Datafied news work: A scoping literature review

Datafication is the assignment of numerical values for digital registration of informational aspects to phenomena in the (physical as well as virtual) world for the purpose of computational quantification, analysis, and mining (Flensburg & Lomborg, 2021; Hansen, 2015; Lycett, 2013). It “refers to the process of rendering into data aspects of the world not previously quantified” (Kennedy et al., 2015: 1), which follows the global ubiquity of networked digital technology. Datafication is closely connected with the availability of Big Data, which denotes data of such volume that it borders on being exhaustive (Mayer-Schönberger & Cukier, 2013). However, “Big Data is not simply denoted by volume” (Kitchin, 2014: 2), but also by granularity, scalability, and its inherently relational nature (boyd & Crawford, 2012; Kitchin, 2014; Kitchin & McArdle, 2016).

Different types of data are part of processes of datafication. Arguably, the most prominent type of data in this context is user data – that is, data about the users of digital technologies and services (e.g., information about age, gender, geography, devices, etc.) as well as behavioural data (information about what users do). Additionally, data can have the form of structured information about content or formats, metadata (Kristensen & Sørensen, 2023), or the form of data used in journalism coverage. As such, data can originate from both within and outside the news organisation and exist in both confined and publicly available reservoirs.

One consequence of datafication is that the pervasive organisation of information in digital data as well as the knowledge that such data exist in the first place shape how actors think and form knowledge about the world. Such processes of “datastructuring” (Flyverbom & Murray, 2018) lay the groundwork for an epistemological context, where actors’ conditions for knowledge production exist within the frameworks of datafied practices and are shaped by the (infra) structural properties of digital communication technology. The question remains, however, of what such datastructuring looks like in the practical contexts of news work and news workers.

It is important to emphasise that we focus on news work in general rather than on journalism in particular. We do so because the news is a result of the work of a much broader group of actors than only journalists and editors, and the process of making and distributing news includes many other activities than producing journalistic content. The newsroom is, for example, crowded with not only journalists, but also social-media managers and other types of “interlopers” (see, e.g., Belair-Gagnon & Holton, 2018; Vulpius, 2023), and journalists have come to conduct much work in addition to journalism (see, e.g., Bro et al., 2016; García-Avilés et al., 2004). Journalism and editorial decision-making are at the core of this work, but news organisations are complex organisations where many different jobs support the production and distribution of news. On these grounds, we believe that a holistic rather than narrow approach to news work provides a more correct picture of the actual consequences of datafication.

To employ such a holistic approach to reviewing the scholarly literature on news work, we use Kammer’s (2023b) model of “spaces” for datafication in the media industries. Drawing upon Porter’s (1985) managerial model of the “value chain”, this model distinguishes between five different spaces in media industries, each of which conditions and is conditioned by datafication in different ways. While these spaces represent different parts of the process of news work, they are closely integrated with each other.

The first space concerns content production. In the context of news work, newsrooms will typically constitute the physical location for this space, where the journalists and the editors are located and are engaged in making the news.

As news work is organised and takes place across different spaces, it is highly heterogeneous in nature: Different actors, with different understandings of data and their value in relation to their work, need to cooperate. We understand the cooperation of this heterogeneous group of actors to be made possible by the functioning of data as boundary objects; that is, as “objects which are both plastic enough to adapt to local needs and the constraints of the several parties employing them, yet robust enough to maintain a common identity across sites” (Star & Griesemer, 1989: 393). Boundary objects, then, are concrete objects (or concepts) which may have a core meaning but can, at the same time, be understood and constructed in highly different ways by different social groups. Analysing collective heterogeneous work (their analysis concerns the Museum of Vertebrate Zoology at Berkeley), Star and Griesemer (1989) found that it is common for actors from different social worlds to come together and address objects that have different meanings for each social world; mismatches, then, become ground for negotiations about meaning in context.

