, mirroring my experiences of account deletion after receiving swathes of negative comments upon going viral (Are, 2022) and other instances where flags triggered de-platforming (Clark-Flory, 2019; Perrett, 2021; Silverman and Fortis, 2023). Recruitment took place via a blend of targeted outreach to users in my network who shared similar experiences, and of onboarding through a cross-platform call for participants shared across my Instagram, Twitter and TikTok profiles with my combined following of over 400,000. Before confirming their participation, participants were vetted for their experiences and were therefore chosen only if they met both inclusion criteria. Following the interviews, they were paid £50 for their time and expertise.

The data collected was transcribed and subsequently analysed through thematic analysis (TA) in the form of answer excerpts. TA is a qualitative method which allows researchers to identify, analyse and report themes or patterns within data (Braun and Clarke, 2006). Themes are ‘creative and interpretive stories about the data, produced at the intersection of the researcher’s theoretical assumptions, their analytic resources and skill, and the data themselves’ (Braun and Clarke, 2019: 594). The researcher develops them through coding, often reflecting data collection questions (ibid). Characterised by minimal data organisation, TA is a realist method describing data in rich detail and providing insights into relevant lived experiences, particularly when conducting research on stigmatised communities (Braun and Clarke, 2006).

Consistently with my previous experiences of carrying out studies amongst censored, marginalised users (Are, 2023; Are and Briggs, 2023), participants yearned for someone to listen to them, so much that some refused payment for participation because they were ‘happy to help’, while others told me: ‘Thank you for giving me the space to rant’. The data collected was therefore rich, lengthy and nuanced, leading me to conduct a preliminary analysis, to then analyse the responses collected through two rounds of data analysis.

This study presents a set of limitations. Critics of interviews argue that they leave too much room for researcher bias, both in asking questions and follow-up questions and in the framing of answers (Roulston, 2022). And indeed the choice of interviewees, both amongst those to whom I directly outreached to and those who responded to my call for participants, can be both a consequence of legislation that has brought platforms to over-moderate anything mildly sex-related (Paasonen et al., 2019), generating similar experiences of de-platforming, but also a reflection of the user demographics in my network of sex-positive, queer followers, with whom I engage from the positionality of a white, bisexual, cisgender woman with experiences of digital censorship.

Critics of my approach may argue that trying to provide evidence of malicious or misused flagging leading to de-platforming through the experiences of de-platformed users may not be effective, as it may be relying on conjectures or misunderstandings of content moderation processes. Some may wish I would have spoken to those who engage in flagging instead of to those affected by it, or argue that those reporting participants may not be doing so maliciously, and may just be expressing their disagreement with content. However, as the data will show, participants’ content often complies with public-facing community guidelines, showing discrepancies in platforms’ enforcement. Further, in the face of platforms’ known lack of transparency (Are, 2021), flagging either manifests as a lack of change on participants’ profiles or on their de-platforming: it is therefore an action difficult to detect from outside platforms, a silence (Gerrard and Thornham, 2020), making its manifestation from a user’s perspective relevant to consider. Lastly, and most importantly, following a series of investigations (Clark-Flory, 2019; Cox, 2021; Silverman and Fortis, 2023) and Meta’s own Oversight Board’s (2023) findings that repeated reports resulted in the deletion of non-violating content, and since investigating similar conjectures has shed light on equally mysterious moderation practices such as shadowbanning (Leybold and Nadegger, 2023), listening to participants’ experiences, conjectures and suspicions is crucial in platform governance. Indeed, consistently with previous research on creator communities and de-platforming (Are, 2022; Cotter, 2021; Garcia-Rapp, 2017), the specific experiences of creators are often the only type of information we have about platform processes, making users experts in their field. Since Big Tech tends to discredit and patronise participants, I wish to instead believe them, and consider them as collaborators in holding platforms to account instead of entertaining platforms’ black box gaslighting about their governance (Cotter, 2021).

