Automating social media content moderation: implications for governance and labour discretion

Authors:
Sana Ahmad WZB Berlin Social Science Center and Free University Berlin, Germany

Search for other papers by Sana Ahmad in
Current site
Google Scholar
Close
and
Maximilian Greb WZB Berlin Social Science Center, Germany

Search for other papers by Maximilian Greb in
Current site
Google Scholar
Close
Full Access
Get eTOC alerts
Rights and permissions Cite this article

Content moderation is key to platform operations. Given the largely outsourced character of content moderation work and the dynamic character of social media platforms, technology firms have to address the accompanying high degrees of uncertainty and labour indeterminacy. Central to their managerial strategies is the use of automated technology that allows them to organise work by incorporating the social media user activities within the production processes, and control workers for ensuring the accuracy of content moderation decisions. The labour process analysis is informed by two workshops with ten participants at a Berlin-based IT-services firm providing content moderation services to a lead firm based in the USA. The research design combines together the design thinking method and the focus group interview method to examine the worker–machine interaction. The research findings indicate that technical control results in continuous standardising of content moderation work through routinisation of tasks and codification of time. Its combination with bureaucratic control through the supply-side managerial functions aims to ensure the quality service delivery and points to the continued significance of human supervision. Correspondingly, there are two main contributions of our study: first, regarding the governance in content moderation value chains and second, regarding the worker experiences of technical-driven control. On account of the limited resistance observed in the labour process, we conclude that instead of seeing it as the totalisation of technical control, our findings point towards the structural conditions in Germany that restrict migrant workers’ agency.

Abstract

Content moderation is key to platform operations. Given the largely outsourced character of content moderation work and the dynamic character of social media platforms, technology firms have to address the accompanying high degrees of uncertainty and labour indeterminacy. Central to their managerial strategies is the use of automated technology that allows them to organise work by incorporating the social media user activities within the production processes, and control workers for ensuring the accuracy of content moderation decisions. The labour process analysis is informed by two workshops with ten participants at a Berlin-based IT-services firm providing content moderation services to a lead firm based in the USA. The research design combines together the design thinking method and the focus group interview method to examine the worker–machine interaction. The research findings indicate that technical control results in continuous standardising of content moderation work through routinisation of tasks and codification of time. Its combination with bureaucratic control through the supply-side managerial functions aims to ensure the quality service delivery and points to the continued significance of human supervision. Correspondingly, there are two main contributions of our study: first, regarding the governance in content moderation value chains and second, regarding the worker experiences of technical-driven control. On account of the limited resistance observed in the labour process, we conclude that instead of seeing it as the totalisation of technical control, our findings point towards the structural conditions in Germany that restrict migrant workers’ agency.

Introduction

Technology firms have time and again referred to scalable technologies as the solution to the enormous scale of user content generated on their social media platforms and the limited time window permitted by regulatory measures to moderate content (Gillespie, 2020). Against their promises to users, lawmakers and investors of artificial intelligence (AI) as the future of content moderation, it has become increasingly clear that the bulk of this work continues to be outsourced to back office sections of IT service companies in the Philippines and India as well as to IT-service firms in the USA, Germany and other places in Europe (Gillespie, 2018; Ahmad, 2019; Roberts, 2019; Reuter et al, 2019; Ahmad and Krzywdzinski, 2022). Facebook claims to have 15,000 workers moderating content on its social media; at Google, 10,000 workers have been moderating YouTube and other Google products; and Twitter has declared having 1,500 content moderators (in Silver, 2018; Barrett, 2020).

Content moderation can be understood as those ‘governance mechanisms that structure participation in a community to facilitate cooperation and prevent abuse’ (Grimmelmann, 2015: 47). But it is also a multivalent concept that serves different functions for lead firms operating social media platforms. On one hand, it allows for the maintenance of public discourse and has been emphasised by many including Jürgen Habermas in his new book (2022) underscoring the active character of digital users.1 Critics have however argued that the consequences of these governance choices are far-reaching and have proven to be disadvantageous for marginalised communities and people from the Global South (Ebrahim, 2017; Ascher and Noble, 2019; Zahra, 2020). Content moderation on the other hand is also based on data processing practices that allow lead firms to extract processed data from the outsourced firms. Information generated by user activities on social media requires processing by lead firms for it to be used for both generating advertisement revenue (Postigo, 2016) but also for infrastructure expansion processes (Mackenzie, 2019).

Significant to both these functions is the accuracy of content moderation decisions, underpinned by the logic of increasing platform predictivity, and therefore lead firms undertake specific organisational strategies to incorporate the social media user activities within the production processes. Given the research gap on these intersecting processes, our article asks: how do lead firms ensure content moderation accuracy within a limited time and on a large scale? The starting point of our labour process analysis is that services provision is generally characterised by high degrees of intangibility due to ‘asymmetric information, product differentiation and dynamic scale economies’ (Sapir, 1987 in Miozzo and Ramirez, 2003: 63). To ensure the quality of services, managerial strategies have to address uncertainty at the broader level and ‘labour indeterminacy’ (Smith, 2006; Thompson and Smith, 2009) at a more specific level. In the outsourced content moderation processes, the high level of uncertainty is caused by the relocation of service work, different (supplier) management and external technological infrastructure. Added to this is the issue of labour indeterminacy that is linked to the increased variation of user-generated content and the dynamic social media landscape on accounts of external influences on content moderation standards.

This article draws from the assessment that managerial strategies are underpinned by a ‘control imperative’ to reduce the indeterminacy gap (Thompson, 2010). With this, we mean that the managerial strategies are aimed at reducing the scope of error in content moderation work. Our analysis focuses on the use of automated technology for work organisation and control in the outsourced content moderation labour process. In so doing, the flexibility and versatility of content moderation technology is highlighted to show how lead firms are able to tap into the user activities on social media and integrate them into the labour process. The ensuing high degree of standardisation is supplemented with supply side managerial functions to ensure the quality service delivery or accurate moderation decisions. The combination of technical and bureaucratic control mechanisms is accompanied by reduced labour discretion over tasks and time, together with limited spaces for exercising resistance. From this perspective, there are two main contributions of our study: first, the significance of automated technology for governance in content moderation value chains, and second, worker experiences of technically-driven control of the labour process.

Our analysis is informed by two workshops with ten participants who were employed at a German-based IT-services company and worked on a content moderation project for one of the largest social media platforms. The workshops combined the design thinking or deductive reasoning method together with the focus group research method to allow for a retrospective perspective within a group setting. In applying a user (worker) design framework to show how work is organised and to include workers’ experiences of the technology, the article has been able to observe the role of regular updates in the content moderation technology for enabling the standardisation of work, and the resulting weakening of labour discretionary power.

The article is structured as follows. First, we evaluate the use of automated technologies in service work for labour control and worker experiences of knowledge and skill. Second, we refer to the user activities on social media and discuss them as customer-facing extensions of content moderation technologies. Next, the data collection process is described in detail including the background of the research participants. Thereafter, an analysis is presented on control using the content moderation technology, followed by a gradual decline in labour discretion regarding tacit knowledge and working hours. Finally, we discuss the labour market conditions for migrant workers in Germany and their implications for resistance.

IT-services sector: defined by standardisation

Content moderation processes are highly obscured from the public view and increasing criticisms of social media platforms and demands for accountability from lead firms include calls for transparency regarding the content moderation systems. Stakeholders often raise the question of legitimacy of content moderation workers, their proneness to errors and ‘cultural bias’ that get framed within the ‘decision-making responsibilities’ for user content (in Klonick, 2017). Challenging these simplistic claims of cultural bias to account for moderation errors, Ahmad (2019) shows the significance of training mechanisms in outsourced work and restrictions placed on workers’ culture. This refers to the standardised character of work that limits workers from exercising their cultural knowledge in the work. It also corresponds to the broader analysis of outsourced services where firms draw upon labour skills and knowledge that are culturally and linguistically near to their customers, but in a standardised manner (Remesh, 2008; Taylor, 2015).

To understand how the lead firm ensures the supply of local skills and knowledge on the one hand, but restricts the labour indeterminacies for managerial control on the other, our literature review is situated within the historical use of automated technologies in service value chains. Two arguments can be drawn for this preference: first, that we include the role of supplier firm and on-floor supervision for ensuring accurate moderation decisions. We can see that control in the gig economy (Gandini, 2019), particularly in reference to algorithmic control (Kellogg, Valentine, and Christin, 2019; Krzywdzinski and Gerber, 2021) has resulted in producing similar degrees of increased monitoring and reduced labour discretion. Our research however is interested in understanding how control in the content moderation labour process is a site of inter-firm coordination between the lead firm and the supplier. The second argument refers to our focus on call centres due to the similar degrees of labour indeterminacies observed in the content moderation work. While content moderation is back-office work with no interaction between the customers and the workers, the reference to the interactive call centre work shows how despite the significance of tacit knowledge, work is standardised using technology.

Qualitative advances in the information and communications technologies (ICTs) have allowed for radical changes in the information network services or the IT-services sector. The digitalisation of telecommunications networks followed by key innovations such as the automatic call distribution switches (ACD) and predictive dialling technologies have facilitated the spatial relocation of work to locations with labour cost arbitrage (Miozzo and Ramirez, 2003; Ellis and Taylor, 2006). The application of these technologies in the labour process have allowed lead firms the possibility for both automatically directing workers with fragmented tasks and monitoring them in real time, thereby replacing certain managerial functions (Taylor, 2015; Flecker and Schönauer, 2016). It may be said therefore that the use of technology to ‘minimize the production of transforming labour power into labour as well as to maximize the purely physical based possibilities for achieving efficiencies’ (Edwards, 1979: 112) is not just definitive of the transformation of the nature of work but also for developing sophisticated managerial control strategies (see Braverman, 1974; Noble, 1984).

Within the discussion on call centres, several examples from financial services and telecommunications show that technologies have brought about radical changes to clerical processing and customer servicing through work intensification and extension of working times (Ellis and Taylor, 2006; Carter et al, 2011). Using technological ‘systems of work measurement, targets and surveillance’ managers have developed more comprehensive strategies of disciplining workers and performance (Carter et al, 2011: 93). And rather than signalling an end point, the sophisticated call centre technologies are underpinned by the logic of continuous adaptation in the capitalist accumulation process (Ellis and Taylor, 2006). With the result that standardisation is the chosen strategy to ensure both capital accumulation and labour productivity (Howcroft and Richardson, 2012).

