‘A Part of Our Work Disappeared’: AI Automated Publishing in Social Media Journalism

Abstract

This study explores the implementation of artificial intelligence (AI) in social media journalism. We apply a labour process approach to examine why German newspaper editors adopt AI publishing and how it influences journalistic work. Automated publishing services (APSs) are used in newsrooms to select, edit, and publish content on social media platforms. In-depth interviews with German news editors revealed that the reasons for implementing APSs include economic dependence on platforms, the centralisation of news roles, and the intensification of work. Furthermore, resistance to fully automated social media publishing in some newsrooms has resulted in semi- or hybrid-automated approaches. Resistance is primarily expressed through concerns over the loss of editorial control, content diversity, and the quality of user engagement.

1. Introduction

Economic, technological, and cultural changes have increasingly prompted news organisations to integrate artificial intelligence (AI) into their news production processes, fundamentally transforming journalists’ work (Simon, 2024). News publishers face economic challenges from declining revenue and substantial income shifts to major tech companies such as Meta, Alphabet, and ByteDance (see Bleyer-Simon et al., 2024) These companies utilise personalisation algorithms and AI models to reach large audiences. At the same time, news publishers rely on these companies’ market and data dominance because they function as gatekeepers. From a commercial perspective, a major challenge for news publishers is engaging audiences in a highly competitive and fragmented digital news market (Fletcher & Nielsen, 2017). From a public service perspective, the challenge lies in reaching critical groups, such as underserved audiences (Benson, 2019).

Meta’s Facebook and Instagram are crucial platforms for news consumption, although Facebook’s market share and cultural relevance have declined (Newman et al., 2024). Incidental news consumption on these platforms is critical for reaching audiences who otherwise avoid news. However, Meta’s corporate policies and ongoing conflicts over platform regulations have made Facebook traffic to news sites unreliable. The opaque algorithms and AI models behind the ranking system further complicate the process of maintaining or increasing audience reach. Editors require knowledge about how algorithms influence content promotion and demotion across diverse audiences and times. Consequently, journalists in digital and social media-focused roles have developed routines, best practices, and strategies to grow audiences and engagement while negotiating professional journalistic standards (Neilson & Gibson, 2022).

Amid tensions between social media logic and journalistic standards, many newsrooms have integrated AI-supported automated publishing services (APSs) into their news production and distribution, significantly affecting journalistic work. APSs are among several newsroom software tools, including content management systems (CMSs) and audience metrics dashboards. News publishers use APSs to publish content across various social media platforms, increase workflow efficiency, and relieve ‘scarce editorial (human) resources’ (Nielsen & Cherubini, 2022). Newman (2022) identifies AI-driven APSs, such as Echobox, as an emerging trend in journalism, a finding also confirmed by respondents in this study. APSs import publishers’ Really Simple Syndication (RSS) or content feeds, select posts that drive engagement, and suggest or publish them directly on platforms. This news production process can be conducted with varying degrees of automation.

A growing number of studies have addressed the changes, challenges, and effects of automation, algorithms (see Siitonen et al., 2024), and AI (see Calvo-Rubio & Ufarte-Ruiz, 2021; Simon, 2024) in the context of journalistic work. Other studies have used labour process theory (LPT) to examine the impact of technological innovation on the structural conditions of journalistic work (see Cohen, 2015, 2019; Hayes, 2024; Örnebring, 2010). Employing a Marxist approach, LPT examines capitalist relations of production, providing insights into changes in news work and their broader economic context (Hayes & O’Sullivan, 2023). Our study complements existing research by examining the transformation of journalists’ labour processes involving AI-based APSs (Chillas & Baluch, 2019; Smith, 2015). We conducted an exploratory study with ten journalists from eight major media groups in Germany to address our research questions: What motivates news organisations to use APSs? How do they change journalistic work, and what are their effects? What forms of resistance or adaptation strategies do journalists develop?

The reasons for using APSs include economic dependence on social media platforms, centralisation of news roles, and intensification of journalistic work. Our findings reveal that journalists appreciate the efficiency APSs bring to daily tasks and processes, but this occurs within a broader context of corporate mergers and restructuring, where fewer journalists are expected to take on more social media work. As a result, introducing AI services has also triggered conflicts, resistance, and new strategies for journalists. APSs have been integrated into news routines in various ways, ranging from fully automated social media posts to semi-automated or hybrid approaches. These differences are partly shaped by concerns over editorial control, the diversity of news content, and the types of comments that automated posts receive from social media users.

2. Literature

2.1. AI and Automation in Journalism

The popularisation and commercialisation of AI have spurred a wave of studies focusing on the impact of AI, automation, and algorithms on journalistic work (see Calvo-Rubio & Ufarte-Ruiz, 2021; Siitonen et al., 2024). AI is the most recent of a series of technological changes in the news industry—including the digitisation and platformisation of journalism—which have introduced new challenges and anxieties into the sector (Mari, 2024). It is characterised by Broussard et al. (2019) as a field of computer science that deals with the simulation of human intelligence, with a focus on the subarea of machine learning: ‘the training of a machine to learn from data, recognise patterns and make subsequent judgements, with little to no human intervention’. This definition is broad enough to include so-called generative AI and other types of self-adjusting ranking and personalisation algorithms. Machine learning subfields include deep learning and various forms of natural language processing (NLP) (Simon, 2024). Research in journalism studies has investigated the impacts of algorithms, especially those in social media and search ranking systems (see Siitonen et al., 2024). Many of these algorithms, such as those developed by Meta (formerly Facebook) and Google, involve machine learning (Hazelwood et al., 2018). These uses of AI predate the popularisation of generative AI, although there has been increased interest in the field as tech companies race to integrate AI into new and existing services.