Analysis

Participants were all over 18 years old and based in the countries where most of my social media following lies: the United Kingdom (6), Italy (2), the United States (2), Ireland (1) and Australia (1). Interviewees were given the choice to be kept anonymous (and therefore protected from further harms and referred to with a pseudonym and an asterisk) or to be named and credited for their experiences and expertise.

Each participant provided a preferred description of their accounts’ aim and niche, which this paper has attempted to follow. They were: two transgender men working as writers posting about their transition journey as well as education and activism - one, Rob*, based in the UK and one, Elia, based in Italy; Sylvia*, a UK-based cisgender woman journalist and content creator posting about sex education; Nicole*, a US-based cis woman meme creator; Roxie*, an Australian creator and artist; Malli, a UK-based cis man creating content about mental health; Valery*, a UK-based non-binary model and performer; Marta, a UK-based Ukrainian cis woman creating content about the war in Ukraine; Bel, an Irish transfeminine ‘wannabe’ creator building an audience through pole dance and expressing their identity; Gin, an Italian queer, cis woman posting about emotional and sexual health; Alex, an Italian, UK-resident non-binary activist using Instagram on a personal and activist basis and Reed, a sex, nudity and sex work advocate, sex educator and podcast host. Most participants had been de-platformed at least once, with some claiming to have had their accounts disabled up to 10 times, sometimes in one year alone. Some deletions were never overturned, while others lasted from days to months on end. Only one user’s account had never been de-platformed, but had different types of content removed.

The first round of TA consisted in placing the data in broader, preliminary themes surrounding misused flagging, de-platforming and their overall effects on users’ lives and livelihoods. The second round narrowed themes down to three:

1. Malicious or misused flagging in action

The following sections present the themes through participant quotes and the researcher’s sense-making, with the aim to show how de-platforming after malicious flagging manifests.

Malicious or misused flagging in action

All participants in this study believed they had been targeted with malicious flagging, and provided significant examples of the practice in action. However, their relationship with this functionality was conflicted: while they described flagging as an action platforms allowed users to perform, participants also viewed it as something done to their account rather than something they themselves could do, because often when they did it, platforms would not listen.

Participants identified a blend of negative comments after their posts had gone viral, the disappearance of older posts and the accumulation of community guidelines violations for disparate reasons unrelated to their behaviour as examples that flagging was being weaponised against their account. For instance, transgender activist and educator Elia became suspicious he was being reported by other accounts when ‘a lot of posts and stories that were getting deleted didn’t have anything explicitly against community guidelines’ (i.e. they were not nude or explicit), resulting in a barrage of deletions of even older posts, when his profile was personal and not professional. Users connected misused or malicious flagging initially to a series of ‘punishments’ (i.e. removal of the ability to go live, to comment on or save posts, followed by account deletion warnings) and then to full de-platforming.

The experiences shared by participants highlight major flaws within platforms’ flagging affordance, ranging from platforms triggering de-platforming after flags by users reporting others as a mode of feed curation, to remove content they did not wish to see, all the way to maliciously reporting others according to completely unrelated community guidelines violations to silence creators. For example, see the experiences of Elia, of Ukrainian activist Marta, of transfeminine ‘wannabe’ creator Bel, and of mental health advocate Malli:

• ‘I got the notification from Instagram that my post was against community guidelines and that it had been removed, sometimes for things that really astounded me, such as sex acts or violence against minors’.

As shown by the above quotes, participants use algorithmic gossip (Bishop, 2019) or folk theories (Eslami et al., 2016) to make sense of their experiences. This attempt at sense-making happened throughout data collection, and can be interpreted as an example of the confusing nature of opaque content moderation rather than as an attempt by participants to justify their actions. And indeed, this discrepancy between the content users posted and the community guidelines according to which it or their profile was removed seems to lend legitimacy to their own folk theories (ibid) or sense-making: it seems odd that mistaken detection could be the only reason behind this censorship, particularly in instances when they were not posting any new content. Instead, it seems that platforms’ flag function may be assembling with their decision to de-platform content and profiles they already deem controversial based on the assemblage of existing prejudice against content, platform policies, machine-learning and enforcement (Gerrard and Thornham, 2020), in a similar vein to findings by Silverman and Fortis (2023).