Several scholars have however cautioned against a simplistic reading of standardisation in services (in Flecker et al, 2013). Highly controlled work processes also involve labour agency in coping with the manifold situations in which standards and technical controls are insufficient (Smith and Thompson, 1998; Thompson, 2003). Some have also claimed that the degree of routinisation and standardisation is limited by the variability posed by customer requirements (Frenkel et al, 1995; Korczynski, 2002 in Ellis and Taylor, 2006: 109). There is much value in recognising the influence of ‘customer-oriented bureaucracy’ (Korczynski, 2002) in customer service work, most importantly because it shows how labour’s discretionary power over work-related knowledge and working hours is significantly shaped by their interactions with the customers. Fuller and Smith’s (1991) seminal work showed the use of ‘consumer feedback’ as a labour control mechanism where workers needed to continually utilise their tacit knowledge. Even in back-office work, Gabriel, Korczynski, and Rieder (2015) have attempted to address the ‘rise of consumers’ by attending to the significance of consumer knowledge within the sphere of production.

Clearly, with the rise of internet and mass communication consumers undertake important activities that have implications for an organisation’s business activities (Cova and Dalli, 2009; Ritzer et al, 2012; Dunkel and Kleemann, 2013; Gabriel and Lang, 2015). At stake here however is the increasing ‘convergence in the direction of standardisation’ caused by changes in technology (Houlihan, 2004: 68 in Ellis and Taylor, 2006: 109). Characterised by customer-facing functions and asymmetric information, the call centre industry has included technological innovations such as the self-dialled phones in the 1950s for segmenting tasks to users for the benefit of increasing call volume and labour productivity (Miozzo and Ramirez, 2003: 69). And even with variations among call centres in terms of ‘strategic choice, configuration, call centre type and market segment’, the organisation of work is underpinned by the logic of standardisation (Batt and Moynihan, 2002; Glucksmann, 2004 in Ellis and Taylor, 2006: 109). This refers to the extended control over the labour process by fragmentation, routinisation and the ‘removal of informal controls over output and time’ (Ellis and Taylor, 2006: 120).

Drawing from the above discussion on the role of technologies in services provision, we can see that there is general tendency towards standardisation of work. Given that social media platforms are unique socio-technical infrastructures in that they rely on end-users and multiple partners to develop parts of their product or service (van Dijck, 2013), we can see the increased variation of user-generated content and the multi-stakeholder influence on content moderation standards. To understand how lead firms are able to bridge the dynamic social media landscape with the content moderation production so as to eliminate waste, the following section refers to the media and communication scholarship on the continuous capture of user activities.2 Our analysis shows later that the resulting standardised process allows lead firms to realise economies of scale while at the same time ensuring quality service delivery.

Social media and content moderation: shaped by user input

In 2017, German regulators enacted the Netzwerkdurchsetzungsgesetz (NetzDG) or the Network Enforcement Act, with the aim of holding large technology firms accountable for user-generated content posted on their social media platforms. In the case that the content is found to be manifestly unlawful based on user complaints, firms are legally required to remove it within 24 hours or face a fine of up to €5 million.3 On the one hand, the regulation aims to limit the safe harbour for firms who have managed to enjoy non-liability for ‘third-party’ content generated by users on their social media platforms through Section 230 of the Communications Decency Act (CDA) in the United States and its European equivalent of the e-Commerce Directive (Directive 2000/31/EC) (Wang, 2018). On the other hand, it involves users in servicing the public discourse on social media. In this sense, the NetzDG immunises lead firms from liability as long as they lack the knowledge of infringing social media content.

The content moderation architecture is designed by lead firms to constitute the flagging mechanisms by which social media users can flag or report content that violates the community standards. Flags are observed as the first step in the moderation process, enabling lead firms a system of self-regulation at a vast and dynamic scale. Over one million pieces of worldwide content are flagged by users every day (Klonick, 2017). Flagging also functions as a ‘powerful rhetorical legitimation’ for moderating content on social media, according to the interests of user communities (Crawford and Gillespie, 2016: 412). Within the discussion on value extraction from the ‘unpaid labour’ of social media users (Fuchs, 2014; Beverungen et al, 2015), flags can be understood as ‘data points’ that are used in two ways: one, as ‘undetectable patterns of user sentiments’ and two, as controlled user judgements, that can be ‘handed off’ to the next stage in the moderation systems (Crawford and Gillespie, 2016: 412–13).

The design choices of social media platforms allow lead firms to control users’ expression of concerns and integrate this tailored information in the moderation process. In 2013, YouTube, the largest video sharing social media platform, introduced a feature in its flagging mechanism whereby users could indicate the ‘time code of the video’ where the offending material appears and describe ‘additional details about the offending content in 500 characters or less’ (Crawford and Gillespie, 2016: 414–15). Depending on the subject of concern, users are required to tick off the menus and sub-menus and fill in the required information (Figure 1). The integration of NetzDG option on the platform interface for users residing in Germany further enables the consolidation of information that depends on user knowledge of the context, publisher, and other details such as the time of and language in which the content has been published on the platform.

Youtube’s feature to report videos is designed to integrate detailed information on users’ expressions of concerns. In fact, users are obligated to select between different labels of violation. An optional selection can be made as to whether the reported content falls under the the German Network Enforcement Law. It is also required to provide a specific timestamp for the violation and to select the issue type and affiliation. A description of the complaint in your own words of up to 1500 characters is necessary. Providing the full name is optional.
Figure 1:

YouTube flagging pop-up window

Citation: Work in the Global Economy 2, 2; 10.1332/273241721X16647876031174

Source: YouTube

The flagging mechanisms on other big social media platforms such as Facebook and Twitter do not have such an ‘elaborate vocabulary’ of reporting as YouTube, but they still export the important work to users for ‘policing content, placating (other) users, and suggesting to external bodies that they represent a functioning system of self-regulation’ (Crawford and Gillespie, 2016: 418). Increasingly lead firms practice the ‘trusted flagging’ mechanism, where certain stakeholders such as the European Union’s Internal Referral Units, civil society organisations and third parties are privileged over individual users, for flagging content (Appelman and Leerssen, 2022). Close coordination with their trusted flaggers includes training sessions and workshops at the headquarters of YouTube, Twitter and Facebook (Jourová, 2019). Scholars have pointed out that while ‘trusted flagging does not scale’, it does ‘feed reference files directly into the platform’s automated removal logics’ (Appelman and Leerssen, 2022: 17). Correspondingly, there has been an increase in automated flagging and even moderation practices such as in the case of copyright and child assault-related content (Appelman and Leerssen, 2022).4

While flagging activities are conditioned by terms set by the technology firms, they do not translate into large scale ‘proactive moderation’ (Grimmelmann, 2015) practices given the reputation risks for lead firms and their relations with the end-users and multiple partners on social media platforms. Correspondingly, ‘reactive moderation’ (Grimmelmann, 2015), wherein the flagged content is transferred to workers employed in the supplier firms is an established practice (Klonick, 2017). By undertaking an inquiry of the technological interface with which content moderators interact, our research shows first how work is organised to include the knowledge of social media users and stakeholders, and second, how continuous standardisation is marked by work intensification and extension of working time. The resulting worker experiences include deskilling and reduced labour discretion regarding tasks and time. Although not the main focus of this article, we argue that the reduced resistance observed in our research is on account of labour market positions of the participating migrant workers, rather than a consequence of totalisation of technical control.

Methodology

The research setting was a German-based subsidiary of a German multinational conglomerate. The subsidiary is an IT-services company (SU – meaning supplier) and is contracted by the largest social media platform (LF – meaning lead firm) to provide content moderation services for its different language and regional markets. With covert practices of outsourcing content moderation and lack of reliable data specific to national labour markets, the exact size of the content moderation labour force in Germany remains unknown. Moreover, workers are required to sign non-disclosure agreements (NDAs) at the onset of their work. These NDAs are descriptive of the retribution mechanisms available to employers in the case that workers disclose work-related information to a third party. Across December 2020, two online workshops with ten research participants in total were conducted. Each workshop was scheduled for four and a half hours. Workers with former and current employment status at SU working on the content moderation project for LF (CMB-01 – CMB-10) were invited to take part in the workshop. Enquiries to managements at SU and LF remained unanswered. Due to the high level of secrecy surrounding their work and lack of trust, the online format provided for increased degrees of anonymity. This enabled workers to turn off their webcam and use a pseudonym to identify themselves.

Methods belonging to two different disciplines were brought together: a) the design thinking technique that provides a unique way to understand the human–machine interaction via ‘reverse engineering’ the labour process, that is, applying a deductive reasoning method to understand how the content moderation technology functions in it and to render a detailed model of the software’s user interface with its different pages, multiple menu options as well as location and function of buttons;5 and b) the focus group interviewing method that is unique in combining both the elements of the interview method and participant observation (Colucci, 2007). Against the backdrop of the high level of secrecy and the worker challenges of remembering the user interface specificities (Ahmad, nd), the focus group setting drew from collective memory especially regarding sensitive topics.6 The combination of both these methods has provided evidence on an essential feature of technology, that is, its regular updates, which underpinned by the integration of social media user knowledge in the labour process, enable the continuous standardising of content moderation work.

Most of the participants were approached on an international networking website for professionals. Out of the 200 participants approached, only six (CMB-01, CMB-02, CMB-05, CMB-08, CMB-09, CMB-10) confirmed that they would take part, with the remainder either not responding or citing reasons related to confidentiality. One participant (CMB-06) was already known to one of the co-authors of this article in the context of earlier research. The other three participants (CMB-03, CMB-04, CMB-7) were approached via snowball sampling. Each participant was remunerated with €100 for their engagement in the workshops. To ensure participation from different employment status, language background and work experience, prior communication was made with all potential participants. Additionally, we also conducted two separate interviews: one with a team leader of content moderators at SU and the other, with a representative of the trade union with membership at the works council at SU.