Approaching data in news work as a boundary object is a useful conceptual framework for understanding datafied news work. The properties of the platform infrastructures that support the capturing and use of data allow for data to be employed by different groups of actors, without necessarily explicitly negotiating their meaning. Indeed, platform infrastructure properties make it easier to “transcontextualise” datafied categories “that are projected onto narrower contexts or broader ones, projected onto different ‘objects’ – activities, interaction or spaces” (Lury, 2021: 150). That is, platform infrastructures make it possible for data to be detached from their context and for their epistemological value to be “constituted in circulation [emphasis original]” (Lury, 2021: 168). In Lycett’s (2013: 382) words, data are dematerialised and liquified before they reach density as a “(re)combination of resources, mobilised for a particular context, at a given time and place”.

In the context of news work, data circulate in organisations across entities and departments as boundary objects, assuming different values and relevance. At the same time, they are embedded and employed in different contextual settings, which gives the data new value and meaning. Data are not one stable thing in a news organisation; rather, they are a productive construction, the meaning of which is constantly formed, and which constantly forms meaning within the organisation. These differences in meaning also serve to distinguish different social and professional groups from each other. This way, understanding data as a boundary object that gets transcontextualised in news organisations allows us to understand the ways in which data are perceived and used differently as well as the properties of data that remain stable across spaces of news work.

The review takes the form of a scoping literature review, which treats scholarly publications as its empirical data. This type of literature review typically searches a range of relevant databases and uses specific criteria for inclusion and exclusion to systematically sample the appropriate literature for the review. The literature is then “described and mapped to provide both an overall picture of the current state of the evidence in the field and to identify and highlight knowledge gaps in the area” (Munn et al., 2018: 4; see also Arksey & O’Malley, 2005, for further elaboration); in this sense, the scoping literature review not only identifies the state-of-the-art of the research but also critically examines it to identify ways forward for future research. Indeed, this type of literature review typically deals with emerging concepts and research and with the available evidence, identifying paths for future research as the field settles.

Scoping reviews “share several characteristics of the systematic review in attempting to be systematic, transparent and replicable” (Grant & Booth, 2009: 101). However, the systematic review originates in the medical sciences tradition where results are synthesised quantitatively, and it is generally better suited for synthesising results in research areas of a certain maturity (Munn et al., 2018). The scoping review combines the methodological rigor of the systematic review with the awareness and sensibility of an emerging and developing research area where the scholarly focus points in various directions.

For the sampling of our empirical material (publications), we initially used the Scopus database. Through several rounds of brainstorming and pilot testing, we identified the most adequate search terms to capture the publications we were interested in, namely those that reported studies of datafied news work; these terms are “data*”, “metrics”, or “analytics” in combination with “newsroom”, “{news work}”, or “{news industry}”. Furthermore, at this point, we limited the search to publications in English and within the subject areas “Arts and Sciences”, “Business Management and Accounting”, “Computer Science”, “Decision Sciences”, “Economics, Econometrics and Finance”, “Social Sciences”, and the somewhat broader “Multidisciplinary”. ⁽¹⁾

As only a few Nordic peer-reviewed journals are included in the Scopus database, we additionally manually screened all publications 2005–2024 from the Nordic journals MedieKultur, Journalistica, Nordicom Review, Nordic Journal of Media Studies, and Norsk medietidsskrift. This was done to ensure that the journals with the greatest geographical proximity to the empirical focus of the literature review were also included. Finally, we searched the Research Portal Denmark (which indexes some Nordic research), using the broad search terms “nyh*” [news] or “journalis*” [journalism]; these search terms work in Danish, Norwegian, and Swedish.

Taken together, these enquiries resulted in 678 potentially relevant publications. By reading abstracts and keywords, we screened each of these publications to extract those relevant for our purposes, dividing the work equally between us. In this process, we excluded literature reviews (since they would not contribute with new knowledge about the subject matter but, rather, synthesise existing knowledge), conference papers, publications without a Nordic focus, and “grey literature” (i.e., non–peer-reviewed publications like reports and whitepapers). These criteria mean that some publications are excluded from this literature review even though they do explore aspects of datafied news work (e.g., Andreassen, 2020; Jørstad, 2022; Lindskow, 2015; Ministry of Culture, 2017, 2024; Møller, 2022b, 2022a). Finally, we collected additional relevant literature through the bibliographies of the publications that made it through the screening process. The titles we identified this way then went through the same screening process as the publications we had initially found.

Through this process of separating the signals from the noise, we ended up with 32 publications for the literature review. The steps in the process of identifying, screening, and selecting the relevant literature are documented in Figure 1.

