‘24/7 horny&training’: porn bots, authenticity, and social automation on Instagram

ABSTRACT

This article presents a conceptual and methodological account of a small porn bot network, focusing on its embeddedness within Instagram use. The analysis explores the gendered design of bots as platform-native personas, particularly their capacity to perform within the confines of Instagram’s increasingly strict sexual content controls. We address three performative trajectories in the bot-exploited ‘Instagrammatics’ of identity play, social influence, and attention capture. We argue that a bot programmed to operate with sexual content to generate attention relies on the paradoxical blend of pornographic ‘imagination’ and social media ‘authenticity’. For our analysis, we manually identified 30 porn bot accounts spamming in the comment sections of highly visible Instagram posts (those published by @justinbieber). We then collected associated metadata – bot profile names and images, comments and comment likers, followers and followings, bot content, and links in the bot profile bios. By variously situating and combining these data, we discuss how networked automation taps into the sexualized social scripts imitated by ‘artificial’ and ‘authentic’ users alike. Our findings point to how porn bots re-enact gender as a programmed set of instructions, adapting to Instagram’s vision of acceptable sexuality and revealing its normative order.

KEYWORDS:

• Instagram
• porn bots
• spam
• authenticity
• platform governance
• situated methods

Introduction: why porn bots Instagram comments

Whenever a celebrity posts a photograph on Instagram, it is almost inevitable that the first comment on it will be made by a porn bot (Alvarez 2020). Observations of this kind – some focusing on the distorted perception of ‘sexiness’ and social media authenticity (Thing 2020; Salim 2023), others contesting Instagram’s community guidelines prohibiting both nudity and automation (Santos 2023) – have been ubiquitous in online tech reports for years. In academic discussions, scholars usually define porn bots as automated or semi-automated software agents that operate via sexual solicitation to capture attention (Narang 2019). Twitter porn bots using Not Safe for Work (NSFW) hashtags to advertise commercial links have been addressed in their generic patterns (Paasonen, Jarrett, and Light 2019; Maddocks 2020). Tumblr porn bots have generated significant controversy due to their uninterrupted operation during the platform’s so-called ‘porn ban’ (Pilipets and Paasonen 2020; Tiidenberg 2021). Within Instagram, as we will show, porn bots continuously intervene in a network of relationships that users forge with the platform, avoiding detection through a range of circumvention tactics. In all these instances, bots operate as medium-specific, heterosexually scripted ‘personas’ (Bucher 2014), demonstrating remarkable adaptability to the platforms’ dominant traits.

In this article, we further develop ‘Instagrammatics’ (Highfield and Leaver 2016) as a method to study bot performance across the trajectories of identity play, social influence, and attention capture. From identifying hybrid situations in which bots become actors, we expand our focus to the platform environments and artefacts that make such situations possible (Dieter et al. 2019; Marres 2020; Rettberg 2020). Much like the unsolicited digital phenomenon of spam embedded in the internet’s architecture (Parikka and Sampson 2009; Steyerl 2011), porn bots operate within social networks by adapting themselves to platform-specific communicative and technical standards. Rather than trying to determine the likelihood of an account being automated (Martini et al. 2021), we discuss how such standards, formalized into different action possibilities or ‘grammars’ (Agre 1994), may allow researchers to understand bot interventions from a relational angle. This involves addressing how Instagram’s ‘mechanisms of computerised tracking’ (Agre 1994) not only enable networked encounters but also capture these encounters in distinct data formats and metrics (Gerlitz et al. 2019), setting the stage for social automation.

In the highly competitive realm of online engagement, the metrification of social media, alongside concerns about data capture and amplification, has produced the need to move beyond a priori dichotomies such as ‘real’ and ‘fake’ (Burton and Chun 2023; Lindquist and Weltevrede 2024). On Instagram, comments on posts, likes in stories, tags in reels, direct messages, follower-following relations, and links in profile bios, among other pre-structured possibilities to interact, provide paths for software-supported activities programmed to mimic ‘authentic’ engagement (Boshmaf et al. 2011; Guilbeault 2016). As we demonstrate through associated metadata, the application of ‘Instagrammatics’ (Highfield and Leaver 2016; Rogers 2021) as a method to study platform cultures of use highlights bot operators’ techniques of working with and around platform affordances and constraints (Bucher and Helmond 2018). Conversely, it draws attention to the infrastructural hierarchies and norm-producing power (Tiidenberg 2021; Diepeveen 2022) that porn bots simultaneously perpetuate and disrupt.