In preparation for the workshop and to get a broad demographic overview, all participants were requested to take part in a short online survey. Personal information such as age, nationality and language skills were requested along with work-related information about the length and status of employment and the type of content reviewed. The survey also played the function of informing the participants about the legal data protection provisions and for taking permission for collecting and processing the information from the survey and the workshop.

By making use of the design thinking method together with focus group interviewing, we undertook five activities in each workshop that could initiate discussions on a given topic, especially considering the sensitivity of the topic. Each activity sequentially drew from the previous one (also in Sinders and Ahmad, 2021). First, participants were requested to write down their daily workflows, which included details such as logging in the system, subsequent tasks, available functions and other broad information on the content moderation work process. Such an exercise was useful in deciphering that the software was web-based. Additionally, we discovered that apart from the moderation-related function, the software enabled internal communication, including email communication with the SU management and with the team members.

The second activity drew from the first one where participants had provided the general workflow to us. This exercise did not elicit as much information as the authors had hoped for, possibly because the participants had already detailed the software workflows and preliminary layouts in the first exercise. The third activity focused on granular elements of the different software pages that were visible to our participants during their work. This includes the visibility of corporate logos of LF or SU in the software, the different menu options, location and functions of buttons, and other options. Together with this, broader questions relating to the management control embedded in the software and workers’ perception of it were discussed.

In the fourth activity, participants were questioned about surveillance in the workplace and monitoring strategies that they knew were embedded in the software. And finally, the fifth activity discussed the question of labour agency and the possibilities for exercising resistance through technology. All activities required active participation from workers by drawing out illustrations of their software interface and submitting the drawn images during the workshop, using email or social media. Regarding activities four and five that were more sensitive in nature, we made use of ‘projective techniques’ (Colucci, 2007) such as an interactive whiteboard called Jamboard, wherein the participants could anonymously share their perspectives, in the form of text, images and other formats. All the participants gave textual responses.

The use of a method-based interdisciplinary approach to study the given research questions at hand has generated previously unknown information on the human–machine interaction in the content moderation labour process. We took great care to make our interview questions comprehensible to the participants, provide them more time along with visual prompts to draw out the wireframes, and enable more channels, such as email and personal messaging applications, to submit the drawn images to us. Due to high sensitivity of the research questions and potential retribution for workers, all anonymisation and security related protocols were strictly maintained during both phases of data collection and analysis.

Workers’ background and motivations

In terms of the participant composition (Table 1), all workers are migrants and nationals of Brazil, Finland, Iran, Italy, Portugal and Turkey, respectively. They constitute a relatively young workforce between 26 years and 40 years of age. With only a single exception, our participants did not have proficiency in the German language and English was a second or third language for most of them. Their educational qualifications include marketing studies, design studies, linguistics and their work experiences encompassed customer service work, language teaching, web and user interface designing. Six of them were employed on non-permanent or limited contracts for up to two years, while the other four had unlimited contracts. While 60 per cent of our participants were full time employees, the rest were part timers.7 Seven of the ten participants took part in their capacity as former content moderators, while the remaining three continued to be employed during the time of their participation. Because of their different language skills, our participants moderated content for specific language markets, including Finish, Italian, Persian, Portuguese and Turkish markets. Some also moderated content in English and one participant also moderated content in German after having moderated content in Portuguese for more than a few months.

Table 1:

Participants’ information on employment status, language specialisation and work experience

S. No. Employment status Full time or part time Language specialisation Work experience (in years)
CMB-01 Former Full time Finnish > 1
CMB-02 Former Full time Italian < 4
CMB-03 Current Part time Persian < 3
CMB-04 Current Part time Persian < 1
CMB-05 Current Full time Turkish < 2
CMB-06 Former Full time Persian < 1
CMB-07 Former Part time Persian < 1
CMB-08 Former Full time Italian < 2
CMB-09 Former Full time Portuguese < 3
CMB-10 Former Full time Portuguese, German < 1

Economic security and material conditions were cited as the primary motivations for participants to apply for content moderation work. Even though the wages for this work are low, many participants confirmed that economic necessity along with the immigration related issues persuaded them to work at SU. In most cases, the visa status was dependent on their employment contract. Two of our participants were also in the process of applying for an asylum status in Germany, which required them to not leave the country until they received their status.8 In the absence of German language skills, participants were unable to apply for better paying jobs in other sectors.

Our participants were recruited at SU based on generic skills such as communication, basic computer knowledge and language proficiency for the respective consumer markets. They did not have information on what the work would entail as they applied using generic job advertisements for the content moderation work. All participants saw this job as a stop-gap arrangement before moving to another career or educational opportunity. The gross monthly wage (that is, before taxes) of a full-time moderator at SU is between €1400 and €1500 and for part-time moderators, it is between €900 and €1000. In terms of the health insurance, half of it was covered for full-time workers according to German legal regulations of payroll taxes and employer’s contribution (Arbeitgeberbeitrag) and the part-time workers received no health insurance coverage (CMB-03, CMB-04, CMB-07). Apart from the monthly wages, content moderators receive bonus payments during night shifts. Content moderation work at SU runs 24 hours and seven days a week, with workers employed in three different shifts across a day. Excluding night-time bonuses, the monthly wage roughly corresponds to the statutory minimum wage in Germany and can therefore be described as a case of a low-income level.

Work organisation

Technology is integral to labour control in the content moderation process at SU and the transformation of moderators’ labour power into productive labour is mediated using a browser-based software provided by the client. The three main operational interfaces of this software are a) HR Platform wherein workers can access emails from SU management and other human resources related aspects such as leaves and performance metrics, b) Task Interface wherein the moderators can access the tasks, and c) Workplace Chat interface wherein workers can communicate with their team members. Control over the main software for work is centralised with LF, whereas the HR Platform is managed by SU.9

The SU office comprises of different language floors and depending on their language qualification and other skills, workers are assigned to specific language markets and content queues during the mandatory training from SU. This training is provided at the onset of their work and lasts between two and four weeks. The main structure of the content moderation software remains the same for all the different language floors. This includes the common HR Platform, and the general features of the Task Interface that consists of three important elements: the content queue menu, the policy section and performance metrics section. At the same time, the user interface of the software modules differs according to different language markets and content queues. The different content queues in which flagged content arrives at SU are namely nudity, hate speech, user comments, fake news, spam and news articles. They constitute different categories of content, such as text, video and image, and are enlisted with different colour gradations highlighting the number of new tickets in the queues. Content queues showing the highest number of tickets get automatically assigned to the workers.

The highly confidential nature of content moderation outsourcing can be observed in the organisation of work. Both the physical setting of the workplace and the labour process are designed to control workers’ movement and interaction. Every language floor constitutes a couple of teams and these teams are separated from each other and members are barred from interacting across the teams. They are required to wear their work identification card at all times and are not allowed to carry any personal items to the workspace. As they work with long and detailed content moderation policies, many participants recounted the need to carry pen and paper for making notes on the policies, this however is not allowed (CMB-02, CMB-04, CMB-05, CMB-06, CMB-07, CMB-10). Instead, workers can access the policies using the Task Interface function that opens in another page.

An important feature of the content moderation labour process is the continuous updating of the content moderation technology, which allows for adapting to the uncertain characteristics of social media user content and externally-influenced moderation standards. These updates have implications for the moderation labour process in two main ways: one, in terms of the increasing routinisation of tasks and two, in terms of the codification of time. While the updates are followed by short training sessions at SU during the weekly meetings with team leaders, our research provides evidence on the continuous standardisation of content moderation work and worker experiences of deskilling. Underpinned by the logic of fragmentation and codification of knowledge, the managerial strategies of control on task and time are described in detail below.

Routinisation of tasks

Content queues showing the highest number of tickets get automatically assigned to the workers. Where workers could earlier choose between different queues, this possibility was later removed with an update in the software. At the opening of each content queue, workers can click on a particular ticket which opens into a new tab called the Ticket Window wherein the moderation decisions are made. It consists of the task area with the flagged content and sometimes further information about the related users. Depending on the language market and content queue, the interface can have the decision tree on the left or the right side. The decision tree is essentially a list of labels and sub-labels which have been predeveloped by LF according to the moderation policies. Using the decision tree, moderators must choose the appropriate labels and if required, the sub-label, for the respective content.

In the case that the content is labelled with a high concerning option such as child abuse, it will automatically get deleted. In case it does not, then the moderator can manually do it using the delete option on the bottom of the page. Most of the times, workers simply label the content, and they are automatically moved on to the next ticket. The delete option for content is often used with caution by workers because it can have repercussions for LF’s customer relations. The apprehension towards deleting content also comes from concerns related to workers’ performance metrics. Workers are only allowed to make a few mistakes every month and in case they delete a content incorrectly, either they cannot go back to change their decision using the reverse button or must change the decision within five seconds of having been automatically routed to the new ticket window.

In this way, the software functions as a means to tightly control the allocation of work tasks, first by separating the content into different content queues and automatically assigning them to the workers. Second, with the predetermined ways of labelling content, the technology further directs the labour process in a highly standardised manner, leaving little space for individual discretion. Depending on the queue, workers may be able to see content-related context such as information on the user who posted the content and also the user who flagged the content. Before a software update, workers were able to see both the profile picture and names of users in the hate speech queue, but this was later changed to reveal only the names of the users. At the same time, in some queues, workers also receive additional information which they either do not have the time to use or do not know how to use, for example the specific location and time at which the content was posted. In those situations where workers are confused about a moderation policy label, they can go back to the HR Platform page and read the policy documents. They can also consult with their team leaders who are always present on the floor and subject matter experts who are content policy experts who are often also physically present. However, because of high target numbers, they have limited time to consult, and mostly make use of the Workplace Chat tool to quickly consult with their team members.

Codification of working time

Workers are assigned with certain quantified targets, which means that they need to correctly moderate a certain amount of content every month. Their individual performance is monitored against both the average handling time per ticket and the total production time. Workers’ time starts tracking as soon as the In-Production tab is opened using the Activity Code button on the HR Platform. Each page has the Activity Code button that workers are required to click and select their productivity status. Apart from the Idle Code (which means that the workers are not engaged in any productive task) and the In-Production code, there are also codes relating to going for lunch break, leaving the desk for other tasks, taking a short break, and for meeting with the management or team leaders.