At present, the number of flags required for content or account removal is not publicised by platforms. Users receive confirmation that their account is being reported only when platform workers or users tell them about it. After her personal account was de-platformed, sex worker and educator Reed received a message from a Facebook worker on her podcast’s official account:

‘This girl, when she looked into it she said it looked like there were a couple of accounts targeting me and constantly reporting posts on my account, up to a point where it basically filled up the quota that Instagram needed for it to be automatically taken down’.

Flagging creates power imbalances between followers and users, who repeatedly found that content requiring the application of nuance and context to understand the meaning of posts often became a way in for malicious flaggers. For example, participants shared that some of the content they believed triggered flagging and de-platforming included: the words ‘bitch’ or ‘fag’ used by way of reclamation by women and queer participants removed as hate speech; humour by white people towards white people removed as hate speech or bullying; war reporting removed as violence and graphic content; sex education mistakenly removed as nudity and sexual activity. Users have therefore had to negotiate what they could and could not post, or even which type of content they chose to try to recover: no swearing, no close-ups of anything beyond their face, not even Renaissance paintings showing nudity.

A blend of algorithmically, automatically censored content and posts flagged by others, taking advantage of algorithmic and moderation loopholes within Instagram and TikTok, have therefore resulted in post or profile take-downs for participants. Users can effectively game the system, taking advantage of the assembling of unclear, puritanical and broad community guidelines, automated moderation and underfunded, under-paid human moderators to attempt to remove content and profiles they disagree with. This is particularly evident with nuanced but potentially controversial content, such as sex education, LGBTQIA + content, nudity and war reporting, in an algorithmic rendition of Jane’s ‘lascivious contempt’, simultaneously over-focusing on and frowning upon women and marginalised communities (Jane, 2014: 560). Yet, despite known controversies about this content (e.g. see Perrett, 2021; Stokel-Walker, 2022), platforms do not seem to have added any layer of protection to detect nuance in flags or understand their context. Indeed, it appears that a report by an anonymous account set up for harassment or lurking has the same relevance as any legitimate user looking to report concerning content.

Participants also shared that platforms lack specific functionalities to signal that malicious flagging may be taking place. At present, there is no ‘misuse of flagging’ specific ‘report abuse’ category beyond ‘bullying and harassment’ to highlight that rogue activity is taking place, and in a scenario where hate speech and harassing comments are not demoted or removed by platforms, users find it even more challenging to report accounts for misuse of flagging as bullying. Malli highlighted that, due to platforms’ lack of vigilance, flags could come from the same person creating multiple anonymous accounts: ‘People are afforded a certain amount of anonymity on it, because there's so many accounts that don't have anything, no picture, no information, and each one of those is potentially a troll’. Additionally, trans activist Rob* shared that the only way he could secure a take-down for harassment and coordinated flagging by a known anti-trans activist was by reporting the pictures she lifted and reposted from his account as a copyright violation.

In this sense, users feel targeted from multiple fronts: from users who mistakenly report content they disagree with; by harassers who engage in visible abusive behaviours through comments and, supposedly, coordinated flagging efforts; and by platforms, who consider these efforts as valid without considering the context, and without providing users with an explanation or a form of complaint for what is going on, showcasing a combination of platform-enabled and platform-perpetrated harms against the most marginalised users (Schoenebeck and Blackwell, 2021), and the assemblage of different layers of platform governance (Gerrard and Thornham, 2020), perpetrating censorship and online abuse.

Cross-platform abuse and lack of nuance in moderation

Online abuse via misused flagging escalated when it became a coordinated cross-platform action. Indeed, although this study focuses on Instagram and TikTok, a variety of participants experienced cross-platform abuse: they were targeted by users on one platform and then the abuse spread to other platforms and offline. This included harassment and accusations - e.g. transgender users being accused of paedophilia - and flagging, leading to content and account take-downs on their main content creation platforms (Instagram and TikTok) following specific interactions with other users or after viral posts.

