• Čeština


Keynote Speakers

Scaling public access to journalism using automation
David Caswell

(BBC News Lab)

This talk will focus on one particular opportunity that is available from the application of AI technologies in newsrooms: the communication of news stories in ways that are personally relevant to audiences and are genuinely accessible to them. This opportunity to adapt news content to the individual news consumer will be contrasted with the traditional one-size-fits-all model of journalism, in which the individual news consumer was expected to adapt to the news content. Experimental and production examples of this application of AI to news at the BBC will be reviewed.

Deepfakes and synthetic media: Global threats and solutions for journalists
Sam Gregory

(Program Director of WITNESS)


When algorithms recommend what is new(s): new dynamics of gatekeeping and agenda-setting in newsroom innovation labs
Hannes Cools, Baldwin Van Gorp, Michaël Opgenhaffen
University of Leuven (KU Leuven), Belgium.

Newsroom innovation labs have been created over the last ten years to develop algorithmic news recommenders (ANR) that suggest and summarize what new(s) are. Although this ANR is still in an early stage and has not been implemented in the entire newsroom yet, they have the potential to change how news workers fulfill their gatekeeping and agenda-setting role (Deuze, 2005; Helberger, 2019; Möller et al., 2018). With an increased news worker-ANR interaction, it has become more complex and unclear who makes the decisions, who selects the news, and who sets the digital agenda (Thurman et al., 2019). First, this study focuses on how the ANR-news worker interaction can potentially change the process of gatekeeping in the newsgathering section. Second, this chapter investigates how this ANR could have an impact on the autonomy of the news workers’ role as agenda-setter. To advance our understanding of the changing roles in the newsroom, this study conducts expert interviews with 16 members of newsroom innovation labs of The Washington Post, The Wall Street Journal, BBC, Der Spiegel and Bayerische Rundfunk (BR). Results show that when news workers interact with ANR, they rely on the suggestions and the summaries to evaluate what is newsworthy, especially when there is a newspeak (elections, a worldwide pandemic). With respect to the agenda-setting role, the news worker has still full autonomy, but the ANR creates a ‘positive acceleration effect’ on how certain topics are put on the agenda. Although there is some distrust in using the ANR, if the news worker’s autonomy is safeguarded and the technology provides enough quality leads and summaries, then it appears from the interviews that the news worker will continue to use the ANR. This research could contribute to a better understanding of the use and influence of ANRs on the role of news workers, both for journalism scholars and journalism practitioners.

Automated Fact-Checking in Czech Language
Jan Drchal

Czech Technical University

We give an overview of the current state of the AI-based automated fact-checking project aimed at the Czech language. The task is to evaluate textual claims with respect to the ground truth document database. The system’s expected output is one of the three categories: the claim may be supported, refuted, or there is not enough information to decide. In the case of the former two outcomes, the output is further augmented by a selection of documents providing evidence. The claim has a form of a single sentence typically. Our ground truth database is based on the Czech News Agency archive of roughly 2 million 2000-2020 articles.

Similar to state-of-the-art approaches, which primarily focus on English, the architecture of our system is based on two modules: 1) Document Retrieval (DR) module, which searches the documents for the evidence, and 2) Recognizing Textual Entailment (RTE) module providing the final classification of the claim.

Our approach utilizes large BERT (Bidirectional Encoder Representations from Transformers) [1] neural network models for both stages. We primarily employ multilingual models pretrained on related language modeling tasks allowing relatively small training data for our target problem.

Still, the main obstacle for deploying state-of-the-art fact-checking methods is the sparsity of Czech language data needed to train the models. We have translated the English Wikipedia-based FEVER [2] dataset aimed for the document retrieval stage to overcome this problem. More importantly, we are currently in the process of building and annotating a genuine Czech fact-checking dataset that is based on the Czech News Agency data.

We have designed several proof-of-concept tools that employ the models mentioned above as well as other methods, allowing for side-by-side comparison with, e.g., classical keyword-search approaches.