In this way, the firms strive to monitor every aspect of the working day. Further on, there are weekly and monthly schedules for the content moderators to follow, therefore assigning a certain time for each activity. Time tracking is also made visible to the workers, usually at the top of the Task Interface and the Ticket Window interfaces. In addition to the time tracker, they can also see their performance metrics such as accuracy, working speed, number of tickets completed and others. While we can see that the routinisation of tasks and codification of working time facilitates the high standardisation but also individualisation of the work, at the same time, team-based work is observed with small teams of 15 to 20 members. Each team is assigned a team leader who plays the part of a ‘foreman’ (Braverman, 1974) in resolving work-related queries but also monitoring workers on the floor and pressurising them regarding productivity time. In this manner, instead of technology eliminating close supervision and obscuring social relations, team leaders in fact accentuate the monitoring work of technology by ensuring both the quantitative targets or the average handling time per ticket and qualitative targets.

Ensuring labour productivity by controlling working hours is primarily undertaken using the Activity Codes system wherein if workers are logged into Idle Code, their productive hours are deducted. Further on, monthly reports based on individual performance metrics are also used by the SU management for extending or revoking the employment contracts. Managerial control over workers’ time has also introduced certain updates that restrict workers from accessing external websites such as YouTube and SoundCloud, which some workers used for listening to music and for alleviating stress during work. Muting the video-based flagged content and using music in the background was seen as a self-help measure by participants for protecting themselves against the shooting or screaming sounds in the video (CMB-06, CMB-08).

Deskilling and labour agency

Our article goes on to argue that the decline in their discretionary power over 1) tasks and 2) working hours, has resulted in labour deskilling. Before certain updates, workers were often required to apply tacit knowledge over content where the specific policy label was not available in the software, or the content was confusing. Having not acquired the appropriate corrective from their immediate supervisors, workers exercised their individual agency and created their own workflows based on their own judgements. This was however possible with less time pressure by management. However, with the technological updates on their work time and tasks, workers have been restricted from using their discretion.

Instead of making work more efficient for the workers, the automatic routing of tickets with designated choices for moderating the content has created more challenges. The high-ticket volume, unclear information in the policy and software related errors, especially during updates, were of particular concern to our participants (CMB-02, CMB-05, CMB-09). In addition to this, they do not have control over their working hours and how much time they can spend on each activity, regardless of the target completion. Many of our participants were observed to be recounting patronising experiences and being treated like a child.

‘You can’t even go to toilet for more than 15 minutes in a row. I got into an argument (with the team leader) many times about this toilet time thing. Like if you want to go for longer, it is like a kindergarten. It is really something between a kindergarten and a police station, the rules are just ridiculous.’ (CMB-06)

With restrictions placed on their working hours, workers are unable to spend the extra time to go through the extensive policies or consult with their supervisors. This should be read in light of the reduced number of cases that require different workflows. Considering that flags act as repositories of content knowledge based on user information and context, we can address the codification of user knowledge together with labour’s tacit knowledge in the content moderation software. By fragmenting the content moderation work process and transferring the maximum number of tasks between unpaid users and outsourced workers, technology companies can extract value efficiently with limited waste in terms of performance-related errors due to communication (between workers or between workers and users).

Despite the codification of user knowledge in the automated systems and high task routinisation, participants recounted instances of still being able to create separate workflows. As Littler and Salaman (1982) note that while the introduction of technology allows the reduction of labour discretion, it never gets ‘reduced to zero’. Having not acquired the appropriate corrective from their immediate supervisors, workers exercised their individual agency based on their own judgements. But instead of perceiving these situations as empowerments, workers saw them as stressors because it affected their content average handling time. In fact, the consistent visibility of performance metrics, creates immense pressure and stress for the workers and ensures that their performance metrics remain high.

The extension of working time through the introduction of a time tracking update in the software, together with increased routinisation of tasks and restriction on content-related context has the effect of internalisation of pressure to work. Before several software updates, content moderators found it less difficult to question the quality reports on their work made by the quality analysts or team leaders. But ‘the systematic preparation and presentation of statistics’ (Callaghan and Thompson, 2001) has reduced the confrontation with supervisors at SU. At the same time, workers are conscious of the technology-related errors that create challenges for the workers. Cases of productivity time being affected by software problems such as automatic switching of Activity Codes without notice (CMB-03, CMB-04) and problems in time tracking such as the time tracker automatically pausing during production hours (CMB-05) created troubles for the workers. And even when the software bugs were fixed and updated, bad working practices were blamed on ‘faulty’ supervision by team leaders (CMB-04, CMB-09).

Discontent with their immediate supervision, one of the teams managed to put collective pressure on the SU management to replace their team leader. While the process took almost two years to completion, CMB-07 noted, ‘for us they could change the team leader, but it did not make it better actually. Not worse as well, but not better.’ Exercising collective agency is also immensely difficult because of increasing managerial restriction on workers’ movement. Much of the internal and informal sharing of knowledge between workers of a team takes place on the Workplace Chat tool due to time constraints but also because of the presence of team leaders in the workplace. This further creates feelings of constant surveillance where workers are concerned about supervisors tapping into their private communication with colleagues. This specific organisation of work serves the client and micromanagement at a distance, ensuring labour productivity, and keeping resistance at bay.

Most importantly, by automatically directing workers with tasks and monitoring them in real time, technology firms are able to centralise control without being visible to the workers. Correspondingly, employment conflict is prevalent within the SU-labour relations and does not implicate LF in it. This is again a result of the technological design that ensures obscurity of the client and the suppliers are enforcers of this security. At the same time, governance mechanisms between LF and SU necessitate physical meetings and in the initial years of the outsourcing contract to SU, representatives from LF would often visit the work floors. In one instance, participant CMB-06 reached out to the LF representative due to the problematic redressal mechanisms at SU:

‘One time, there was huge meeting with the client representative and I told him about my problem with one of the floor managers. During work, the manager would not allow me to move from my table to talk to colleagues and I felt like being chained to the table. The representative was like, “Oh my god, who was it?” I told him the name of my manager. And two days later, the head of the site (SU) came to me and started a fight, accusing me of looking for problems. It was horrific. There is no point talking to the client.’

Dissatisfaction among workers on accounts of deskilling and low wages gets amplified by labour market precarity and as a response, they undertake forms of embodied ‘emotional labour’ (Hochschild, 1983) to cope with and navigate their working conditions. In this sense, by completing individual targets and maintain high performance, workers are actively consenting to the valorisation interests of capital. But at the same time, this labour agency is also directed towards ‘combating the devaluation of their work and the disposibility they might feel’ (Mirchandani and Hande, 2020: 277). Unsurprisingly, several participants were observed suggesting solutions during the workshops for making their work more efficient. This includes the introduction of certain control elements in the technology such as zooming in and out for analysing the content details (CMB-02, CMB-05) and making notes on work such as summarising policy points (CMB-02, CMB-04, CMB-05). An anonymous quote on Jamboard during the fifth activity – on improving the software – in the first workshop reads:

Accessibility to the policies and resources should be enhanced to perform (work) more efficiently. And the software should become user friendly regarding the purposes of our jobs and the expectations from the moderators.

But most importantly, participants find managerial control on their working time and restriction on content-related information as excessive and non-productive. Several workers were found asking if the changes in the software were really ‘necessary’ (CMB-01, CMB-02, CMB-03, CMB-05, CMB-06, CMB-07, CMB-08). Taking cognisance of their reduced ‘structural power’ (Wright, 2000) in the labour process, workers get disillusioned with the SU management and lose confidence in collective bargaining and possibilities for meaningful change. Resultingly, the attrition rate in this work is very high and the average work tenure lasts up to two years. The following section further elaborates on this account of reduced resistance by looking at broader institutional arrangements that affect labour agency.

Labour resistance: limited by structural conditions

The research findings are not reflective of explicit resistance practices by content moderators. At the same time, certain observations can be made regarding the structural conditions for migrant workers, which limits their labour agency to coping strategies. By studying the development of low-wage employment in Germany, Krings (2021) has observed that a disproportionately higher number of migrant workers are employed in services. The author argues for a comprehensive examination of qualifications and skills together with employer behaviour, local labour market institutions and immigration laws to understand the position of migrant workers in the labour markets (McGovern, 2007; MacKenzie and Forde, 2009; Anderson, 2010 in Krings, 2021: 529). In our research, two important factors stood out for the limited resistance observed at SU: first, the education and skill requirements in Germany, and second, the national immigration regulations. The formal recognition and certification of skills in Germany has been well observed by researchers to create barriers for migrants holding foreign qualifications (Bauder, 2005; Etzold, 2017; Krings, 2021). In the absence of proficiency in the German language, many participants were unable to apply for better paying jobs in other sectors and their position in the labour market was even more weakened. Consequently, they were only eligible for and pressured into working in low-skilled and low-paid jobs.

The other important factor concerns immigration regulations in Germany. Participants were found to repeatedly express concerns regarding the renewal of employment contracts and their visa statuses (CMB-04; CMB-07). Two of our participants were also in the process of applying for an asylum status in Germany, and they feared that resistance and organising at work could negatively affect their asylum decisions. ‘Everyone is really in survival mode in this job. Students, people on different VISAs or asylum seekers, so it is really difficult. No one wants to risk, and the company knows that you are desperate and that you won’t say anything’, noted one workshop participant (CMB-06). Even though the German constitution protects the right of the workers to be members of a trade union and also to constitute and elect a works council in their firms, our research shows that among the 1000-plus content moderators employed at the German supplier, only 30 are members of the trade union and only five of these are members of the works council (interview with trade union representative at SU).

In the year 2020, both the Italian and Portuguese language projects were shut down at SU and moved by the social media client LF to another supplier located in Portugal. Two of our research participants (CMB-02 and CMB-09) lost their jobs because of the project relocation. There were no instances of organised resistance observed. Being a member of the trade union, CMB-02 was at least provided with a €1000-worth of compensation, which was not the case for CMB-09. A third participant (CMB-10) moved to the German language market on account of having trained themself in the German language and noted that ‘you are in a new country, and you need to survive’. It can be concluded that immigration mechanisms and language barriers, combined with the characteristics of the German skill certification, have negative effects on migrant workers and serve the function of enhancing managerial control and keeping resistance at bay.