In the sections that follow, we first situate the conditions under which porn bots operate on Instagram in the broader discussions of sex and social media, problematizing mainstream platforms’ ongoing tendency towards de-platformization of sex (Molldrem 2018; Tiidenberg and van der Nagel 2020). We then present an ensemble of situated methods, beginning with porn bot interventions in the comment sections of five Instagram posts published by @justinbieber in June 2022. Here, we emphasize the role of automated ‘likers’ that push porn bots’ comments to the top, reflecting on their capacity to achieve an ‘authentic’ balance of (in)visibility. We further discuss how bots promoting adult-themed services through sexy pictures and self-descriptions such as ‘24/7 horny&training’ not only temporarily avoid Instagram’s detection mechanisms but also intervene in the relations of following and being followed. We conclude with the heterosexually programmed characteristics of the ‘attention spam’ (Lee et al. 2012) based on the click trajectories of external sites initiated by the disguised links in bot profile bios. This opens porn bot performance to a plural analysis, where each perspective reveals new connections, whether they are automated, genuinely authentic, explicitly sexual, or, as we suggest, always already social.

Porn bots, authenticity, and social automation

From a perspective that approaches the programmed arrangements of social media as dynamic and performative (Bucher 2018), the notion of social automation comes with two implications. First, it sidesteps binary conceptions of human authenticity and non-human artificiality (Guilbeault 2016). Second, it requires critical considerations of authenticity and visibility as values under the market order of platforms (Burton and Chun 2023). In an environment where the notion of ‘appropriate’ sexual content is tied to these values (Bernstein 2010; Blunt et al. 2020), porn bots not only hijack common attention patterns – for example, by spamming in celebrities’ comment sections – but also point to the normative visions (of gender, sexuality, sociality) embedded in Instagram policies and dominant cultures of use.

Contemporary critique addresses the regulatory authority of platforms, challenging corporate definitions of ‘authentic’ and ‘acceptable’ engagement as driven by capitalist and heteronormative logics (Gerrard and Thornham 2020; Tiidenberg 2021). Especially in the realm of Instagram’s influencer industry, where various forms of content amplification are in frequent use, the boundaries for which accounts may remain active are constantly policed and negotiated, setting the limits for the constitution of the so-called ‘good enough publics’ (Lindquist and Weltevrede 2021). Analyzing bot adaptations of the ‘porn chic aesthetic’ popular among female influencers (Drenten, Gurrieri, and Tyler 2019), we explore how these publics help calibrate a porn bot ‘persona’ – ‘social’ enough to come across as authentic but not ‘too visible’, and thus able to escape moderation. We show that while bot interventions are flexibly attuned to what Instagram’s parent company, Meta, defines as spam (Meta 2023a) and sexual solicitation (Meta 2023b), the communicative conventions they sustain reflect the normative logics of attractiveness and desirability prevalent on social media.

Akin to the scripted dynamics known from commercial culture and reality television, authentic engagement on Instagram often involves sexualized identity play in tandem with a self-promotional attitude attached to the cultivation of ‘Instafame’ (Marwick 2015). Like their Twitter counterparts that formerly operated via searchable NSFW hashtags (Paasonen, Jarrett, and Light 2019, 29–30), Instagram porn bots display authenticity by adapting to the rigid standards of attractiveness and femininity. In terms of self-presentation, they reflect the ‘ready-made conventions of character, setting, and action’, which Susan Sontag (1967, 51) described in her essay on ‘The Pornographic Imagination’. In operational terms, however, maintaining a ‘good enough’ appearance for Instagram accounts distributing sexual content has become more challenging. With the platform’s rigorous hashtag moderation (Gerrard 2018), the playground for porn bot interventions has evolved, demanding an advanced level of social and technical performance.

A porn bot persona, correspondingly, is both socially bound – in that it reproduces common rules for crafting presumably attention-grabbing sexual content – and infrastructurally embedded – in that it relies on the manipulation of metrics. As we argue in the following, this relational understanding elevates striking contradictions inherent in mainstream platforms' strategic frameworks. The first strategy of reducing the ambiguity of interactions (and data) coincides with the interpretative flexibility of following and liking that bots actively exploit (Gerlitz, Herma, and Kyrimi 2015). The second strategy refers to platforms’ ongoing efforts towards ‘deplatforming of sex’ (Molldrem 2018). Within a strictly regulated environment, in which platform policies stigmatize and valorize engagement, particularly around sexually suggestive content (Tiidenberg and van der Nagel 2020), the performance of authenticity presumes a complex interplay of factors: achieving the right balance of visibility and invisibility goes hand in hand with conforming to a normative set of social scripts. The very same scripts, in turn, shape the concept of influence on Instagram and are subject to ongoing contestation within user communities (often resulting in their expulsion) (Are and Briggs 2023).

These double standards have been criticized for their negative impact on human content creators rather than porn bots, resulting in the censorship of bodily displays shared by women, athletes, educators, artists, and queer people (Blunt et al. 2020; Morresh 2023). While porn bot content increasingly emulates sexy celebrity posts that continue to thrive on Instagram, communities whose representations do not fit the same narrow standards have often been disproportionately affected by account deletion and shadow banning (Are and Paasonen 2021). Our argument posits that just as these contradictions manifest within the metadata associated with our collection of porn bots, they also reflect a broader framework of online sexual culture and its conservative corporate definitions (McDowell and Tiidenberg 2023). On the one hand, Instagram’s terms of service prescribed by Meta (2023b) manifest increasingly sex-averse community guidelines in the name of safety (Tiidenberg and van der Nagel 2020). On the other hand, the unequal governance of user cultures paradoxically allows porn bots to reproduce normative visions of sexuality that continue to thrive on the platform, often at the expense of marginalized communities.