AI capabilities are already being used by news organisations, shaping journalists’ day-to-day work. AI services promise improved efficiency and competitive advantages, which correspond with broader cultural narratives about technological progress and rationalisation (Feenberg, 2017). AI services build on, join, and, in some cases, replace existing software systems, such as CMSs, which standardise news production (Neilson, 2021) and audience dashboards that inform editorial and social media publishing decisions (Petre, 2021). News organisations adopt AI for various reasons: for example, to respond to market pressures and financial challenges, cut costs, and the promise of supporting news quality (i.e., freeing up time) (Opdahl et al., 2023; Simon, 2024). However, there are also concerns that AI could generate ‘a flood of low-quality content’, which further reduces interest and trust in news (Newman et al., 2024).

AI is already prevalent in all phases of news production (Beckett & Yaseen, 2023; Simon, 2022), making the interaction between humans and AI a central concern in journalism research (Ferrucci & Perreault, 2021; Primo & Zago, 2015). The phases of AI news production include selection, production, distribution, moderation, and analysis. For example, AI services support the identification of topics and verification of information in the news selection phase. In the production phase, AI-based tools are used to write, edit, and translate content, visualise data, and transcribe audio files. During the dissemination phase, journalists utilise AI-supported recommendation systems for news on their websites and publish content on social media platforms that, in turn, employ AI-based ranking systems. In the moderation phase, they apply AI software such as Conversario (https://www.conversar.io/en; accessed on 26 October 2024) to manage user comments on social media platforms (Opdahl et al., 2023). News organisations have been distributing their news content via AI-driven social media platforms for almost two decades and are now deploying AI services (APSs) to automate publishing on these platforms.

2.2. Social Media Management and APSs

Social media management software (e.g., HootSuite, Buffer, Sprout Social) has been used by news organisations for more than a decade to manage their social media presence, particularly for marketing and brand-building purposes (Thapa & Skinner, 2015). However, news publishers are increasingly employing AI-driven social media management software such as Echobox (see Nielsen & Cherubini, 2022; Newman, 2022; Paik, 2023; Sánchez-García et al., 2023; Verstappen & Opgenhaffen, 2024). At the time of publishing, news organisations, including South China Morning Post, Le Monde, The Guardian, and many more, are listed as Echobox customers, using the AI-driven social media management software to publish content on platforms such as Facebook. This service builds on existing software infrastructure such as CMSs and is explicitly developed for news publishers to automate editorial processes by taking content from news organisations’ RSS or content feeds (see Echobox, 2024).

To distinguish traditional social media management software from AI-based services for news publishers, we refer to Echobox as an APS in this paper. According to the service provider, APSs can predict timing and social virality with an AI-supported ranking system, select and suggest posts, generate news headlines using NLP, and publish automatically on Facebook, Instagram, TikTok, X, and LinkedIn while also reacting to social media algorithm updates. APSs promise to increase website traffic and engagement, offer tools for monitoring and analysing social media, and enable comparisons with competitors (Echobox, 2024).

Echobox appears in a few studies. Newman (2022) and Sánchez-García et al. (2023) refer to it as an emerging tool in their reviews of AI journalism. There is a brief discussion in Paik’s (2023) study on digital journalism ethics, where respondents express interest in APSs for convenience but are also concerned about data bias, reliability, and ethical standards. Nielsen and Cherubini (2022) state that publishers automate processes with this APS to ‘reduce their investment in Facebook’ and avoid engaging scarce human resources while ‘maintaining their presence on the platform’. None of these studies explicitly focused on APSs in newsrooms, although this service has been mentioned as a trend in news organisations’ workflows.

2.3. Social Media Platforms and News

News organisations are struggling with declining revenue, which is shifting to large social media platforms (Keller & Eggert, 2024). The economic challenges faced by news media have led to a concentration of ownership, resulting in declining local offerings, job losses, and the centralisation of news roles. Although Germany has comparatively high news media diversity, structural changes are underway (Bleyer-Simon et al., 2024). Social media platforms are crucial for disseminating news and reaching underserved groups (Benson, 2019).

News organisations depend on social media platforms such as Facebook to increase website traffic, user engagement, brand awareness, and subscriptions (Chen & Pain, 2021). User engagement refers to audience responses (likes, comments, shares, and reactions). Meta’s Facebook and Instagram and ByteDance’s TikTok are among the most important social media platforms for news organisations (Newman et al., 2024) and are core components of the news publishing process. However, adjustments to the ranking systems of social media platforms have significantly impacted content prioritisation, affecting website traffic and engagement (Bailo et al., 2021). Facebook’s significance has decreased in most Western countries because of algorithm and product changes in response to government platform regulations, an ageing demographic and management strategies (Newman et al., 2024; Nielsen & Cherubini, 2022). Newspaper publishers are caught between short-term opportunities offered by platform technologies (to increase engagement) and long-term strategies that ensure their independence from social media platforms (Nielsen & Ganter, 2022). Dependence on big tech companies has been an important research subject for years, emphasising the ongoing importance of platform economics and logic (Meese & Hurcombe, 2020).