Conclusion

Content moderation is key to maintaining some semblance of public discourse on social media and for extracting processed data for generating advertisement revenue and expanding the platform infrastructure. At stake here is the question of accuracy within a limited time and at a large scale, which is addressed by the lead firm through the use of automated technology. Our labour process analysis is informed by a rich literature on outsourced IT-services with a focus on call centres, providing evidence on how technically driven control facilitates the extraction of labour productivity in services characterised by high labour indeterminacies. Our research takes a step further and shows the strategic integration of social media users’ knowledge in the labour process using automated technologies resulting in routinisation of tasks and codification of time. This technical control is combined with supply side functions to ensure quality service delivery at a large scale. From this perspective, we make two main contributions to the labour process theory: first, the significance of automated technology for governance in content moderation value chains, and second, reduced labour discretion on account of technically driven control of the labour process. Rather than a consequence of totalisation of technical control, the reduced resistance and organised bargaining is on account of low labour market position of the participating migrant workers.

The findings have been drawn using a mixed-method approach with user interface design and focus group interviewing methods to analyse the use of technology for reducing labour indeterminacy and uncertainty at a German-based IT-services firm. By applying a user design framework for analysing the organisation of work and workers’ experiences of the technology, our analysis is able to tap into an important development in the content moderation labour process, that is, the role of regular updates in the technology. Although a novel approach for studying the human–machine interaction in an otherwise lesser-known labour process, there are limitations of our methodology. First, our sample size is limited to ten people and it is not representative of the workforce at the supplier firm. While our participants noted that the large majority of workers at the firm are migrants, we are unable to verify this claim due to high degrees of obscurity of this work. Second, in terms of the software updates that have informed our analysis of continuous changes to work, a longitudinal study can provide more details on these updates. At the same time, this could be challenging considering the high attrition rate in the work considering that 70 per cent of our research participants are former employees.

Our research findings offer opportunities for further research along two strands: first, regarding the future of content moderation labour. With the objectives of dynamic scaling and high fault tolerances, technology firms have built social media platforms upon specially developed software that enables modularisation of service delivery and developing parts of the software architecture independently (Ziegler, 2021). Future research can explore in detail the intersections between the content moderation technology and the platform architecture of social media, and examine how this could result in reducing labour indeterminacies to further degrees. This would also mean linking further the labour process analysis with the content moderation value chain dynamics, and examining the implications for power asymmetries. In her ongoing dissertation on content moderation work in India, this article’s co-author includes the case of a supplier exit from our target lead firm and examines how the lead firm-owned automated technology is not only crucial for limiting supplier upgrading within market competition but also for transferring the work along with workers to another firm during exit (Ahmad, nd).

Second, our findings offer an opportunity for retrospection over situating labour market condition as an important factor in the labour process analyses (Krzywdzinski and Gerber, 2021). This includes looking further into how immigration policies, language barriers and education and skills together with IT-services sector-based challenges such as frequent restructuring and exit, can explain structural constraints for collective bargaining at the workplace (Holtgrewe and Doellgast, 2012). Additionally, given the increase of direct actions such as ‘wildcat’ strikes are popular strategies among migrants in industrial conflicts (Birke, 2022; Fairwork 2022), further research needs to unpack the complex relations between low-paid migrant workers and traditional bargaining institutions in Germany. In this vein, our suggestions for future research are directed towards examining the conditions for strengthening intersectional organising and solidarity in precarious occupations.

Notes

1

The title of the new book can be roughly translated as a new structural change of the public sphere and deliberative politics, building upon his habilitation thesis on the structural change of the public (Habermas, 1962).

2

Waste here refers to performance-related errors due to information sharing (between workers or between workers and users).

3

Additionally, those companies receiving more than 100 complaints a year are legally required to submit bi-annual transparency reports about their content moderation practices. Article 1, Section 3, Network Enforcement Act.

4

Instances of fully-automated proactive moderation include technologies such as the Microsoft’s image-recognising algorithm PhotoDNA and YouTube’s copyright scanner ContentID (Ith, 2015 in Klonick, 2017).

5

One of the study designers has repeatedly conducted research in the field of user interface design and has considerable experience in it.

6

Prior to the workshops, individual telephonic conversations were made with the participants to build trust and to ensure functional diversity in the workshops. These conversations also touched upon the question of software interface and several workers mentioned challenges in remembering the details.

7

According to the ILO (1994), ‘the term part-time worker means an employed person whose normal hours of work are less than those of comparable full-time workers’. The average working hours per week for full-time employees in Germany was 41 hours in 2019 (Eurostat, 2020). The work contract is provided for one year and then after every half a year, it must be renewed. After three years of full-time work, according to the German law, the work contract is to become permanent. However, before the permanency of the work contract many workers are either fired within the probation time (Probezeit) or leave within the probation time.

8

Due to the precarious status connected with immigration and asylum procedures, we do not make a specific reference to who these participants are.

9

Additionally, ongoing research by this article’s co-author has discovered that both the Task Interface and Workplace Chat interfaces remain same in the content moderation work which LF has outsourced to an Indian supplier company.

Acknowledgements

We would like to first thank the participants of this study for their valuable time and grit in view of immense managerial control. We would also like to thank Martin Krzywdzinski, Virginia L. Doellgast, Paul Thompson, Florian Butollo, Robert Koepp and other colleagues at the Globalization, Work and Production department at the WZB Berlin Social Science Center for their valuable comments and feedback. Furthermore, the workshops informing this article would not have been possible without the ideas and support of Caroline Sinders and Michelle Nagel, former colleagues at the Weizenbaum Institute for the Networked Society and the WZB Berlin Social Science Center, respectively.

Conflict of interest

The authors declare that there is no conflict of interest.

References

  • Ahmad, S. (nd) Work and Workers in Content Moderation Value Chains [Unpublished doctoral dissertation], Berlin: Free University Berlin.

  • Ahmad, S. (2019) ‘It’s Just the Job’: investigating the influence of culture in india’s commercial content moderation industry, preprint, SocAXiv. doi: 10.31235/osf.io/hjcv2

    • Search Google Scholar
    • Export Citation
  • Ahmad, S. and Krzywdzinski, M. (2022) Moderating in obscurity: how Indian content moderators work in global content moderation value chains, in M. Graham and F. Ferrari (eds) Digital Work in the Planetary Marke, Cambridge, MA: The MIT Press, International Development Research Centre, pp 7795.

    • Search Google Scholar
    • Export Citation
  • Anderson, B. (2010) Migration, immigration controls and the fashioning of precarious workers, Work, Employment and Society, 24(2): 30017. doi: 10.1177/0950017010362141

    • Search Google Scholar
    • Export Citation
  • Appelman, N. and Leerssen, P. (2022) On ‘trusted’ flaggers, December. https://law.yale.edu/sites/default/files/area/center/isp/documents/trustedflaggers_ispessayseries_jul2022.pdf.

    • Search Google Scholar
    • Export Citation
  • Ascher, D.L. and Noble, S.U. (2019) Unmasking hate on Twitter: disrupting anonymity by tracking trolls, in Free Speech in the Digital Age, New York: Oxford University Press. doi: 10.1093/oso/9780190883591.003.0011

    • Search Google Scholar
    • Export Citation
  • Barrett, P.M. (2020) “Tech - Content Moderation June 2020.” NYU stern center for business and human rights. June 2020, https://bhr.stern.nyu.edu/tech-content-moderation-june-2020.

    • Search Google Scholar
    • Export Citation
  • Batt, R. and Moynihan, L. (2002) The viability of alternative call centre production models, Human Resource Management Journal, 12(4): 1434. doi: 10.1111/j.1748-8583.2002.tb00075.x

    • Search Google Scholar
    • Export Citation
  • Bauder, H. (2005) Institutional capital and labour devaluation: the non-recognition of foreign credentials in Germany, European Journal of Economics and Economic Policies: Intervention, 2(1): 7593. doi: 10.4337/ejeep.2005.01.09

    • Search Google Scholar
    • Export Citation
  • Beverungen, A., Böhm, S. and Land, C. (2015) Free labour, social media, management: challenging marxist organization studies, Organization Studies, London: SAGE Publications.

    • Search Google Scholar
    • Export Citation
  • Birke, P. (2022) Grenzen aus Glas: Arbeit, Rassismus und Kämpfe der Migration in Deutschland, Wien: Mandelbaum Verlag eG.

  • Braverman, H. (1974) Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century, New York and London: Monthly Review Press.

    • Search Google Scholar
    • Export Citation
  • Callaghan, G. and Thompson, P. (2001) Edwards revisited: technical control and call centres, Economic and Industrial Democracy, 22(1): 1337. doi: 10.1177/0143831X01221002

    • Search Google Scholar
    • Export Citation
  • Carter, B., Danford, A., Howcroft, D., Richardson, H., Smith, A. and Taylor, P. (2011) ‘All they lack is a chain’: lean and the new performance management in the British civil service. doi: 10.1111/j.1468-005X.2011.00261.x

    • Search Google Scholar
    • Export Citation
  • Eurostat (2020) Durchschnittliche Wochenarbeitszeit Von Vollzeitbeschäftigten Erwerbstätigen in Deutschland von 2009 bis 2019 nach Geschlecht [Graph], Statista.

    • Search Google Scholar
    • Export Citation
  • Chen, A. (2012) Inside Facebook’s outsourced anti-porn and gore brigade, where Camel toes are more offensive than crushed heads, Gawker, https://www.gawker.com/5885714/inside-facebooks-outsourced-anti-porn-and-gore-brigade-where-camel-toes-are-more-offensive-than-crushed-heads.

    • Search Google Scholar
    • Export Citation
  • Colucci, E. (2007) ‘Focus groups can be fun’: the use of activity-oriented questions in focus group discussions, Qualitative Health Research, 17(10): 142233. doi: 10.1177/1049732307308129

    • Search Google Scholar
    • Export Citation
  • Cova, B. and Dalli, D. (2009) Working consumers: the next step in marketing theory?. doi: 10.1177/1470593109338144

  • Crawford, K. and Gillespie, T. (2016) What is a flag for? Social media reporting tools and the vocabulary of complaint, New Media & Society, 18(3): 41028. doi: 10.1177/1461444814543163

    • Search Google Scholar
    • Export Citation
  • Dijck, J.van. (2013) Facebook and the engineering of connectivity: a multi-layered approach to social media platforms, Convergence, 19(2): 14155. doi: 10.1177/1354856512457548

    • Search Google Scholar
    • Export Citation
  • Dunkel, W. and Kleemann, F. (2013) Customers at Work: New Perspectives on Interactive Service Work, London: Springer.