Bot methodologies: from situations to relations

Our explorations in the following refer to how and when porn bot interventions in user activities take place on Instagram. The situations in which porn bot performance is rendered visible are likely to have stable, ritualized features, but they also come into being as moments of disruption: primarily, bots are designed for ‘selective amplification’ (Agre 2000) of community-building routines through commenting, liking, and following. At the same time, the built-in capacity to embed themselves within different situations of Instagram use renders bot actions somewhat off-sync with the contextual dynamics these situations afford. Ongoing enhancements in Instagram’s (2023) community guidelines continuously transform the operational landscape for both social automation and sexual solicitation. In turn, bot operators face the challenge of adjusting the accounts’ functions, altering the ways in which porn bots operate in different contexts.

For our analysis, we identified 30 bot accounts starting with a typical situation of porn bots in action – spamming in celebrities' comments – through manual exploration of Justin Bieber’s five Instagram posts subsequently published in June 2022 (one of these posts from 3 June 2022 screenshotted and annotated by authors is shown in Figure 1). We selected these accounts based on the shared patterns of comment amplification described below and then collected associated metadata – bot profile names and pictures (n = 30), bot comments (n = 30) and comment likers (n = 1297), bot followers (n = 15,729) and followings (n = 4550), bot profile posts (n = 150), and links in profile bios (n = 28). The research affordances (Weltevrede 2016) of these metadata scraped with Phantombuster’s Instagram modules derive from the situations that porn bots are designed to be part of and that come with certain application programming interface-based access restrictions. Another limitation concerns the fleeting nature of interactions studied: bot accounts associated with other bot accounts appear as quickly as they disappear, resulting in the contrast between the initially visible metrics and the actual number of porn bot followers and comment likers available for scraping.

Figure 1. Typical situation of Instagram porn bots and their ‘likers’ in action (annotated screenshot of @justinbieber’s post comments from 3 June 2022). Note the prevalence of blank accounts with repetitive account names deployed for comment amplification.

As a methodological entry point, likes on comments open up questions about how the sociality of porn bots is produced as visible ‘in situ’ (Marres 2020) but not visible enough to be immediately detected and algorithmically removed at scale. Unsolicited comments, a trademark of porn bots, resemble spam in that they interrupt the ‘organic' flow of interactions, redirecting users’ attention through random forms of address (Paasonen 2006). A comment like ‘Goal me ’, not specifically aimed at anyone but liked by rairebefitsehardbloodliecrit, faecarrrorosibibsouffhenrand, reochilosoftsicbackcirctedse, and hundreds of others (Figure 1), suggests an instant but, from a platform moderation standpoint, also an all too ephemeral rise in automated engagements. As was evident in the comments that soon had lost their visibility along with the likes, examples ranged from positive appraisals such as ‘mega class’ or ‘top grade’, to sexual innuendos such as ‘3 × 23 = ’ or ‘want boyfriend !!’, to phatic emotional expressions – ‘A little bit sad today ’, ‘Today is my birthday ’, or ‘Does anyone care that it's my birthday?’ (Figure 2).

Figure 2. Overview of 30 comments amplified by automated likers constituting a sample of porn bot accounts selected for further exploration.

Importantly, this spammy effect is no longer confined to text-based bulk activities. Rather than simply blocking out other interactions with repeated messages (Steyerl 2011), automated users capture attention through carefully orchestrated ‘relational' (Baym 2015) labor. Building connections and seeking attention are two sides of the same coin present in various aspects of Instagram performance that bots adapt to while aiming to entice potentially curious clickers. By making the most liked and replied to comments appear on the top, the non-chronological order of Instagram's comment sections facilitates such interventions. With their bulk-liked contributions and sexy profile pictures that signal availability, porn bots ‘borrow' random celebrities’ audiences in tandem with the aesthetic of Instagram models (Caldeira, De Ridder, and Van Bauwel 2020) who actively pursue visibility through similar exchanges. Departing from thirty amplified comments with a similar pattern, our situated approach prompted us to further interrogate the relations between the public accounts these engagements originated from.