2.4. Social Media Journalism and Social Media Logic

The discrepancy between social media logic and professional journalistic standards is a major concern for journalism researchers (Hermida & Mellado, 2020). Hendrickx and Opgenhaffen (2024) characterise social media journalism as newswork that employs one or more social media platforms in the core aspects of production, distribution, and/or consumption of news content. In this definition, ‘production’ includes the use of platform features, while ‘distribution’ refers to the sharing of news articles on social media platforms, including the adaptation of content to fit the logic of social media (Lamot et al., 2022). Social media logic, according to van Dijck and Poell (2013), is the set of strategies, mechanisms, and economies that underlie the dynamics of these platforms and impact user engagement.

Professional journalistic standards often compete or conflict with social media logic. Lischka (2021) finds that social media editors emphasise emotional and surprising elements in their posts to adapt to user preferences and Facebook News feed logic. These emotions, conveyed via news content, help increase engagement with posts (Heidenreich et al., 2022). In her study of European media organisations, Lamot (2022) concludes that focusing on engagement results in news softening. Qualitative studies have identified evidence of differences in the news headlines of websites and news status messages in Facebook posts (see Lamot, 2022; Lischka, 2021). As DeVito (2017) states, ‘Facebook’s algorithmic values are very different in both content and underlying structure from traditional news values’. Stöcker (2020) points out that the recommendation systems used by these companies prioritise content and user behaviour that can be monetised; ‘however, these metrics often do not reflect relevance, quality, or other desirable properties’. Social media editors’ roles in news marketing and their performance objectives (i.e., engagement and subscriptions) are balanced with (and are sometimes in conflict with) traditional news values and the public service role of journalism (Neilson & Gibson, 2022). This ‘balancing act’ takes place in a context where recommendation algorithms such as Facebook feed can influence editorial decisions and access to news (Bernstein et al., 2021), with APSs increasing this influence.

2.5. AI and Journalistic Labour

The automation of news production processes raises questions about journalistic labour, whether it contributes to redundancies, changes newsroom roles, increases management control, enables journalists to fulfil their work, or involves a combination of these effects (Carlson, 2015). While ‘automation anxiety’ is often expressed as a fear that technology will replace humans, the reality is more complex (van Dalen, 2012). Despite declining employment in journalism and the capacity of algorithms to write like human journalists, Linden (2017) argues that professional ideologies of journalism (see Deuze, 2005) and the artisanal nature of work mean that machines are not simply replacing journalists. Instead, journalists are adapting to new forms of human–computer interaction in their work and changing their workflows accordingly. Human journalists have aligned their work more closely with the logic and demands of audience metrics, CMSs, and curation algorithms (Neilson, 2021). This is evident in the colonisation of newsrooms by audience metrics, which are used to make editorial decisions by aligning news products with what the audience wants or works best on third-party platforms (Petre, 2021). Audience metrics have been used as management technologies to monitor and discipline news work (Bunce, 2019; Milosavljević & Vobič, 2021).

While social media editors have been instrumental in bringing audience metrics into the newsroom, training other journalists, and formalising online engagement strategies (Neilson et al., 2023), some of their work is now automated by news recommender systems (NRSs) and APSs. NRSs are ‘algorithmic solutions that filter, suggest, and prioritise content based on previous or similar users’ behaviour, explicitly stated user preferences, popularity metrics, and other content-specific characteristics’ (Mitova et al., 2023). By developing their own NRS (onsite), news companies can regain control over news curation (Møller, 2023) and better align algorithmic curation with journalistic values (Bastian et al., 2021). Studies on the development and implementation of internal NRS have shown how difficult it is to programme journalistic principles into algorithmic systems. This software builds on existing technologies and relies on journalists’ previous work as a source of training data. Similarly to news-writing algorithms, NRSs and APSs are part of a process to formalise and quantify news work and make it available for automation.

As such, automation does not mean less work for human reporters. As Hayes and O’Sullivan (2023) suggest, digitisation is associated with more frequent deadlines and higher work intensity. Management introduces new technologies to enforce control over the pace, quantity and quality of journalists’ work. Online engagement also involves affective labour. That is, the expectation that journalists make themselves more available to social media audiences, work to foster online communities, and deal with harassment and hate speech (Miller & Lewis, 2022; Pantti & Wahl-Jorgensen, 2021). Approaching the idea of emotional labour from another angle, Olsen (2023) argues that automation can negatively impact journalists by degrading their work and making it seem less meaningful.