  • Ebrahim, S. (2017) There is no freedom of speech on Facebook, activists say, The Daily Vox (blog), (Accessed: 6 Dec 2017), https://www.thedailyvox.co.za/there-is-no-freedom-of-speech-on-facebook-activists-say-shaazia-ebrahim/.

    • Search Google Scholar
    • Export Citation
  • Edwards, R. (1979) Contested Terrain: The Transformation of the Workplace in the Twentieth Century, New York: Basic Books.

  • Ellis, V. and Taylor, P. (2006) ‘You don’t know what you’ve got till it’s gone’: re-contextualising the origins, development and impact of the call centre, New Technology, Work and Employment, 21(2): 10722. doi: 10.1111/j.1468-005X.2006.00167.x

    • Search Google Scholar
    • Export Citation
  • Etzold, B. (2017) Capitalising on asylum – the reconfiguration of refugees’ access to local fields of labour in Germany, Refugee Review, 3: 82102.

    • Search Google Scholar
    • Export Citation
  • Fairwork. (2022) Fairwork Germany Ratings 2021: Labour Standards in the Platform Economy, Berlin, Germany and Oxford, UK.

  • Flecker, J., Haidinger, B. and Schönauer, A. (2013) Divide and serve: the labour process in service value chains and networks, Competition & Change, 17(1): 623. doi: 10.1179/1024529412Z.00000000022

    • Search Google Scholar
    • Export Citation
  • Flecker, J. and Schönauer, A. (2016) The production of ‘placelessness’: digital service work in global value chains, in J. Flecker (ed) Space, Place and Global Digital Work, London: Palgrave Macmillan, pp 1130.

    • Search Google Scholar
    • Export Citation
  • Frenkel, S., Korczynski, M., Donoghue, L. and Shire, K. (1995) Re-constituting work: trends towards knowledge work and info-normative control, Work, Employment and Society, 9(4): 77396. doi: 10.1177/095001709594008

    • Search Google Scholar
    • Export Citation
  • Fuchs, C. (2014) Digital Labour and Karl Marx, New York: Routledge.

  • Fuller, L. and Smith, V. (1991) Consumers’ reports: management by customers in a changing economy, Work, Employment and Society, 5(1): 116. doi: 10.1177/0950017091005001002

    • Search Google Scholar
    • Export Citation
  • Gabriel, Y., Korczynski, M. and Rieder, K. (2015) Organizations and their consumers: bridging work and consumption, Organization, 22(5): 62943. doi: 10.1177/1350508415586040

    • Search Google Scholar
    • Export Citation
  • Gabriel, Y. and Lang, T. (2015) The Unmanageable Consumer, London: SAGE.

  • Gandini, A. (2019) Labour process theory and the gig economy, Human Relations, 72(6): 103956. doi: 10.1177/0018726718790002

  • Gillespie, T. (2018) Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media, New Haven, CT and London: Yale University Press.

    • Search Google Scholar
    • Export Citation
  • Gillespie, T. (2020) Content moderation, ai, and the question of scale, Big Data & Society, 7(2): 2053951720943234. doi: 10.1177/2053951720943234

    • Search Google Scholar
    • Export Citation
  • Glucksmann, M.A. (2004) Call configurations: varieties of call centre and divisions of labour, Work, Employment and Society, 18(4): 795811. doi: 10.1177/0950017004047965

    • Search Google Scholar
    • Export Citation
  • Grimmelmann, J. (2015) The virtues of moderation, Yale Journal of Law and Technology, 17: 42.

  • Habermas, J. (1962) Strukturwandel Der öffentlichkeit, Neuwied, Berlin, Vol 6, pp 4275.

  • Habermas, J. (2022) Ein Neuer Strukturwandel der Öffentlichkeit und die Deliberative Politik, Berlin: Suhrkamp Wissenschaft, https://www.suhrkamp.de/buch/juergen-habermas-ein-neuer-strukturwandel-der-oeffentlichkeit-und-die-deliberative-politik-t-9783518587904.

    • Search Google Scholar
    • Export Citation
  • Hochschild, A.R. (1983) The Managed Heart: Commercialization of Human Feeling, Berkeley, CA: University of California Press.

  • Holtgrewe, U. and Doellgast, V. (2012) A service union’s innovation dilemma: limitations on creative action in German industrial relations, Work, Employment and Society, 26(2): 31430. doi: 10.1177/0950017011432913

    • Search Google Scholar
    • Export Citation
  • Houlihan, M. (2004) Tensions and variations in call centre management strategies, Call Centres and Human Resource Management, London: Palgrave Macmillan, pp 75101.

    • Search Google Scholar
    • Export Citation
  • Howcroft, D. and Richardson, H. (2012) The back office goes global: exploring connections and contradictions in shared service centres, Work, Employment and Society, 26(1): 11127. doi: 10.1177/0950017011426309

    • Search Google Scholar
    • Export Citation
  • Ith, T. (2015) Microsoft’s PhotoDNA: protecting children and businesses in the cloud. Stories, https://news.microsoft.com/features/microsofts-photodna-protecting-children-and-businesses-in-the-cloud/.

    • Search Google Scholar
    • Export Citation
  • Jourová, V. (2019) How the code of conduct helped countering illegal hate speech online, Directorate-General for Justice and Consumers. European Commission, https://ec.europa.eu/info/sites/default/files/hatespeech_infographic3_web.pdf.

    • Search Google Scholar
    • Export Citation
  • Kellogg, K.C., Valentine, M.A. and Christin, A. (2019) Algorithms at work: the new contested terrain of control, Academy of Management Annals, 14(1): 366410. doi: 10.5465/annals.2018.0174

    • Search Google Scholar
    • Export Citation
  • Klonick, K. (2017) The new governors: the people, rules, and processes governing online speech, Harvard Law Review, 131: 73.

  • Korczynski, M. (2002) Human Resource Management in the Service Sector, Basingstoke: Palgrave.

  • Krings, T. (2021) ‘Good’ bad jobs? The evolution of migrant low-wage employment in Germany (1985–2015), Work, Employment and Society, 35(3): 52744. doi: 10.1177/0950017020946567

    • Search Google Scholar
    • Export Citation
  • Krzywdzinski, M. and Gerber, C. (2021) Between automation and gamification: forms of labour control on crowdwork platforms, Work in the Global Economy, 1(1–2): 16184. doi: 10.1332/273241721X16295434739161

    • Search Google Scholar
    • Export Citation
  • Littler, C.R. and Salaman, G. (1982) Bravermania and beyond: recent theories of the labour process, Sociology, 16(2): 25169. doi: 10.1177/0038038582016002006

    • Search Google Scholar
    • Export Citation
  • Mackenzie, A. (2019) From API to AI: platforms and their opacities, Information, Communication & Society, 22(13): 19892006. doi: 10.1080/1369118X.2018.1476569

    • Search Google Scholar
    • Export Citation
  • MacKenzie, R. and Forde, C. (2009) The rhetoric of thegood worker’versus the realities of employers’ use and the experiences of migrant workers, Work, Employment and Society, 23(1): 14259. doi: 10.1177/0950017008099783

    • Search Google Scholar
    • Export Citation
  • McGovern, P. (2007) Immigration, labour markets and employment relations: problems and prospects, British Journal of Industrial Relations, 45(2): 21735. doi: 10.1111/j.1467-8543.2007.00612.x

    • Search Google Scholar
    • Export Citation
  • Miozzo, M. and Ramirez, M. (2003) Services innovation and the transformation of work: the case of uk telecommunications, New Technology, Work and Employment, 18(1): 6279. doi: 10.1111/1468-005X.00111

    • Search Google Scholar
    • Export Citation
  • Mirchandani, K. and Hande, M.J. (2020) The hidden work of challenging precarity, Canadian Journal of Sociology, 45(3): 26588.

  • Noble, D.F. (1984) Forces of Production: A Social History of Industrial Automation, New York: Knopf.

  • Postigo, H. (2016) The socio-technical architecture of digital labor: converting play into youtube money, New Media & Society, 18(2): 33249. doi: 10.1177/1461444814541527

    • Search Google Scholar
    • Export Citation
  • Remesh, B.P. (2008) Work organisation, control and ‘empowerment’: managing the contradictions of call centre work, in Upadhya, C. and Vasavi, A.R. (eds) An Outpost of the Global Economy: Work and Workers in India’s Information Technology Industry, London: Routledge India, pp 23562.

    • Search Google Scholar
    • Export Citation
  • Reuter, M., Dachwitz, I., Fanta, A. and Beckedahl, M. (2019) Exklusiver einblick: so funktionieren facebooks moderationszentren, netzpolitik.org (blog), (Accessed: 5 Apr 2019), https://netzpolitik.org/2019/exklusiver-einblick-so-funktionieren-facebooks-moderationszentren/.

    • Search Google Scholar
    • Export Citation
  • Ritzer, G., Dean, P. and Jurgenson, N. (2012) The coming of age of the prosumer, American Behavioral Scientist, 56(4): 37998. doi: 10.1177/0002764211429368

    • Search Google Scholar
    • Export Citation
  • Roberts, S.T. (2019) Behind the Screen: Content Moderation in the Shadows of Social Media, New Haven, CT and London: Yale University Press.

    • Search Google Scholar
    • Export Citation
  • Sapir, A. (1987) International trade in services: comments, in Giarini, O. (ed) The Emerging Service Economy, Oxford: Pergamon Press.

  • Silver, E. (2018) Hard questions: who reviews objectionable content on Facebook – and is the company doing enough to support them?, Meta (blog), (Accessed: 26 Jul 2018), https://about.fb.com/news/2018/07/hard-questions-content-reviewers/.