Figure 1

Flow diagram for the selection of literature for the review

Source: adapted from Page et al., 2021: 5

Despite rigorous work to identify the relevant literature, no review can be fully comprehensive. So, potential blind spots of this review include studies that were not yet published at the time of our sampling, studies where the abstracts are not written in English, Danish, Norwegian, or Swedish, and studies that investigate datafied news work but use other terms than our keywords to conceptually frame the work.

After the initial coding of abstracts for relevance and Nordic orientation, we read the relevant publications in their entirety, systematically registering and coding the information we needed to answer our research questions. Again, the task of reading through the publications was divided equally between us. Throughout the coding process, all instances of doubt and uncertainty were discussed to maximise reliability.

In the newsroom, data are predominantly used in two ways, namely to inform editorial gatekeeping decisions and to enable post hoc evaluation of what is already published. First, journalists and editors use data to inform decisions about what to publish. These actors “are strongly committed to metrics in their daily news work, largely relying on audience preferences in deciding what is worth publishing” (Ekström et al., 2022: 762). Kalsnes (2019: 47), likewise, found that the availability of data makes it easier “for the news editor to prioritize between stories, topics, and genres”, while Ahva and Ovaska (2023) laid out how data are also used for editorial experimentation with writing style and the framing of news stories.

Overall, the scholarly literature agrees that audience metrics have an impact on which news is produced – but there are different assessments of what exactly the impact is. In an early study of journalism for the web, for example, a site and community manager argued in an interview that real-time tracking of online readership was “actually the only parameter [for news judgment]. We utterly do not care how relevant we think the story might be” (Kammer, 2013: 149). At the same time, but less bombastically, Karlsson and Clerwall (2013) found that while “clicks” do affect the news judgment and the editorial decisions about what to keep on top of the front page, it is only one signal among many to influence the process of editorial decision-making. Along the same lines, and almost a decade later, Ekström and colleagues (2022: 763) found that “although metrics has become a prime standard in deciding what is valuable to inform about, there are situations in which metrics are temporarily overridden in favor of fundamental epistemic claims of investigative reporting, novelty and immediacy in news journalism”.

According to Salonen and colleagues (2023), knowledge about audience behaviour accumulates in the newsroom over time, influencing and informing everyday gatekeeping practices. That is, journalists and editors learn from audience metrics. Against this background, and based on extensive newsroom observations, Kristensen (2023) proposed “expected reception” as a central news value in the digital newsroom, emphasising how news workers embed their experiences from audience data (including data from algorithmically driven third-party platforms) into the everyday practices of decision-making. This decision-making pertains not only to what becomes news but also to post-publication adjustments of, for example, headlines (e.g., Ahva & Ovaska, 2023; Salonen et al., 2023), the composition of news stories on the front page (e.g., Karlsson & Clerwall, 2013), placement behind or in front of a paywall (e.g., Kalsnes, 2019; Salonen et al., 2023), and self-evaluation by journalists (e.g., Bucher, 2017; Kristensen, 2023).

Finally, automated content production draws heavily upon structures of datafication. Sometimes labelled “robot journalism”, this type of news work is of a different flavour than the practices just mentioned, since no human actors are directly involved in the production of content once the system is developed. In the literature, this type of news work is found to exist in the form of predefined templates, where the blank spaces are filled out automatically with structured data that the newsroom retrieves automatically from elsewhere (see, e.g., Sirén-Heikel et al., 2019). These data would typically concern sports results, weather, or election results, all of which are highly standardised and can be presented in uniform ways. It should be noted that the reviewed literature does not include analyses of the editorial use of generative artificial intelligence, simply because no such literature was found in the sampling process – but we expect such research to figure more prominently in future literature.

Another well-documented datafied newsroom practice is that of data journalism. This genre of journalism can use a variety of data types as evidence for a story, and stories often have infographics or interactive elements driven by large datasets (e.g., Engebretsen et al., 2018; Ishøj, 2009). Data are usually obtained from an outside source, such as national statistics authorities. Furthermore, data can be crowdsourced from readers; for example, Appelgren and Nygren (2014) documented how Swedish news outlets used crowdsourced data from users to report on mortgages and dog breeds.