2.6. Labour Process Theory (LPT)

The labour process theory provides a framework for understanding specific changes in journalists’ work related to the automation of social media publishing and the broader economic context in which these changes occur. As Edwards (2010) argues, LPT combines an ‘empirical interest in the experience of work at the point of production and a theoretical concern with the contradictory relationships between capital and labour’. Furthermore, LPT is well suited for studying technology in the workplace as one of the many competing factors that shape work (Smith, 2015). AI-driven APSs and changing newsroom roles can be considered part of ‘labour-capital struggles over the way in which the technology is implemented and utilised’ (Hall, 2010, cited in Chillas & Baluch, 2019).

Founded in Marx’s political economy, LPT was formalised as a methodology after the publication of Braverman’s (1974) book Labor and Monopoly Capital: The Degradation of Work in the Twentieth Century. The labour process involves an ‘activity in which a natural object or raw material is transformed into a useful product’, satisfying human needs (Smith, 2016). In addition to human labour (journalists’ work), the basic elements of the labour process are the object (news content) being worked on, instruments (software, services, etc.), and purpose (public service, engagement, or profit) (Smith, 2015). Furthermore, LPT investigates the conflict between workers and management and labour and capital (Chillas & Baluch, 2019). Braverman argued that technology is used to deskill labour and disempower and alienate workers. However, Burawoy (1979) and others have challenged Braverman’s degradation thesis, noting that new technologies and management strategies do not always result in lower-skilled work.

While Braverman studied factory labour, his approach has been extended to research on white-collar, creative, and professional work (McKinlay, 2009). Much of this research has focused on management control and worker resistance. For example, Thompson (2024) differentiates between consent and coercion in workplaces. The concept of consent helps to understand the ideological and subjective aspects of work, wherein white-collar managers invoke professional values and career goals to increase productivity (Smith, 2016). At the same time, software systems can impose more granular control and surveillance on creative work. Likewise, white-collar and creative workers can challenge management control by invoking professional norms and values such as autonomy and public service in the context of journalism (Neilson, 2021), or they can adopt individual or collective forms of resistance such as industrial action (Cohen & de Peuter, 2020). Control in newsrooms is also complicated by the contradictory class position of cultural workers (Cohen, 2012) and because editors occupy a complex structural position as both managers and reporters (Hayes & O’Sullivan, 2023).

Journalism researchers have adopted LPT to investigate management control in digital newsrooms (Hayes & O’Sullivan, 2023), the role of audience metrics in news work (Petre, 2021), and the working conditions of freelance journalists (Cohen, 2015, 2019). Örnebring’s (2010) influential study of journalism labour and technology tested LPT approaches, concluding that adopting new technologies in newsrooms leads to complex combinations of specialisation, multiskilling, upskilling, and deskilling. Here, we extend LPT to understand the implementation of AI-driven APSs in newsrooms. The following research questions are addressed using tLPT, and our findings are structured based on its key elements:

RQ1:

What motivates journalists and managers in German news organisations to introduce APSs into their social media news production process?

RQ2:

How do AI technologies such as APSs change journalists’ social media work processes, and what is their impact?

RQ3:

What conflicts, forms of resistance, and adaptation strategies do journalists develop in response to APSs?

3. Method

We conducted a qualitative, exploratory study to address these research questions by asking journalists about their experiences, practises, and opinions. The interviewees were initially recruited via a leading German media network, and two additional respondents through network recommendations. The journalists worked for 10 news organisations within 8 major media groups in Germany, overseeing 23 regional Facebook pages with over 1.3 million followers and more than 200 local and other Facebook pages across 11 federal states. They were responsible for social media content production and, in many cases, the strategic development of editorial offerings. The respondents’ positions varied from editor-in-chief, online chief, and online/social media editor to social media or content specialist, depending on the organisation’s structure and size. Following Anter’s (2024) characterisation of social media editors, all interviewees had an active and far-reaching influence on social media production and were not just social media administrators. Half of them held management positions and provided their perspectives on labour and technology. Interviews were conducted and recorded using Zoom from November 2021 to February 2022, each lasting 45–55 min. Transcriptions were anonymised according to established rules (Dresing & Pehl, 2018) and numbered I1 to I10 (see

Table 1. Interviewee Data: Respondents and their Facebook accounts and Echobox usage (February 2022).

German Interviewees	Regional Facebook Accounts (Followers)	APS
I01 *	4 (200,000+)	interest
I02	3 (270,000+)	interest
I03 *	3 (207,000+)	utilised
I04	1 (80,000+)	interest
I05 *	2 (70,000+)	utilised
I06 *	3 (78,000+)	utilised
I07 *	2 (45,000+)	tested
I08	1 (135,000+)	utilised
I09 *	1 (120,000+)	tested
I10	3 (104,000+)	utilised
* management position	23 1 (1.3 m+)

¹ Additionally, 208 local and other Facebook accounts.

).

The interviews were then coded using qualitative content analysis with deductive and inductive coding to identify topics, interpret data, and answer the research questions (Mayring & Fenzl, 2014). The analysis and coding were performed using MAXQDA 2022 software and checked with a second coding cycle, ‘intracoder reliability’ (Rädiker & Kuckartz, 2020). We developed a semi-structured interview guide based on the research questions, the literature review, and a preliminary interview with a German journalist. To cluster the topics and interpret the results, we used the following LPT dimensions: (1) point of production: company/department structure, news role, responsibilities, collaboration, social media goals, and strategies; (2) instruments and news production processes: software and services in social media news production; (3) object/product: impact on social media news content; and (4) conflicts and resistance to the implementation and utilisation of AI technology (Chillas & Baluch, 2019; Smith, 2015).