    • Search Google Scholar
    • Export Citation
  • Sinders, C. and Ahmad, S. (2021) The labor behind the tools: Using design thinking methods to examine content moderation software, ACM Interactions, https://interactions.acm.org/archive/view/july-august-2021/the-labor-behind-the-tools.

    • Search Google Scholar
    • Export Citation
  • Smith, C. and Thompson, P. (1998) Re-evaluating the labour process debate, Economic and Industrial Democracy, 19(4): 55177. doi: 10.1177/0143831X98194002

    • Search Google Scholar
    • Export Citation
  • Smith, C. (2006) The double indeterminacy of labour power: labour effort and labour mobility, Work, Employment and Society, 20(2): 389402. doi: 10.1177/0950017006065109

    • Search Google Scholar
    • Export Citation
  • Taylor, P. (2015) Labour and the changing landscapes of the call centre, in Putting Labour in Its Place, edited by Kirsty Newsome, Phil Taylor, Jennifer Bair, and Al Rainnie, Basingstoke: Palgrave Macmillan, pp 26686. https://strathprints.strath.ac.uk/51919/.

    • Search Google Scholar
    • Export Citation
  • Thompson, P. (2003) Fantasy island: a labour process critique of the ‘age of surveillance’, Surveillance & Society, 1(2): 13851. doi: 10.24908/ss.v1i2.3350

    • Search Google Scholar
    • Export Citation
  • Thompson, P. (2010) The capitalist labour process: concepts and connections, Capital & Class, 34(1): 714. doi: 10.1177/0309816809353475

    • Search Google Scholar
    • Export Citation
  • Thompson, P. and Smith, C. (2009) Labour power and labour process: contesting the marginality of the sociology of work, Sociology, 43(5): 91330. doi: 10.1177/0038038509340728

    • Search Google Scholar
    • Export Citation
  • Wang, J. (2018) Regulating Hosting ISPs’ Responsibilities for Copyright Infringement, Singapore: Springer.

  • Wright, E.O. (2000) Working-class power, capitalist-class interests, and class compromise, American Journal of Sociology, 105(4): 9571002. doi: 10.1086/210397

    • Search Google Scholar
    • Export Citation
  • Zahra, L. (2020) A new resource to fight unfair content takedowns, Mozilla Foundation (blog), (Accessed: 13 May 2020), https://foundation.mozilla.org/en/blog/new-resource-fight-unfair-content-takedowns/.

    • Search Google Scholar
    • Export Citation
  • Ziegler, A. (2021) Das Tech-unternehmen. Zum fundament eines neuen Unternehmenstypus, ISF München. doi: 10.36194/ArbSozForschung-2021-001

    • Search Google Scholar
    • Export Citation
  • Ahmad, S. (nd) Work and Workers in Content Moderation Value Chains [Unpublished doctoral dissertation], Berlin: Free University Berlin.

  • Ahmad, S. (2019) ‘It’s Just the Job’: investigating the influence of culture in india’s commercial content moderation industry, preprint, SocAXiv. doi: 10.31235/osf.io/hjcv2

    • Search Google Scholar
    • Export Citation
  • Ahmad, S. and Krzywdzinski, M. (2022) Moderating in obscurity: how Indian content moderators work in global content moderation value chains, in M. Graham and F. Ferrari (eds) Digital Work in the Planetary Marke, Cambridge, MA: The MIT Press, International Development Research Centre, pp 7795.

    • Search Google Scholar
    • Export Citation
  • Anderson, B. (2010) Migration, immigration controls and the fashioning of precarious workers, Work, Employment and Society, 24(2): 30017. doi: 10.1177/0950017010362141

    • Search Google Scholar
    • Export Citation
  • Appelman, N. and Leerssen, P. (2022) On ‘trusted’ flaggers, December. https://law.yale.edu/sites/default/files/area/center/isp/documents/trustedflaggers_ispessayseries_jul2022.pdf.

    • Search Google Scholar
    • Export Citation
  • Ascher, D.L. and Noble, S.U. (2019) Unmasking hate on Twitter: disrupting anonymity by tracking trolls, in Free Speech in the Digital Age, New York: Oxford University Press. doi: 10.1093/oso/9780190883591.003.0011

    • Search Google Scholar
    • Export Citation
  • Barrett, P.M. (2020) “Tech - Content Moderation June 2020.” NYU stern center for business and human rights. June 2020, https://bhr.stern.nyu.edu/tech-content-moderation-june-2020.

    • Search Google Scholar
    • Export Citation
  • Batt, R. and Moynihan, L. (2002) The viability of alternative call centre production models, Human Resource Management Journal, 12(4): 1434. doi: 10.1111/j.1748-8583.2002.tb00075.x

    • Search Google Scholar
    • Export Citation
  • Bauder, H. (2005) Institutional capital and labour devaluation: the non-recognition of foreign credentials in Germany, European Journal of Economics and Economic Policies: Intervention, 2(1): 7593. doi: 10.4337/ejeep.2005.01.09

    • Search Google Scholar
    • Export Citation
  • Beverungen, A., Böhm, S. and Land, C. (2015) Free labour, social media, management: challenging marxist organization studies, Organization Studies, London: SAGE Publications.

    • Search Google Scholar
    • Export Citation
  • Birke, P. (2022) Grenzen aus Glas: Arbeit, Rassismus und Kämpfe der Migration in Deutschland, Wien: Mandelbaum Verlag eG.

  • Braverman, H. (1974) Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century, New York and London: Monthly Review Press.

    • Search Google Scholar
    • Export Citation
  • Callaghan, G. and Thompson, P. (2001) Edwards revisited: technical control and call centres, Economic and Industrial Democracy, 22(1): 1337. doi: 10.1177/0143831X01221002

    • Search Google Scholar
    • Export Citation
  • Carter, B., Danford, A., Howcroft, D., Richardson, H., Smith, A. and Taylor, P. (2011) ‘All they lack is a chain’: lean and the new performance management in the British civil service. doi: 10.1111/j.1468-005X.2011.00261.x

    • Search Google Scholar
    • Export Citation
  • Eurostat (2020) Durchschnittliche Wochenarbeitszeit Von Vollzeitbeschäftigten Erwerbstätigen in Deutschland von 2009 bis 2019 nach Geschlecht [Graph], Statista.

    • Search Google Scholar
    • Export Citation
  • Chen, A. (2012) Inside Facebook’s outsourced anti-porn and gore brigade, where Camel toes are more offensive than crushed heads, Gawker, https://www.gawker.com/5885714/inside-facebooks-outsourced-anti-porn-and-gore-brigade-where-camel-toes-are-more-offensive-than-crushed-heads.

    • Search Google Scholar
    • Export Citation
  • Colucci, E. (2007) ‘Focus groups can be fun’: the use of activity-oriented questions in focus group discussions, Qualitative Health Research, 17(10): 142233. doi: 10.1177/1049732307308129

    • Search Google Scholar
    • Export Citation
  • Cova, B. and Dalli, D. (2009) Working consumers: the next step in marketing theory?. doi: 10.1177/1470593109338144

  • Crawford, K. and Gillespie, T. (2016) What is a flag for? Social media reporting tools and the vocabulary of complaint, New Media & Society, 18(3): 41028. doi: 10.1177/1461444814543163

    • Search Google Scholar
    • Export Citation
  • Dijck, J.van. (2013) Facebook and the engineering of connectivity: a multi-layered approach to social media platforms, Convergence, 19(2): 14155. doi: 10.1177/1354856512457548

    • Search Google Scholar
    • Export Citation
  • Dunkel, W. and Kleemann, F. (2013) Customers at Work: New Perspectives on Interactive Service Work, London: Springer.

  • Ebrahim, S. (2017) There is no freedom of speech on Facebook, activists say, The Daily Vox (blog), (Accessed: 6 Dec 2017), https://www.thedailyvox.co.za/there-is-no-freedom-of-speech-on-facebook-activists-say-shaazia-ebrahim/.

    • Search Google Scholar
    • Export Citation
  • Edwards, R. (1979) Contested Terrain: The Transformation of the Workplace in the Twentieth Century, New York: Basic Books.

  • Ellis, V. and Taylor, P. (2006) ‘You don’t know what you’ve got till it’s gone’: re-contextualising the origins, development and impact of the call centre, New Technology, Work and Employment, 21(2): 10722. doi: 10.1111/j.1468-005X.2006.00167.x

    • Search Google Scholar
    • Export Citation
  • Etzold, B. (2017) Capitalising on asylum – the reconfiguration of refugees’ access to local fields of labour in Germany, Refugee Review, 3: 82102.

    • Search Google Scholar
    • Export Citation
  • Fairwork. (2022) Fairwork Germany Ratings 2021: Labour Standards in the Platform Economy, Berlin, Germany and Oxford, UK.

  • Flecker, J., Haidinger, B. and Schönauer, A. (2013) Divide and serve: the labour process in service value chains and networks, Competition & Change, 17(1): 623. doi: 10.1179/1024529412Z.00000000022

    • Search Google Scholar
    • Export Citation
  • Flecker, J. and Schönauer, A. (2016) The production of ‘placelessness’: digital service work in global value chains, in J. Flecker (ed) Space, Place and Global Digital Work, London: Palgrave Macmillan, pp 1130.

    • Search Google Scholar
    • Export Citation
  • Frenkel, S., Korczynski, M., Donoghue, L. and Shire, K. (1995) Re-constituting work: trends towards knowledge work and info-normative control, Work, Employment and Society, 9(4): 77396. doi: 10.1177/095001709594008

    • Search Google Scholar
    • Export Citation
  • Fuchs, C. (2014) Digital Labour and Karl Marx, New York: Routledge.

  • Fuller, L. and Smith, V. (1991) Consumers’ reports: management by customers in a changing economy, Work, Employment and Society, 5(1): 116. doi: 10.1177/0950017091005001002

    • Search Google Scholar
    • Export Citation
  • Gabriel, Y., Korczynski, M. and Rieder, K. (2015) Organizations and their consumers: bridging work and consumption, Organization, 22(5): 62943. doi: 10.1177/1350508415586040

    • Search Google Scholar
    • Export Citation
  • Gabriel, Y. and Lang, T. (2015) The Unmanageable Consumer, London: SAGE.