Outside of Instagram and TikTok, users found that malicious actors mainly used Twitter and Telegram to coordinate cross-platform abuse to harass and de-platform them from their main platforms. Although based in different countries, this study’s transgender participants, for instance, shared similar experiences: Rob* found that anti-trans Twitter users from the UK would periodically go through ‘all prominent trans people's accounts [on social media] and find things to report’, while Elia, based in Italy, said:

‘Through a friend who sent me a screenshot I noticed the existence of a Telegram group – which was triggering in itself because it was called something like, “Let’s off them all.” […] The screenshots I was sent were referring to my posts, so people were sharing screenshots of my posts in the Telegram group saying, “Go flag this person,” as well as sharing a series of slurs. All the screenshots were of my face, or just of me, so they weren’t criticising my work, or the fact I did sex education, or that I posted sex toys – it was a personal attack’.

The identities of flaggers mentioned by participants seem to indeed vary by country. Marta, a high-profile Ukrainian creator, found Russian bots or accounts were the main players who tried to de-platform her, judging by the negative comments she received. UK-based transgender participants largely attributed mass reporting and de-platforming to Twitter mobs of so-called gender critical accounts, or those excluding transgender women from cisgender spaces and debates (Zanghellini, 2020). Italian users Alex, Elia and Gin referred to members of their hometown communities or to Telegram groups of middle-aged fascist sympathisers instead. Participants became aware of abusers’ identities after they entered the groups, which regularly shared the Italian ‘quando c’era lui’, or ‘when he’ – ‘he’ being Mussolini – ‘was around,’ implying they longed for times in which LGBTQIA+ people were not allowed to exist as themselves. In other cases, users who had rejected advances from their followers believed their deletions were a direct result of this rejection, a personal campaign against them.

Geopolitics, social and cultural inequalities and nuances therefore showcase the global in- equalities of platform governance, assembling with platform infrastructure against users’ identities as much as against their content. This has meant that platforms have de-platformed sexuality or identity-focused content and posts by marginalised groups, but not identical content posted by celebrities or hate against the same marginalised groups. Non-binary activist Alex, for example, described their experiences of being de-platformed after reclaiming homophobic slurs, showcasing discrepancies between content moderation actions against marginalised communities and against dominant groups (Diaz and Hecht-Felella, 2021). Users, therefore, seem to think social media companies are selective about whose freedom of speech they prioritise: transphobic slurs, accusations of paedophilia or racism against people of colour was not de-platformed in participants’ experiences, while expressions of their gender identities, jokes about white people posted by white creators or sexual expression tend to be targeted by platforms’ moderation.

The emotional impact of malicious flagging and de-platforming on users and their circles

Consistently with previous studies on de-platforming and flagging (Are and Briggs, 2023; Myers West, 2018), users cited severe adverse emotions resulting from the assemblage of misused flagging and de-platforming, resulting in loss of work and inspiration to create content, and in serious stress on the back of what they understood as bullying from both platforms and users.

The work, community building and creativity that go into building oneself a platform have left users extremely distressed upon de-platforming. Consistently with Myers West (2018), being targeted by both platforms and their own followers or viewers left participants sharing feelings of low mood, stress, anxiety, and fear. The uncertainty on the back of malicious flagging left users feeling insecure, fearful and distrusting of their own communities.

Users also found de-platforming and harassment to be an extension of the hate and exclusion they faced offline, in a space they once viewed as safe, creative and generative. This was particularly true for LGBTQIA+ and transgender users such as sex educator Gin and non-binary transfeminine creator Bel:

• ‘I created my profile to say, “Enough with my past,” with my family who insisted I had to be female on their terms, find a husband, become a housewife. It was a rebellion, saying, “This is me,” but once the profile was removed, I felt that I couldn’t be who I was anymore’.

Even more concerningly, the harassment users faced resulted in increasing, never-ending spirals of abuse and de-platforming that continued throughout users’ online and offline lives particularly against transgender participants, who were periodically targeted and even stalked and harassed to invalidate their earnings, their academic work, their book publishers and their earning prospects.