4. Findings

4.1. Changes at the Point of Production: Newsrooms and the News Industry

The implementation of AI in the production process is a response to the centralisation of news functions, reorganisation of workplace structures, intensification of work, and continuing platform dependency (see Bleyer-Simon et al., 2024; Newman et al., 2024; Simon, 2024). Interviewees stated that the organisations they worked for were either part of a media group, undergoing a consolidation process, or collaborating with a larger media organisation. One of the journalists hinted at an upcoming merger with a larger media organisation but was not sure yet if and when the editorial team would be affected (I04). These consolidations and structural changes in newsrooms not only centralise media power but also lead to insecure working conditions, job losses, and work intensification (see Carlson, 2015). The centralisation of news functions means that the work of many journalists is conducted by fewer employees, who often rely on automation and other labour-saving tools.

The interviewees described centralised approaches to managing social media accounts, with only a few still being edited by local journalists. Collectively, the organisations had 23 primary or regional Facebook pages, more than 200 local (or other) Facebook pages, and many more accounts on other platforms such as Instagram, YouTube, TikTok, and LinkedIn. The journalists indicated that managing a growing number of accounts is challenging; hence, the need for APSs. One of the interviewees reported the following: ‘We have the support of Echobox…, which has become important… when you have so many pages as a single online editor… even three people can’t manually manage 32 Facebook pages and handle engagement simultaneously’ (I10). The journalists also mentioned the increasing workload of moderating toxic comments on Facebook; therefore, they implemented additional AI software to automate this process (Conversario). As one journalist explained, ‘It really is such a big task that we are unable to moderate all of our channels’ (I03). Existing staff cannot keep pace with the intensification of social media work without the help of automation (see Nygren, 2014).

Journalists who use APSs explained that they have wide-ranging responsibilities employing multiple skills. As one journalist who uses an APS stated, ‘I also do other things, like managing the website for the online team or reporting, where you do research, so in addition to everything else. I am not just a social media manager; that is not our only workload’ (I03). APSs promise to manage media group accounts across multiple platforms and relieve scarce human resources of repetitive and time-consuming tasks. Consequently, journalists can balance other journalistic tasks with social media administration (see Nygren, 2014; Anter, 2024).

Some of the journalists interviewed said that APSs saved time and optimised their workflow. ‘It really has facilitated workflow and freed up capacity for other things… we used to spend a lot of time posting and writing teasers and planning… we don’t have to do that anymore’, said one (I10), while another said, ‘A part of our work disappeared… so we can focus on the stories’ (I08). The journalists feel freed from repetitive, low-skilled tasks. The flip side of this efficiency is that the software has allowed news organisations to centralise and downsize digital teams, leaving fewer social media editors to manage more accounts.

In some media groups, specialised strategy and innovation departments strive to introduce efficient workflows and performance targets. A social media specialist from a major media group responsible for the strategic development of new services described the implementation of a social media roadmap and intended introduction of new CMS and APS software because of a complex company merger. The roadmap includes optimised and automated workflows and standardised performance metrics, such as user engagement. He also noted that his job was to introduce key performance indicators (KPIs) (I02). In the absence of formal KPIs, some journalists used metrics for comparison or had expectations regarding daily achievement, engagement, and subscriptions (I10, I07). This is evidence of news organisations’ attempts to increase their efficiency and profitability by optimising work processes and introducing KPIs (see Hayes & O’Sullivan, 2023).

Many news organisations depend on social media platforms to achieve engagement, brand awareness, visibility, referral traffic, and digital subscriptions (see Chen & Pain, 2021; Meese & Hurcombe, 2020). According to the respondents, dependency is most noticeable when Meta changes its guidelines or publishes algorithm updates. However, according to an interviewee, the APS was able to react and adapt quickly to the algorithm (I05). The interviewees stated that they believed Facebook, Instagram, Twitter, and YouTube were the most important platforms (see Newman et al., 2024). Respondents estimated that approximately 80% of their news organisations’ social media traffic originates from Facebook. One of the publishers even received 50–60% of all its website traffic from Facebook (I10). According to the interviewees, Facebook’s significance is gradually declining, yet they do not want to leave Facebook (see Nielsen & Cherubini, 2022). News publishers are still indecisive about short-term opportunities to increase engagement and traffic and long-term strategies for platform independence (see Nielsen & Ganter, 2022). Currently, the economic dependence on social media platforms is a significant factor in the introduction of APSs.

4.2. Instruments of Social Media Production

The interviewees said three categories of software were integrated into labour processes related to social media news production: CMSs, audience analytics dashboards, and AI-driven publishing and moderation services. These technologies are implemented partly in response to organisational consolidation and are used to control or monitor work processes (see Chillas & Baluch, 2019; Smith, 2015). Journalists reported plans to introduce new CMSs (e.g., CUE) for print and online in their groups to optimise the entire news production process and provide journalists with a single user interface (I01, I02, I09, and I10) and/or advance online-first strategies (I10). Therefore, traditional print, online, and social media news roles use only one system to publish content.