Using ‘Instagrammatics’ (Highfield and Leaver 2016; Rogers 2021) as a method to study associated metadata comes with some ethical concerns. The latter might encompass scenarios such as non-bot likers that challenge the task of bot detection based on patterns, compromised accounts unknowingly boosting someone else’s visibility through following, or a more carefully crafted formation of ‘instant and real’ (Lindquist and Weltevrede 2021) users programmed to stand in as a marker of authenticity. Approaching the accounts’ aesthetic properties, we work with different visual techniques of deidentification as bot avatars and posts usually include images taken from other, often sensitive, contexts. Account names and textual content, however, deliberately remain explicit as their shared recurrent features allow us to study bot engagement from a relational angle. Understanding these relations against the backdrop of Instagram’s ‘click farms’ and ‘follower factories’ (Confessore et al. 2018) invites considerations of social performances assembled by porn bots as heterogeneous and scalable. Accordingly, there are several action paths that allow a porn bot persona to unfold. The next sections pursue these paths, methodologically repurposing image-sharing, following, and linking to study ‘authenticity’ as an ensemble of ‘programmed visions’ (Chun 2011) – human and automated.

Programmed identity: hetero-sexiness and a porn bot persona

Like all social bots, porn bots take advantage of software for coordinating their actions and require a face (Boshmaf et al. 2011) – a persona designed to appeal to a collective formation of norms, desires, and imaginaries. Persona, in terms of its capacity to fabricate a role, implies performance and masquerade but also rule-bound repetition (Marshall, Moore, and Barbour 2019). It indicates ‘the projection of a character, or role, by an actor, through a mask out into an audience’ (2019, 25). When it comes to gender impersonation in particular, feminist scholars consider the act of wearing a mask as serving a dual purpose: it both reflects a normative precondition of (hetero)sexual role models and bears subversive potential, allowing for a temporary escape. Queer theorist Judith Butler (1990) has influentially analyzed these dynamics in their description of gender as performative and social. Like masks worn during drag performances, ‘natural’ or ‘authentic’ identities also are ‘manufactured through a sustained set of acts’ and ‘positioned through the gendered stylization of the body’ (1990, xv) – two intertwined dimensions of heterosexual social masquerade that Instagram porn bots rely on and which they expose as taken-for-granted through the very act of their under-moderated existence.

Programmed identities in this context indicate the work of porn bots as social personas that, by stubbornly upholding binary figurations of gender, expose the artificiality of platforms as the ‘shapers of sex’ (Tiidenberg and van der Nagel 2020). Examining which scripts are deployed in the specification of such personas through a composite visual analysis of bot content (Figure 3) may thus capture Instagram-native vernacular performances of embodied self (Caliandro and Graham 2020) that porn bots reproduce. As indicated by the pattern shown in Figure 3, the bodies featured in our compilation of bot images echo a highly standardized concept of attractiveness, aligning with the idea of ‘hetero-sexiness’ observed in mainstream social media (Dobson 2011).

Figure 3. Deidentified image stacks compiled from the 10 most liked porn bot images, based on patterns found in 150 posts: ‘butts’, ‘faces’, ‘lingerie’, ‘breasts’, ‘spread legs’, and ‘leaning’. The patterns were specified using qualitative coding and visualized with ImageJ.

Reinforcing broader social norms present in pornography (Sontag 1967), advertising (Goffman 1976), and other popular genres, this pattern highlights attributes such as slimness, youthfulness, whiteness, and physical fitness. It embraces the voluptuous forms and provocative attire typical of heterosexual adult content while valuing ‘traditional’ feminine qualities like cuteness and playfulness: depictions of female models with parted lips and poses that accentuate breasts or butts, therefore, present an ambiguous target for moderation algorithms programmed to separate porn from commercial entertainment – an endeavour that is difficult to accomplish given the mainstreaming of pornography and its constantly changing cultural role (Attwood 2002; Paasonen, Nikunen, and Saarenmaa 2007; Smith 2010).

By close-looking at the visual analytical artefacts assembled through qualitatively coded layers of porn bot images, scholars can consider various aspects: recurrent tropes may refer to how multiple accounts adapt the same gesture or pose, how specific body parts are elevated or cropped, how established conventions of sex appeal are replicated, or how changing Instagram nudity standards are implemented to escape moderation. According to German artist and media theorist Hito Steyerl (2014), such choreographies of imitation under the algorithmic order of platforms are neither neutral nor accidental. The capacity of porn filter algorithms to draw a line ‘between face and butt’ derives from an ‘order of images, a grammar of images, an algorithmic system of sexuality, surveillance, productivity, reputation, and computation’ (Steyerl 2014) that platforms such as Instagram facilitate and that bots reveal through their automated interventions.

Contemporary porn detection systems learn to recognize sexual content by combining human moderation with supervised computer vision algorithms trained on large collections of nude images (Gillespie 2018; Crawford 2021). Meta’s training system uses artificial intelligence (AI) to determine whether a Facebook or Instagram post contains nudity or sexual activity and decides whether to remove content based on its degree of confidence. To make automated decisions, machine learning models must be able to detect relations between categories such as gender and body parts (e.g. ‘uncovered female nipples’) as well as specific actions and objects (e.g. ‘sex toys placed upon or inserted into mouth’) (Meta 2023c). If the confidence is high, the content will be automatically removed; if it is low, the system will call for manual review (Meta 2023d). Criteria for removal not only involve factors such as ‘severity’ or ‘likelihood for violating’ that porn bot content partially fulfils but also ‘virality’, the latter translating into ‘how quickly’ the content generates attention (Meta 2023d).