  • Gandini, A. (2019) Labour process theory and the gig economy, Human Relations, 72(6): 103956. doi: 10.1177/0018726718790002

  • Gillespie, T. (2018) Custodians of the Internet: Platforms, Content Moderation, and the Hidden Decisions That Shape Social Media, New Haven, CT and London: Yale University Press.

    • Search Google Scholar
    • Export Citation
  • Gillespie, T. (2020) Content moderation, ai, and the question of scale, Big Data & Society, 7(2): 2053951720943234. doi: 10.1177/2053951720943234

    • Search Google Scholar
    • Export Citation
  • Glucksmann, M.A. (2004) Call configurations: varieties of call centre and divisions of labour, Work, Employment and Society, 18(4): 795811. doi: 10.1177/0950017004047965

    • Search Google Scholar
    • Export Citation
  • Grimmelmann, J. (2015) The virtues of moderation, Yale Journal of Law and Technology, 17: 42.

  • Habermas, J. (1962) Strukturwandel Der öffentlichkeit, Neuwied, Berlin, Vol 6, pp 4275.

  • Habermas, J. (2022) Ein Neuer Strukturwandel der Öffentlichkeit und die Deliberative Politik, Berlin: Suhrkamp Wissenschaft, https://www.suhrkamp.de/buch/juergen-habermas-ein-neuer-strukturwandel-der-oeffentlichkeit-und-die-deliberative-politik-t-9783518587904.

    • Search Google Scholar
    • Export Citation
  • Hochschild, A.R. (1983) The Managed Heart: Commercialization of Human Feeling, Berkeley, CA: University of California Press.

  • Holtgrewe, U. and Doellgast, V. (2012) A service union’s innovation dilemma: limitations on creative action in German industrial relations, Work, Employment and Society, 26(2): 31430. doi: 10.1177/0950017011432913

    • Search Google Scholar
    • Export Citation
  • Houlihan, M. (2004) Tensions and variations in call centre management strategies, Call Centres and Human Resource Management, London: Palgrave Macmillan, pp 75101.

    • Search Google Scholar
    • Export Citation
  • Howcroft, D. and Richardson, H. (2012) The back office goes global: exploring connections and contradictions in shared service centres, Work, Employment and Society, 26(1): 11127. doi: 10.1177/0950017011426309

    • Search Google Scholar
    • Export Citation
  • Ith, T. (2015) Microsoft’s PhotoDNA: protecting children and businesses in the cloud. Stories, https://news.microsoft.com/features/microsofts-photodna-protecting-children-and-businesses-in-the-cloud/.

    • Search Google Scholar
    • Export Citation
  • Jourová, V. (2019) How the code of conduct helped countering illegal hate speech online, Directorate-General for Justice and Consumers. European Commission, https://ec.europa.eu/info/sites/default/files/hatespeech_infographic3_web.pdf.

    • Search Google Scholar
    • Export Citation
  • Kellogg, K.C., Valentine, M.A. and Christin, A. (2019) Algorithms at work: the new contested terrain of control, Academy of Management Annals, 14(1): 366410. doi: 10.5465/annals.2018.0174

    • Search Google Scholar
    • Export Citation
  • Klonick, K. (2017) The new governors: the people, rules, and processes governing online speech, Harvard Law Review, 131: 73.

  • Korczynski, M. (2002) Human Resource Management in the Service Sector, Basingstoke: Palgrave.

  • Krings, T. (2021) ‘Good’ bad jobs? The evolution of migrant low-wage employment in Germany (1985–2015), Work, Employment and Society, 35(3): 52744. doi: 10.1177/0950017020946567

    • Search Google Scholar
    • Export Citation
  • Krzywdzinski, M. and Gerber, C. (2021) Between automation and gamification: forms of labour control on crowdwork platforms, Work in the Global Economy, 1(1–2): 16184. doi: 10.1332/273241721X16295434739161

    • Search Google Scholar
    • Export Citation
  • Littler, C.R. and Salaman, G. (1982) Bravermania and beyond: recent theories of the labour process, Sociology, 16(2): 25169. doi: 10.1177/0038038582016002006

    • Search Google Scholar
    • Export Citation
  • Mackenzie, A. (2019) From API to AI: platforms and their opacities, Information, Communication & Society, 22(13): 19892006. doi: 10.1080/1369118X.2018.1476569

    • Search Google Scholar
    • Export Citation
  • MacKenzie, R. and Forde, C. (2009) The rhetoric of thegood worker’versus the realities of employers’ use and the experiences of migrant workers, Work, Employment and Society, 23(1): 14259. doi: 10.1177/0950017008099783

    • Search Google Scholar
    • Export Citation
  • McGovern, P. (2007) Immigration, labour markets and employment relations: problems and prospects, British Journal of Industrial Relations, 45(2): 21735. doi: 10.1111/j.1467-8543.2007.00612.x

    • Search Google Scholar
    • Export Citation
  • Miozzo, M. and Ramirez, M. (2003) Services innovation and the transformation of work: the case of uk telecommunications, New Technology, Work and Employment, 18(1): 6279. doi: 10.1111/1468-005X.00111

    • Search Google Scholar
    • Export Citation
  • Mirchandani, K. and Hande, M.J. (2020) The hidden work of challenging precarity, Canadian Journal of Sociology, 45(3): 26588.

  • Noble, D.F. (1984) Forces of Production: A Social History of Industrial Automation, New York: Knopf.

  • Postigo, H. (2016) The socio-technical architecture of digital labor: converting play into youtube money, New Media & Society, 18(2): 33249. doi: 10.1177/1461444814541527

    • Search Google Scholar
    • Export Citation
  • Remesh, B.P. (2008) Work organisation, control and ‘empowerment’: managing the contradictions of call centre work, in Upadhya, C. and Vasavi, A.R. (eds) An Outpost of the Global Economy: Work and Workers in India’s Information Technology Industry, London: Routledge India, pp 23562.

    • Search Google Scholar
    • Export Citation
  • Reuter, M., Dachwitz, I., Fanta, A. and Beckedahl, M. (2019) Exklusiver einblick: so funktionieren facebooks moderationszentren, netzpolitik.org (blog), (Accessed: 5 Apr 2019), https://netzpolitik.org/2019/exklusiver-einblick-so-funktionieren-facebooks-moderationszentren/.

    • Search Google Scholar
    • Export Citation
  • Ritzer, G., Dean, P. and Jurgenson, N. (2012) The coming of age of the prosumer, American Behavioral Scientist, 56(4): 37998. doi: 10.1177/0002764211429368

    • Search Google Scholar
    • Export Citation
  • Roberts, S.T. (2019) Behind the Screen: Content Moderation in the Shadows of Social Media, New Haven, CT and London: Yale University Press.

    • Search Google Scholar
    • Export Citation
  • Sapir, A. (1987) International trade in services: comments, in Giarini, O. (ed) The Emerging Service Economy, Oxford: Pergamon Press.

  • Silver, E. (2018) Hard questions: who reviews objectionable content on Facebook – and is the company doing enough to support them?, Meta (blog), (Accessed: 26 Jul 2018), https://about.fb.com/news/2018/07/hard-questions-content-reviewers/.

    • Search Google Scholar
    • Export Citation
  • Sinders, C. and Ahmad, S. (2021) The labor behind the tools: Using design thinking methods to examine content moderation software, ACM Interactions, https://interactions.acm.org/archive/view/july-august-2021/the-labor-behind-the-tools.

    • Search Google Scholar
    • Export Citation
  • Smith, C. and Thompson, P. (1998) Re-evaluating the labour process debate, Economic and Industrial Democracy, 19(4): 55177. doi: 10.1177/0143831X98194002

    • Search Google Scholar
    • Export Citation
  • Smith, C. (2006) The double indeterminacy of labour power: labour effort and labour mobility, Work, Employment and Society, 20(2): 389402. doi: 10.1177/0950017006065109

    • Search Google Scholar
    • Export Citation
  • Taylor, P. (2015) Labour and the changing landscapes of the call centre, in Putting Labour in Its Place, edited by Kirsty Newsome, Phil Taylor, Jennifer Bair, and Al Rainnie, Basingstoke: Palgrave Macmillan, pp 26686. https://strathprints.strath.ac.uk/51919/.

    • Search Google Scholar
    • Export Citation
  • Thompson, P. (2003) Fantasy island: a labour process critique of the ‘age of surveillance’, Surveillance & Society, 1(2): 13851. doi: 10.24908/ss.v1i2.3350

    • Search Google Scholar
    • Export Citation
  • Thompson, P. (2010) The capitalist labour process: concepts and connections, Capital & Class, 34(1): 714. doi: 10.1177/0309816809353475

    • Search Google Scholar
    • Export Citation
  • Thompson, P. and Smith, C. (2009) Labour power and labour process: contesting the marginality of the sociology of work, Sociology, 43(5): 91330. doi: 10.1177/0038038509340728

    • Search Google Scholar
    • Export Citation
  • Wang, J. (2018) Regulating Hosting ISPs’ Responsibilities for Copyright Infringement, Singapore: Springer.

  • Wright, E.O. (2000) Working-class power, capitalist-class interests, and class compromise, American Journal of Sociology, 105(4): 9571002. doi: 10.1086/210397

    • Search Google Scholar
    • Export Citation
  • Zahra, L. (2020) A new resource to fight unfair content takedowns, Mozilla Foundation (blog), (Accessed: 13 May 2020), https://foundation.mozilla.org/en/blog/new-resource-fight-unfair-content-takedowns/.

    • Search Google Scholar
    • Export Citation
  • Ziegler, A. (2021) Das Tech-unternehmen. Zum fundament eines neuen Unternehmenstypus, ISF München. doi: 10.36194/ArbSozForschung-2021-001

    • Search Google Scholar
    • Export Citation
Sana Ahmad WZB Berlin Social Science Center and Free University Berlin, Germany

Search for other papers by Sana Ahmad in
Current site
Google Scholar
Close
and
Maximilian Greb WZB Berlin Social Science Center, Germany

Search for other papers by Maximilian Greb in
Current site
Google Scholar
Close

Content Metrics

May 2022 onwards Past Year Past 30 Days
Abstract Views 639 5 0
Full Text Views 1634 1033 57
PDF Downloads 908 503 42

Altmetrics

Dimensions