The interviewees used a wide range of social media analytics software, including CrowdTangle, Facebook Insights, Google Analytics, and custom dashboards. One of the interviewees said that their results were being reported to management (I10). AI-powered APSs now complement or replace analytics software, as one journalist explained, ‘We usually also look at the analytics section in Echobox itself’ (I03). These tools can support journalists by providing comprehensive information about the audience but also make it easier for managers to monitor and discipline news work (see Bunce, 2019; Milosavljević & Vobič, 2021).

To manage extensive social media activity and platform engagement (as well as the changes caused by Facebook’s AI ranking updates), news organisations have introduced social media management software. Some use HootSuite or Meta-Business Suite to manage their Facebook and Instagram accounts, but most use or have used the APS Echobox (see Table 1), which centralises and automates social media administration. To moderate user comments (including toxic comments and hate speech) on Facebook, journalists applied the AI-based software Conversario (see Opdahl et al., 2023). One of the journalists described its functionality: ‘We now have a new tool, called Conversario, that is supposed to help us moderate critical comments…the AI scans for certain topics or words and then hides these comments’ (I03). These AI tools help journalists cope with increasing workloads.

Contrary to claims that workflow management and automation systems improve productivity, a journalist who uses Echobox and Conversario emphasised that a proliferation of technologies could be difficult to manage. ‘The more tools you have, the more complicated it becomes. We have two different CMS systems for publishing; therefore, you always have to juggle back and forth. Anyway, I always work on three monitors, and my browser always has at least ten tabs open’ (I10). Journalists use various technologies during the social media news production process. For some, this makes the work more complex, requiring multiskilling across the various instruments, adaptation to new technologies, and retraining of journalists (see Nygren, 2014).

We asked journalists how these technologies shaped their work routines and identified three social media news production processes: automated, semi-automated, and manual (see Figure 1). Considering that Facebook was the most important social media platform for news organisations at the time, we limited our analysis to the Facebook publishing process.

Journalists primarily used Meta Business Suite (MBS) or the APS Echobox to modify and publish news on Facebook. In MBS, journalists create news stories manually or import them from the news organisation’s CMS using APIs. News content is curated and edited via MBS and published on Instagram and Facebook using a relatively manual approach. Accordingly, many interviewees said they used their professional knowledge and results from social media analytics software or feedback from an audience or conversion manager service to maximise engagement.

In the APS work routine, an RSS feed from their CMS delivers relevant news content to the APS. Two subcategories of APS approaches were identified: semi-automated and fully automated social media publishing. In the semi-automated approach, journalists continue to select, edit, or delete text and decide when and what to publish, while the fully automated approach does not involve human intervention regarding content feed and social media publishing. Instead, Echobox decides if, when, and how the imported RSS content is posted on Facebook. AI software utilises publishers’ metadata to make decisions about the best content to post at different times of the day. One of the journalists used a hybrid approach: a semi-automated process for regional Facebook pages and a fully automated process for local pages. They intervened in the automated process when they recognised an issue (I05).

In both approaches, APSs shape journalists’ work routines and decision-making. Echobox claims to predict social virality using an AI-powered ranking system and suggests posts accordingly (see Echobox, 2024). The interviewees elucidated that APSs predict how posts perform based on the times users are on Facebook and the topics that have performed well in the past. APSs then calculate a score and list articles by ranking. A journalist illustrated how she processes the suggested list using a semi-automated process:

Echobox displays all new articles, and I can decide whether to place them in a queue for Facebook, where AI decides when to publish them or whether to skip the article. The next step is… what do I do with the teasers? There are two ways: one is to write another teaser that will also entice people to click… or if we have a strong title—the only thing we want to keep—then we publish an article without a teaser… The third decision is the timing. Should this article be published immediately? … Or can this article be published during the day, at a time chosen by AI? Or is it an evergreen article for the region, meaning it can be published several times? My last step is to check whether to moderate comments somewhere.

(I10)

The software determines journalists’ work routines, and their decisions are based on information provided by the APS. A second journalist explained she wants to retain some editorial control over the publishing process on social media while also relying on the virality score of the APS:

You can run an automatic feed with Echobox; you wouldn’t have to look at it at all; Echobox would do it all by itself, pick out teasers, pick out text, and post it. However, we don’t want that because we still say, okay, it’s the mix that makes it work, the mix of AI and journalists. We also want to benefit from our experience.

(I03)

The journalist framed the automation issue in terms of editorial control. The semi-automated approach was the most popular among interviewees who wished to maintain editorial control but continued to be informed by the system’s suggestion function. In each case, Echobox’s virality score standardises engagement work and shapes the labour process.