These criteria notably do not apply within the domain of celebrity culture, where the trend of ‘porn chic aesthetic’ (Drenten, Gurrieri, and Tyler 2019), often amplified by ‘fake’ followers (Buckler 2022), continues to flourish, blatantly disregarding Instagram’s regulations concerning sex and automation. Embedded in the same environment, porn bots not only reproduce the gendered aesthetic of mainstream models but also reveal the platform’s arbitrary social standards. By reusing sexy images circulating on and beyond the platform, porn bot operators create accounts that combine ‘low-confidence’ sexual imagery with moderate metrics ratios to evade detection. Of the 150 analyzed porn bot posts, only two accounts, contributing a total of 15 images, gathered 100–1842 likes per post. In contrast, 87 out of 150 images barely received any interactions. Such findings indicate that porn bot content – despite its easily manipulated vanity metrics and ‘thirst trap’ aesthetic (Leaver, Highfield, and Abidin 2020, 68–69) – is unlikely to be designed for high visibility. Instead, the images operate as ‘poor images’ (Steyerl 2009) or ephemeral copies, fostering the exploitation of digital sexual labour and its widespread aspirational arrangements.

The scale of these arrangements examined using automated methods for web detection of identical images (Google Cloud 2021) unsurprisingly goes beyond the boundaries of singular platforms. Copies of images from our dataset were found across different layers of the web, including Facebook with 47 matches, Twitter (17 matches), reddit (13 matches), and tumblr (7 matches), as well as various porn aggregators such as leaxx.com (10 matches), nsfw.xxx (10 matches), and erofound.com (22 matches). Porn bot ‘selfies’ also occasionally matched pictures of models on analisa.io (18 matches) and pixvox.com (16 matches) – two social media analytic websites that allow for downloading Instagram profile information. Challenging the notion of authenticity through borrowed identities, the image content itself – ‘compressed, reproduced, ripped, remixed, as well as copied and pasted into other channels of distribution’ (Steyerl 2009) – becomes interchangeable and thus loses relevance. What matters is the momentary capacity of Instagram accounts to ‘put on a mask’ or appear authentic just long enough to capture the audience’s attention.

Programmed sociality: following and being followed

The specificity of how porn bots appear to users while remaining invisible (at least temporarily) to Instagram’s moderation devices is strongly intertwined with the workings of bot follower ecologies, which involve different qualities of automation. The interplay of these qualities in the social construction of a porn bot ‘persona’ (Bucher 2014) is fuelled by the simultaneous de-stabilization and reproduction of Instagram-specific habitats, such as displaying popularity scores (Guilbeault 2016). As with other mechanisms of relational design, bot profiles can be understood not only in terms of their immediate appearance but as contingent upon the practices of influencer communities populating the platform.

Central for Instagram’s influencer industry is the idea that, like popularity, authenticity is maintained through ongoing management of social relations, using ‘strategic intimacy’ (Marwick 2016, 333) to appeal to the targeted audience. Departing from the premise that ‘self-presentation is carefully constructed to be consumed by others’ (Marwick and boyd 2011, 140), affiliative techniques of following and being followed (ideally) aim to achieve a carefully curated (and yet presumably authentic) social image. Follower metrics not only facilitate the ‘procedural display of authenticity’ (Burton and Chun 2023, 12) but also create symbolic power, making an online persona appear relevant enough to be followed. The actions of porn bots can be seen as an intervention into these influential tropes, entwining programmed visions of sexuality with the social aspirations of Instagram content creators.

The resulting entanglements of bot accounts with broader user formations can be explored through a network analysis of patterns across shared followers and followings (or followed accounts). Drawing on Marshall, Moore, and Barbour (2019, 97), we argue that these patterns constitute porn bots as online personas with ‘amplified’ components. The process of amplification unfolds in a series of relational actions, revealing bot performance as a human–non-human ensemble of attention-seeking practices. The acts of following – by porn bots themselves and their followers – can thus be studied as indicators of ‘programmed sociality’ (Bucher 2018) and explored by interpreting the distribution of network clusters.

Starting from a list of associated account names (represented as nodes), we built a network where a link is established each time an account follows a bot (and vice versa) (see Figure 4). The resulting relations between bot audiencing practices (represented as edges) were then further explored in view of their unique and imitative qualities. Clusters of followers (white nodes) bridging two or more porn bots help identify which bots appear to attract the same audiences as well as whether these audiences themselves have shared characteristics (such as account names or profile images). Clusters of followees or followed accounts (grey nodes) situate porn bots as social personas performing ‘relational labor’ (Baym 2015), the main motivation behind which is to pass as ‘authentic’ according to Instagram-native popularity criteria. By association with aspiring influencers (Duffy 2016), bots pull into focus a forward-looking form of attention capture, simulating signals of authentic connectivity. The ways through which bots become agents accordingly require a networked effort rather than an individual capacity to attract ‘eyeballs’ constrained within a rigid structure of profile templates.