There are two central insights to gain from this part of the literature. First, in the context of research of the newsroom, “data” is basically another term for audience metrics. As mentioned above, the datafied society is characterised by an abundance of different types of data, but in this particular context, there seems to be a quite narrow approach to what the data are and how they are useful. Second, actors in the newsroom largely perceive data as a resource, even if neither they nor the research articulates it this way. But what is clear is that the data are something that the journalists and editors can use in the process of deciding what is news, either by reflexively consulting the data repertoires that are available to them (through, e.g., dashboards in the newsroom or knowledge graphs) or by tacitly drawing upon the experience they have built over time (about, e.g., the “expected reception”).

A well-documented use of data in this space is the development of technical systems for personalisation and automated content curation (Borchgrevink-Brækhus, 2022; Sørensen, 2020). This typically comes in the form of recommender systems, which match behavioural and user data with content to algorithmically expose the individual user to the content that is deemed most relevant for them. Across the literature, there is agreement among informants that recommender systems only work well in combination with human actors. Borchgrevink-Brækhus (2022), for example, documented how news organisations use algorithmic curation to put together the front-page selection of online news – but also how this automated curation is not left on its own. For instance, at Aftenposten and Bergens Tidende, the first six items on the front page are manually curated by humans, and the rest are the result of algorithmic curation. Andersson Schwarz (2016) found a similar effort to ensure diversity in news selection, exploring how data are used in recommender systems as a means to increase diversity in the content that is presented to individual users.

Along the same lines, some industry actors employ machine learning to identify patterns in the user data that can both lead to better user experiences and increase traffic. Svendsen and colleagues (2019), for example, documented how machine learning techniques at Polaris Media identified old content that could be reactivated to improve performance. Other industry pilot tests, however, find that algorithmic curation would not necessarily lead to increased page views (Borchgrevink-Brækhus, 2022). With a slightly different focus, Holand and Engan (2020) studied automation in practice, exploring how the Norwegian wire service NTB can automatically publish content on parts of the digital news offerings of Polaris Media.

In this space, the term “data” does not only refer to audience data, but also to structured content data or metadata. Indeed, a prerequisite for algorithmic recommendation and for machine learning to work is data annotation – that is, the labelling of data to categorise it in a way that is machine-readable. In this process, news content becomes datafied through descriptions in the form of metadata tags; it is the job of machine-learning algorithms to then match the datafied user with the appropriate datafied content. In this context, discussions revolve not only around availability of data, but also around data quality.

For instance, the annotation efforts documented and analysed in case studies of Amedia (Kalsnes, 2019), Sweden’s Radio (Rolandsson et al., 2022), and the Danish Broadcasting Corporation (Sørensen, 2020) show that these organisations were lacking in terms of both the amount and quality of metadata. The Danish Broadcasting Corporation, for example, used the European Broadcasting Union metadata standard to categorise video content. This specification ensured that the public broadcaster could document its fulfilment of content obligations to the Danish Radio and Television Board. However, the EBU standard and textual content descriptions did not always make for sensible recommendations, and the broadcaster had to experiment with a more tailored combination of taxonomies for the recommender system. A similar course of action occurred at Danish tabloid Ekstra Bladet (Kristensen & Hartley, 2023), illustrating how old and new datafication strategies can be challenging to merge and manage in the face of technological developments. In this space, then, data are perceived as a resource because they allow for datafied algorithmic distribution, but only insofar as they are of good quality.

The third space of datafication relates to news work that sustains news organisations financially, more specifically work concerning subscription sales, customer data management, and advertising sales. As mentioned earlier, few publications look directly into this space, although questions of monetisation and commercialisation are themes in research of other spaces (e.g., Kalsnes, 2019; Tenor, 2024). The studies centre around the transformation from paper to digital and the challenges this brings to the traditional business models of news media. Datafication of user characteristics and behaviour is front and centre, and a central goal is to reclaim advertising revenue lost to tech companies through programmatic advertising.