4.3. Products of Automated Social Media Publishing

Incorporating AI into the selection, editing, and publishing processes affects the quality and quantity of social media news products (see Newman et al., 2024). Some interviewees noted that with fully automated publishing, errors or misinformation in the content feed could go undetected; others have observed algorithmic biases that limit news diversity. For example, one interviewee who tested the automated approach said, ‘The Echobox algorithm says “Crime is the best! Why aren’t you posting police reports?” Suddenly, only police reports were posted, and then, of course, we come to the point where media reach isn’t everything’ (I07). Another journalist who used the semi-automated approach commented on the need to post stories that do not rank highly in the suggested posts, noting that ‘sometimes we also have text that has a low score, but where we say that these are topics that are important to us and that we still want to post’ (I03). Conversely, an interviewee noted that AI could provide useful suggestions that human journalists had not considered (I08). As such, APSs may help counter journalists’ biases in some cases while introducing biases for particular types of content in others (Gutierrez et al., 2022). In the semi-automated approach, journalists incorporate a combination of journalistic standards and values, learned social media logic, and APS scores. For some journalists, this means emphasising human interest and emotional news frames, regardless of AI recommendations. One of the journalists explained the following:

We ask: What effect does the text have on users? … Does it have a service aspect? Can it… somehow evoke emotions? Is it an emotional story because it makes me happy… or is it a topic that can trigger interactions? We want to make users aware of our content, and we also want to get interactions and feed this Facebook algorithm.

(I03)

Another journalist said she sometimes edits suggested posts to make them more emotional (I08). She applies human-learned social media logic in addition to the machine-learned suggestions and emphasises that she tries to fulfil the given KPIs: ‘you try to achieve them… you also know which topics are performing well from personal experience’ (I08).

Respondents who adopted a manual approach were also influenced by their understanding of algorithms. They described making publishing decisions based on judgements about ‘topics that affect as many people as possible’ and favouring more emotional content (I02). In contrast, another journalist argued, ‘We try to get our journalistic message across…we consider Facebook a fully fledged channel, and if someone follows us on Facebook, they receive all important news’ (I01). Respondents also reflected on the challenges in understanding algorithms, with one suggesting that ‘The algorithm is the black box… adapting to Facebook’s algorithm is one of the biggest types of homework Facebook poses to newsrooms’ (I09). Essentially, journalists’ decisions are shaped by algorithms (and their beliefs about them), even when they do not directly employ APSs. This finding has implications for news products reaching social media users.

The news is not the only product of social media publishing. Journalists are also in the business of producing (loyal) audiences who will either subscribe or be offered to advertisers (Smythe, 1981). Therefore, social media audiences are a product and measure of success. Although employing APSs increased overall audience engagement (I03, I06, I05), the types of audience engagement attracted by automated posts could include more toxic comments, such as hate speech and other forms of undesirable discourse. Community management is a key commercial feature of social media (Gillespie, 2018) and an important aspect of producing and maintaining news audiences on these platforms. Essentially, APSs affect audience engagement with news content and the diversity and quality of news on social media platforms.

4.4. Conflict and Resistance Following the Implementation of APSs

Implementing new technologies in newsrooms entails negotiation, resistance, or conflict. APSs such as Echobox promise to improve workflows and increase audience engagement (see Echobox, 2024). However, news organisations often integrate new technologies at the expense of journalists’ autonomy to enhance efficiency, reduce costs, and increase profits (see Cohen, 2015; Örnebring, 2010; Chillas & Baluch, 2019). As APSs are designed for engagement and follow social media logic, they may also conflict with the public service mandate and journalists’ professional ideologies (see Bernstein et al., 2021).

Some journalists piloted automatic APSs but returned to a manual approach to ensure brand reputation, journalistic standards, adherence to editorial guidelines, and news diversity. One of the interviewees explained, ‘Media reach is not everything… we want to offer a mix of topics and show the diversity for which the brand is known. That’s why we switched off Echobox. It was only beneficial when resources were limited’ (I07). Despite these concerns, other journalists have adopted fully automated methods. After testing, an interviewee found that AI performed better than local journalists.

Echobox knew better when to display which content. It’s the ‘editor’. The interaction with human-controlled Facebook posts was worse than those controlled by AI. You really have to say, okay, maybe I don’t know better. It is possible that AI is smarter than us.

(I05)

An editor-in-chief of a smaller organisation explained, ‘We run it automatically… initially we monitored the posts, and sometimes Echobox’s suggestions seem nonsensical, but either you do it or you don’t. So you need a certain amount of trust, and traffic tends to increase’ (I06). Automation can meet organisational goals but also creates a sense of alienation from the production process (Braverman, 1974), contravening journalists’ professional ideologies and requiring emotional labour from journalists (Ahva & Ovaska, 2023; Cohen, 2019).

While some journalists expressed satisfaction with the support and additional time provided by APSs (I10, I03, I08, and I05), their resistance was evident through individual actions. Many journalists use a semi-automated or hybrid approach to maintain editorial control; revise content with the potential for toxic comments, rectify errors and misinformation, or ensure appropriate news diversity. These journalists emphasised the need for intervention by experienced human editors. One of them explained how full automation harms her reputation as a journalist, ‘And if there are mistakes… and they are shared automatically… then the user’s reaction is “Did you have an intern do this?” These are typical comments’ (I08). These journalists maintain a degree of autonomy in the news production process, which would not be possible using a fully automated approach (cf. Olsen, 2023).