Figure 4. Gephi network analysis of porn bot followers (white nodes) and followings (grey nodes) showing individual and shared affiliations. Porn bot accounts are shaded black.

To remain undetected, porn bots embed themselves in existing attention ecologies, using strategies of gendered representation and leveraging Instagram's popularity measures and connective features. As Figure 4 shows, among the accounts followed by some of our porn bots, celebrities (including Snoop Dog, Beyoncé, Paris Hilton), politicians (including the account of Vice President of the United States Kamala Harris), entertainment outlets (such as Netflix or Disney), as well as high-end brands (such as Prada or Lamborghini) constitute the main pattern. Like Justin Bieber’s comment sections, the pre-existing visibility these accounts have to offer provides a fruitful basis for porn bots’ affiliative efforts and strategies of extracting ‘value lying dormant within audiences’ (Steyerl 2011, 74). Some of the accounts share aspired affiliations (see clusters of followings connecting storm.huffman, dorothygikas, and lucy_wifkinsonql418), while others operate ‘in the shadows of celebrity’ (Are and Paasonen 2021) in a more targeted way (see the individual followings of _lauren.dickson, maryeedwardss, and others).

Porn bot followers, in turn, bring us back to a relational configuration of social automation that further questions the binary concept of authenticity and standardization. Most porn bots are programmed to ‘always already’ have an audience, imitating influencers’ self-branding and social aspirations (Senft 2008). By identifying coordinated follower ecologies based on the clusters of shared followers and account name patterns (highlighted in Figure 4), we can reflect on how bots simulate authenticity in terms of adjusting the followers-to-following ratio (Cotter 2019). The extent to which these connections are gendered, however, breaks the calculation of the heterosexual male audience a porn bot would be expected to ‘attract’ by design. This is evidenced by the prevalence of female account names (note, in particular, Slavic name patterns) in the clusters of shared followers, the appearance of which is captured in Figure 5.

Figure 5. Fixity and flexibility in the account name patterns and profile images of shared porn bot followers identified through network analysis.

As this visual analysis shows, followers’ engagement in the construction of a porn bot persona relies on a mechanism of random social amplification. The serial naming patterns that are not always ‘properly’ attributed to the profile images involve both fixity and flexibility, creating a glitch in the normative gender matrix. With the standardized female account names haphazardly matched against avatars showing groups of people, men, cars, nature, and AI-generated art, dedicated porn bot followers invoke the notion of a ‘script caught up in the dress code double bind’ (Steyerl 2014). Equating authenticity with metrics, porn bot followers simultaneously drive the strategic possibilities of social heterosexual performance ad absurdum and reveal the normative prescriptions it entails.

Authenticity here plays out as a relational value according to which the question of what an account represents becomes whether (or with whom) it strives to belong (Burton and Chun 2023, 10). Such calibrations of persona – shared by automated and human actors performing digital labour – are not permanent but adjustable to the platforms’ legal and economic pressures (Tiidenberg and van der Nagel 2020; Jarrett 2022). Reinforcing Instagram’s reductionist conceptions of ‘appropriate’ sexual content, porn bots perfectly master these calibrations in that they reflect broader platform visions of social influence. Some of these visions are dictated by the default metrics, moulding online interactions into a programme that bots can reproduce (Guilbeault 2016, 5003–5004). Others have to do with how the image of the authentic ‘self’ is being recognized on the basis of platform activities designed to nudge users towards self-promotion. Especially when it comes to moderation of sexual content, social status often equals inviolability, making evident the opacity of social media community standards, as well as the unequal ways in which they impact different users (Are and Paasonen 2021). As porn bots adopt these values, inherent to online commercial culture, they expose the artificial construct of Instafame itself, reflecting a notion of sociality that is far from being reciprocal. Similar to celebrity accounts they follow, none of the 30 examined porn bots follow back their followers.

Programmed attention: click trajectories and links in profile bios

While the sociality of our porn bots highlights Instagram’s role as a medium of entanglement but also indifference (Wilkie, Mike, and Plummer-Fernandez 2015), their profile bios frame sexual identity performance in terms of its capacity to capture and redirect user attention. At the time of data collection in 2022, this included providing concise and frequently sexually suggestive ‘personal’ information in tandem with disguised short links connecting to dubious online dating sites. Like theatre actors wearing masks to signify character types, maryeedwardss, florencejenkins, _lauren.dickson, eliie_dyson, and other sampled accounts imitate occupations that are common amongst popular ‘sexy’ Instagram influencers and celebrities (Drenten, Gurrieri, and Tyler 2019). Professions such as ‘fashion model’, ‘content creator’, or ‘fitness coach’ demarcate social desirability shaped by gendered vernacular conventions (Figure 6). Additional identity markers such as ‘bad girl’, ‘naughty girl’, ‘sweet booty’, and ‘your adorable muse’ activate pornographic imagination that registers in the heterosexually coded ‘theater of types’ (Sontag 1967, 53). Incorporating playful emojis, invitations to click on the link feed into the attention economy of sexualized digital labour (Hernández 2019; Jarrett 2022), which can be monetized.