Additionally, participants shared that being the target of this form of platform-enabled and platform-perpetrated abuse had an impact not just on them directly, but on their partners, friends, loved ones and communities too – a previously over-looked aspect in platform governance and de-platforming research, as showcased in these quotes by Gin and Elia:

• ‘The deletion of [partner’s profile on the back of flagging] has been the worst experience of my life so far. We already had a series of ongoing collaborations where we were waiting for payment, we’d just moved in together not even three months prior, we didn’t even know what living alone meant and had no chance to ease ourselves in without this bomb dropping on us. I don’t have a good relationship with my parents, so I didn’t want to ask them for money, but the collaborations we had in place either didn’t happen or didn’t pay us because [partner’s] profile was gone’.

In this sense, participants’ experiences seem to show that platform governance does not only affect them personally: it makes them unable to perform and fulfil the duty of care that their following, their content has bestowed on them, showing the impact of de-platforming not just on individual users, but on wider communities.

Lastly, consistently with Are (2023), participants perceived receiving community guidelines violations, particularly for nudity, sexual activity and even soliciting, as a form of consent violation, as an interpretation of their content, activities and persona that they did not agree with and that they found upsetting. This consent violation assumed even sexist and transphobic dimensions: non-binary participants found their posts would stay up when they behaved and dressed in ways socialised as male, and would be removed when they were posing or dressing in ways socialised as female. For queer and trans people, algorithmic labelling was a denial of self-determination: ‘I don’t think, “Today I am female, today I am male” when I post pictures of myself, but the fact that the algorithm moderates me differently depending on which gender it thinks I am is worrying’, said Alex, arguing that it is ‘transphobic that your choice of clothing, or the way that you’re posing, determines how your algorithm understands you and defines you’. The algorithm was engaging in an action - labelling their gender – that they had rejected in their lives and gender expression.

Conclusion

Flagging is characterised by silence (Gerrard and Thornham, 2020), making its assemblage with de-platforming difficult to investigate without speaking to users: platforms share very little information about its uptake (Crawford and Gillespie, 2016), and users only see its result – that is, de-platformed content or profiles, often following online abuse or of virality (Are, 2022, 2023; Perrett, 2021; Stokel-Walker, 2022). This study therefore defines misused or malicious flagging on Instagram and TikTok as the misuse of platforms’ flagging functionality to report content not because it violates social media community guidelines, but due to personal disagreement and/or attempts to harass targeted users to trigger their de-platforming. De-platformed users’ experiences show a user can ascertain that malicious or misused flagging is taking place because either those carrying out the reporting tell them they have flagged their content, or because contacts in and outside platforms alert them to coordinated, cross-platform abuse. However, often, users have no way to confirm they are being targeted with malicious flagging. Therefore, following this study’s findings, it is reasonable to believe that misused or malicious flagging is happening when a user’s account is targeted with one or more of the following: receiving swathes of negative comments on viral content and/or on old posts, experiencing repeated content take-downs of new and old posts, removal of posting privileges or of the opportunity to go live, to like, comment or follow.

This study has found significant loopholes within Instagram and TikTok’s flagging affordance, meaning that it can assemble with existing prejudice against specific content and profiles rooted in community guidelines and with users’ agendas putting specific accounts and content particularly at risk of de-platforming via reporting. In particular, nuanced content such as nude or sex-related imagery and videos, LGBTQIA+ expression, journalism and activism seem to be prime misused or malicious flagging targets.

Given that the users interviewed have shared deep, troubling emotional and financial impacts the combination of abuse and de-platforming has on them, stunting their abilities to express and be their true selves, to reach their communities and networks and to be able to work, platforms’ over-reliance on automation to highlight dangerous content through flagging is facilitating, rather than preventing, online abuse. Yet, despite the crippling effects of this protection affordance turned abuse tool, users have found that nothing amongst platforms’ reporting tools can help them notify Instagram and TikTok that malicious reporting is taking place. It is therefore recommended that platforms create ways for users to notify that malicious flagging is happening, and that they re-target the functionality to ask for more specific reasons behind reports to better identify malicious or misused activity, while also investing in contextual, human moderation to prevent the misuse of this tool.