In this space, the term “data” refers to behavioural user data, and discussions revolve around the issue of quantity and scale. Sjøvaag and Owren observed how Norwegian companies Nordsjø Media and Agderposten Media benefitted from consolidating with larger newspaper chains, as it gave them access to “skills in porting users to the digital platforms, their sophisticated algorithms feeding into editorial product development, and the audience metrics that cater to advertising agencies’ need for data” (Sjøvaag & Owren, 2021: 11). The same study shows that while the CEOs of five news corporations in Denmark, Norway, and Sweden realised that their unique local content was hard to mimic, their offering to advertisers was subpar to the offerings of, for example, Google and Facebook. As such, mitigating losses might only be done through mimicking the datafication strategies and scale of these data-intensive companies. The inevitability of scale in this space of datafication has also been recounted by Barland (2013), who documented how media conglomerate Schibsted advertised for digital products or services owned by the organisation itself on news websites VG and Aftonbladet and thereby created new revenue streams.

The fourth space of datafication is connected to the upper management of news organisations, and the data referred to in the literature is predominantly audience metrics. In many ways, the same data and data sources are used here and in both the newsroom and commercial department. However, in the managerial setting, the use takes the shape of “audience intelligence” (Andersson Schwarz, 2016). Furthermore, in many news organisations, the management space has the final say when it comes to the presence and emphasis on audience data in the organisation. Lindén and colleagues (2022: 336), for example, documented how Amedia developed a “strong leadership and organisational culture” across its media offerings which entailed a focus on monetisation of journalism on the basis of audience measurement data. Likewise, they found that Dagens Nyheter had success with growing their subscription figures by focusing on key performance indicators and visual audience measurement dashboards in the newsroom; a similar top-down approach to datafication was also found in Jysk Fynske Medier (Kristensen, 2023). Research into public service media emphasises how their management goals differ from commercial management goals, and how that shapes views and uses of data (e.g., Andersson Schwarz, 2016; Rolandsson et al., 2022; Sørensen, 2020). At SVT, data were perceived as a means to better service the viewers; this way, instead of being linked to an outside commercial pressure of maximising “clicks” and “viewers”, data use was linked to achieving the diversity aim of reaching as much of the Swedish population as possible (Andersson Schwarz, 2016). In this space, then, data have the form of audience metrics and function both to identify and evaluate goals, providing insights as to whether organisational ambitions are met.

The fifth and final space for datafication concerns the systemic space that exists beyond individual news organisations: the space between industries, where the news industry interfaces with other industries or sectors in exchanging data of various types. The literature about this particular space is still quite limited, and it mainly concerns mapping infrastructural relations and dataflows and discussing organisational dependencies on platform companies. This research provides insights into how data cross the boundaries of news organisations. The understanding of data that is at play in this space is related to actors outside of the news industry who work to capture and structure data coming from news organisations through provisions of platform and infrastructural services.

The literature that explores this space is highly focused on the continuous negotiation of dependencies and autonomies in the complicated relationship between news organisations and technological infrastructures. On the one hand, news organisations (like all other providers of technological infrastructures that exist in the digital realm) need the tools and services of technology companies in order to work and to reach their audiences; on the other hand, the news organisations have no interest in being dependent upon the technology companies that are also competitors for advertising revenues as well as data and power. Unpacking these tensions, Kristensen and Hartley (2023) mapped the externally developed infrastructural systems that news organisations use to collect and display audience metrics, placing ads, hosting, and building services (including recommender systems). In a study of newsroom and management practices, Salonen and colleagues (2023) furthermore found that data from social media are not experienced as reliable in the same way as data that are collected by the news organisation itself.

This part of the literature also explores the relationship between news media and platforms, looking into how technology companies may capture audience data through various types of collaboration with the media. Kammer (2023a), for example, explored longitudinally how drawing upon technical infrastructures enables “resource exchanges” between news organisations and third parties, as mobile news apps connect to servers that belong to others than the news organisations. Over the analysed period of time (2016–2021), the complexity of the network of third parties increased, even if a few actors (e.g., Alphabet, Amazon, and comScore) maintained both a stable and far-reaching presence in that network. The study, however, does not map which data travel between news organisations and third parties. Along the same lines, Appelgren (2017) analysed privacy-agreement texts and cookie-consent information on news websites. She found, among other things, that data-sharing with third parties occurs on numerous news websites and often has a paternalistic reasoning behind it (i.e., the benefits for the audience rather than for the news organisation are emphasised).