Figure 6. Age, profession, and additional ‘personal’ information in bot profile bios presented as invitations to click on the link.

In online environments, in which the promise of sexual content holds engaging potential (Tiidenberg and van der Nagel 2020), female attractiveness is programmed to conform to the normative prescriptions of social roles and values. Porn bots extend the norms reproduced in conventional heterosexual pornography and popular culture (Dobson 2011) to exaggerated forms of address (Paasonen 2006). Under the rule of Instagram’s sexual content controls, bios of female bot models containing suggestive profile pictures in tandem with links and self-descriptions such as ‘24/7 horny&training↶Adult content I make’ stand out as adverts promoting external sites that an implied male user is expected to visit. Attention here is crucial – it informs the very capacity to engage while simultaneously attaching the promise of engagement to well-recognizable heterosexual tropes (see, for example, Paasonen 2006, 408; Paasonen, Jarrett, and Light 2019, 79–82). According to Meta’s (2023b) current policies, sexual solicitation is defined by a ‘suggestive’ pose combined with a request for interaction, which includes links redirecting to external sites. It is a difficult playground for both human users and porn bots to navigate, for engagement in the form of clicks implies both the promise of attention and the risk of being de-platformed. In June 2022, links in porn bot bios still served as a means of circumventing adult content moderation – an important detail that has changed after Instagram introduced new regulations. Showcasing certain ideas about envisioned audiences, the click trajectories of these links reveal a range of mediators employed to monetize flows of activity beyond the platform’s boundaries (Levchenko et al. 2011).

As a method, navigation through the click trajectories can be used to explore the infrastructurally confined scripting of the so-called ‘attention spam’ (Lee et al. 2012). At its best, spamming of this kind targets online encounters where user attention coalesces and then temporarily focuses on a phenomenon – celebrity posts and popular meme accounts being common examples. Since opportunities to share clickable links on Instagram have always been scarce, porn bot linking – as captured by our analysis (Figure 7) – involves a long trajectory. Starting from implicit sexual innuendos embedded in platform interactions with porn bot profiles, curious users can be redirected to explicitly pornographic intermediary sites, also known as ‘pre-lander’ pages, which then link to affiliate dating websites that pay bot operators for generating traffic (Narang 2019). With two different virtual private networks (VPNs) employed to switch between geolocations, the redirection process has led us to the landing pages of seemingly different dating sites with the same ownership statement. In all instances, the publicly available site information and company names aligned with entries found on online dating scam watchlists. Depending on the chosen geolocation, clicking through the links would yield different page results; the rotating trajectories shown in Figure 7 derive from the German context.

Figure 7. Linking logic of Instagram porn bots through click trajectories directing to external pages. The page owners’ names were anonymized due to their similarity to unrelated online businesses.

Short links were present in 28 out of 30 bot profiles, employing services like bit.ly, cutt.ly, and linkr.bio, among others. URL shorteners grant porn spammers a dual advantage – they obscure the actual web address of the shared content and allow for the tracking of click counts (Gupta, Aggarwal, and Kumaraguru 2014). Porn bot interventions are therefore not limited to a singular platform. Like any software agent, a porn bot persona is designed with specific ‘character development’ (Dieter et al. 2019; Burton and Chun 2023) in mind: ‘Just as Instagram influencers can earn cash every time someone clicks their affiliate link to their cosmetics product, porn accounts can make money from getting you onto porn sites – whether or not you spend any money’ (Thing 2020). In the realm of platforms, porn bot performance translates into sequences of disguised actions associated with sexually suggestive but not explicit content. Outside this realm, most click trajectories guided us through overtly pornographic pre-landers aimed at directing traffic to deceptive dating websites via ‘interactive’ surveys about users’ sexual preferences (Figure 8).

Figure 8. Screenshots capturing pre-landing pages (pages that appear before a landing page promoting a service) with interactive ‘sex surveys’ scripted to entice straight male users.