The journalists, regardless of the use of APSs, also described spending considerable time moderating, blocking, or reporting toxic comments on Facebook, which is often emotional work. One of the journalists linked the increasing volume of hate speech to social media logic, stating, ‘news topics that provoke engagement also provoke hate speech… it’s a snowball effect’ (I03). Another journalist emphasised the enormous workload: ‘Sometimes it feels like tilting at windmills, but we fight every day’ (I10). They acknowledged that unchecked automated posting could escalate hate speech. The challenges of moderating comments led several organisations to adopt a second AI-driven software, which may raise new ethical concerns. The balance between commercial goals, editorial control, and maintenance of professional standards is at the core of conflicts regarding AI use in newsrooms.

5. Conclusions

Rapid technological advancements, challenging economic conditions, media consolidation, and the centralisation of news functions are among the factors influencing the implementation of AI tools in the production processes of German news organisations. Alongside dependence on social media platforms, these factors have prompted some companies to introduce AI-based APSs to improve workflow efficiency in newsrooms and secure or increase social media engagement as a means of generating profits. APSs have the potential to automate the entire social media publishing process based on RSS or content feeds. LPT provides a framework for examining the various uses of APSs, the effects on journalists’ work, and, ultimately, the impact on news products. In addition, it examines negotiations, resistance, and conflicts regarding the implementation of this technology.

APSs relieve journalists of repetitive and tedious tasks by automatically importing, selecting, editing, and publishing news content on social media platforms. This automation frees time for more demanding, creative, and fulfilling journalistic tasks, which increases the quality and quantity of overall news work (Simon, 2024). Rather than simply deskilling journalists, social media journalists are still required to possess multiple skills, although this may reduce the need for specialised social media editors (Örnebring, 2010). This trend will make some social media work redundant and continue to transform their roles in the news production process.

APSs can also enable increased managerial control and reduce journalists’ autonomy (see Carlson, 2015; Chillas & Baluch, 2019). The degree to which social media publishing is automated determines the extent of editorial control. We identified two main approaches to automation among the interviewees: fully automated and semi-automated. A fully automated approach eliminates the need for human input at the publishing stage; it is completely AI-controlled. However, this fully automated approach may lead to a decline in news diversity and quality on social media platforms. From the perspective of LPT, a fully automated approach leads to the alienation of journalists from the social media production process. The struggle over automation is articulated in terms of journalists’ professional values (autonomy), while the acceptance of semi-automated social media publishing indicates consent for new types of technological exploitation (Cohen, 2012). Automation can also negatively impact the news product. AI continues to prioritise platform logic (see Simon, 2024), with implications for news consumers, especially young people, who predominantly consume news via these platforms (Newman et al., 2024).

For our interviewees, the tension between commercial imperatives of efficiency and increased user engagement on one side and the desire to maintain editorial control on the other manifested in their resistance to fully automated AI publishing. They indicated concerns over journalistic ethics, professional status, impact on the public and additional emotional work (see Ahva & Ovaska, 2023; Cohen, 2019). Therefore, most respondents opted to use a semi-automated approach and retain the ability to select and adjust their AI-suggested posts. However, they still rely on a list of news suggestions generated by AI that influence their professional decision-making processes.

For the most part, the journalists interviewed insisted that human editorial resources are still required to constantly monitor, optimise and feed AI with human-produced content. They see themselves as bulwarks, ensuring the diversity and quality of news and trust in journalism. They do not reject AI-driven APSs outright, as they hope to improve AI systems by developing and implementing journalistic standards.

6. Limitations and Further Research

We encourage researchers to expand the study of AI and automated publishing to other national contexts and conduct larger qualitative and quantitative studies to gain a deeper understanding of the impact of AI on social media news publishing. We would also encourage further research on how to implement journalistic standards in AI-driven tools, considering democratically relevant characteristics (Mitova et al., 2023). A third avenue of future research is the impact of APS-based social media publishing on the journalistic tasks of news gathering, topic selection and editorial style.

The rapid development of AI technologies in recent years, particularly with the launch of ChatGPT, may have introduced new features in APSs that were not considered in this study, as the interviews were conducted in 2022. According to the Echobox website (Echobox, 2024), the company launched a ChatGPT functionality feature to produce AI-powered news content. Although we cannot predict how these tools will be adopted, some of the concerns and challenges outlined in this study are likely to remain.

It is also important to stress that the quality and type of data provided by publishers for AI processes and how it is structured can significantly influence the content published on Facebook. Furthermore, we do not know to what extent journalists used the available options and functionalities of APSs. For example, journalists could apply keyword-based rules to exclude content that contains specific terms. We considered the perspectives of journalists who may not have been involved in the technical processes of data provision. Therefore, it would be beneficial for future studies to include a technical or interdisciplinary perspective.

According to the company’s website, thousands of news publishers across more than 100 countries use APSs (see Echobox, 2024). Recent qualitative studies examining AI trends in journalism show that journalists in various European countries and the United States have also implemented the service (see Echobox, 2024; Nielsen & Cherubini, 2022; Newman, 2022; Paik, 2023; Sánchez-García et al., 2023; Verstappen & Opgenhaffen, 2024). This highlights the need for further research on AI and social media publishing.