Here, the heteronormative excess of ‘pornoscripts’ (van Doorn 2010) designed to capture and redirect user attention played into the realm of ‘pornographic peekaboo’ (Paasonen, Jarrett, and Light 2019, 52) in which sexual displays are perpetually ‘one click away’, simultaneously visible and hidden. Promoting the possibility of ‘finding sex with local girls’, the backgrounds of pre-landing pages presented a mix of popular porn genres, from oral sex and ‘money shots’ to hentai and furry. The featured ‘sex surveys’ mainly targeted straight male audiences or presented limited binary choices, ultimately directing users to a seemingly ordinary dating site. This exemplifies how attention is orchestrated through the allure of realism and closeness commonly found in online porn spam, which generally invites viewers to follow the action from the male protagonist’s perspective (Paasonen 2006). Of 28 available links, just three redirected us to web destinations that were unrelated to pornography. However, even in this deviation, the click trajectories still adhered to gendered social expectations, leading to the landing page of a major North American sports league.

Conclusion: ‘they always delete my acc ’

Our analysis shows that porn bots operate as heterosexually scripted and infrastructurally embedded social agents. As techniques for attracting audiences (Hernández 2019), porn bot interventions largely depend on the pursuit of attention through a certain extent of sexual solicitation via an approachable public identity. The entanglements of identity performance, social influence, and attention capture constitute a ‘persona’ (Bucher 2014) with a ‘good enough’ appearance and engaging qualities formalized into platform ‘action grammars’ (Gerlitz et al. 2019). The latter can be repurposed as situated dimensions of the method and analyzed as metadata attached to bot comments, posts, follower–following relations, and links. Each dimension reveals a different aspect of social automation, opening up a space for reflection on the hybrid environments that propel bots into action.

In her critique of automation as the general quality of software and computing, Wendy Hui Kyong Chun (2011) points to how digital environments conflate the recurrent and the ephemeral. A ‘24/7 horny&training’ bot persona, flexibly attuned to the ongoing updates in Instagram’s content policies and infrastructure, similarly incorporates both traits. As Instagram increasingly cracks down on both sexual content and automated engagement (Instagram 2023), it perpetuates a cat-and-mouse game in which the platform constantly changes its rules, and bot operators respond by adapting new circumvention tactics. One important aspect of this game is ironically evident in the bot bios stating ‘they always delete my acc ’. Ephemeral by design, porn bots remain prime targets for Instagram moderation policies – liable to be banned either for automating interactions or for appropriating the same kind of sexy displays that Instagram models get flagged for (Schrager 2016). In the three months following data collection, 22 of the 30 accounts studied in this article were removed and the links in the bios of the eight remaining profiles were deactivated. Yet, unlike human labour, social automation can be put on repeat with only little effort. If accounts constantly disappear, they may also reappear with a new variation in the name pattern, or with new functions.

A porn bot persona, therefore, involves examining how mainstream digital culture, porn, and social automation intertwine, despite apparent contradictions. The act of flooding comment sections of influential accounts with random emojis and text – our first methodological entry point – presents a repeatable situation in which interactional scripts break down and simultaneously reinforce bot visibility. Automated likers amplify porn bots’ comments, situating them on the top of the thread. Shared followers and followings lend bots credibility despite strict platform content regulations prohibiting both sex and automation. In the Instagram-mediated situation of spamming Justin Bieber’s comments, porn bot messages, sprinkled with emojis and amplified through automated likes, aimed to make users curious enough to look at the profiles and click on the links. Outside this realm, the attention economy of ‘pornographic peekaboo’ presents an intervention in the control networks of Instagram, relocating the hypothetical audience from the centre of a space that typically avoids adult content to its periphery (Paasonen, Jarrett, and Light 2019).

The overall design, in other words, allows for mutual coordination of Instagram-mediated and web-distributed actions. In an environment where social connections are driven by the intertwined pursuit of intimacy and public attention, porn bots shift the meaning of sexual self-presentation from an aesthetic template to a strategic act of imitation. The excessive repetition, characteristic of spam and pornography, defines bots’ repertoires both on-platform and off-platform, aligning with standardized attentional patterns and commercial motivations. Together, these elements highlight how gender identity shared by porn bots is programmed to appear authentic, as authenticity is always first ‘subject to initial constraints and action paths’ (Burton and Chun 2023, 78) before it results in a particular mode of self-expression. The intertwining of gender stereotypes and imagined audiences feeds into the reproduction of sexualized social scripts imitated by automated and human users alike. Co-constituted by the corporate logic of platforms, porn bot performance re-enacts gender as a programmed set of instructions, adapting to mainstream visions of acceptable sexuality and revealing its normative order.

Acknowledgements

With special thanks to the Center of Advanced Internet Studies (CAIS) in Bochum for hosting our research group. We would also like to thank Marcus Burkhardt, Carolin Gerlitz, and the anonymous reviewers for their encouraging feedback.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by Deutsche Forschungsgemeinschaft [Grant Number Project-ID 262513311 SFB 1187]; Fundação para a Ciência e a Tecnologia [Grant Number PTDC/COM-CSS/5947/2020]; HORIZON EUROPE Marie Sklodowska-Curie Actions [Grant Number 101059460]; Center for Advanced Internet Studies [CAIS Working Group Grant].