We also employ methods of explainable AI to emphasize the textual segments important w.r.t. to the models’ confidence. The aim of the explainability methods is two-fold. Firstly for us, the designers, they help in understanding the inner workings of the complex system. Secondly, they lead to faster navigation in the evidence documents for ordinary users of the fact-checking tools.

BulletinBot: How AI-based text-to-speech could change the news industry
Stuart Duncan

Communication and Culture, Ryerson University, Toronto

It’s no secret that the journalism industry is in a state of upheaval. With declining revenues and shrinking subscriber support, the once-powerful news industry is a shell of its former self. News organizations have seen their positions as a primary distributor of news and information disrupted as online users, particularly younger audiences, turn to social media and search platforms to discover the news and current events. Faced with these challenges, how can the journalism industry adapt to ensure its future relevance? One approach would be to borrow a page from social media and search giants and embrace algorithmic-driven automation and personalization.

There are countless ways that automation will impact the field of journalism and my paper focuses on how improvements in text-to-speech technologies could affect news presentation roles. Text-to-speech systems, digital systems that synthesize text into audio-based speech, no longer feature the tinny robotic results of the past. The adoption of machine learning approaches has allowed these systems to replicate speech that is almost indistinguishable from a human voice. Cloud-based computing systems have introduced relatively simple and affordable implementations of text-to-speech systems, opening up realistic speech synthesis technology to anyone with a credit card and basic digital skills.

This paper will specifically examine how emerging text-to-speech technologies could be combined with news personalization techniques to deliver automated audio-based news bulletins. With the creation of my open-source BulletinBot prototype, a simple automated news bulletin system, my research explores how news organizations could automate elements of the news delivery process and illustrates issues that surround this approach. Within a radio context, news bulletins are short summaries of important news items. Featuring a few items, usually, 30 seconds each in length, bulletins quickly get listeners up to speed on what is going on in the world. News bulletins are a staple of radio broadcasting, and their longevity is a testament to their usefulness, and this project attempts to adapt that format by creating a small-scale working prototype of an automated and personalized news bulletin system.

While text-to-speech technologies have been a significant focus area of computer science research, there has been little inquiry into the sociological effects of the technology from a communications perspective. My research will examine what a more advanced version of an automated news bulletin system might offer to the news industry and explore potential ethical and editorial pitfalls of such systems. Furthermore, my paper examines how news organizations will have to adapt their processes to facilitate novel news distribution approaches, an example of which is illustrated by the prototype created for this project. This project, and the software created as part of it, will help facilitate the continued critical examination of journalism automation and start a discussion on the potentially disruptive impacts of text-to-speech technologies.

Introducing AI inside Al Jazeera newsroom
Ahmed El Gody

Örebro University Sweden

Discussions on Artificial Intelligence (AI) shifted from how much human work they will replace to how AI can support journalists in their work and help improve it. Indeed automation and machine learning is changing newsrooms. Almost 60% of US newsrooms are investing in newsroom automation to personalize content and publish them on different platforms as well as understand the nature of their audience, their preference, and interest. AI refers “to intelligent machines that learn from experience and perform tasks like humans,” according to Francesco Marconi in his book Newsmakers Artificial Intelligence and the future of Journalism (2020 p21). Marconi’s central argument that news organizations are not keeping with the pace of technology which caused several issues including shrinking newsrooms, disorientation with their audience, and fake news phenomenon. Running after failed business models, Marconi suggests that news organizations need to make use of AI to help develop journalism. Today, newsrooms need to make serious investments to attract the skills, knowledge, and innovation that journalism needs to optimize the opportunities of AI and reduce the potential harms.

Similarly, a study conducted by Beckett (2019) surveying the views of a sample of people working with AI in a variety of news organizations across the world showed that news organizations have to adopt some form of AI strategy. They need to change their workflows, systems, and recruitment to adapt to the requirements of AI technologies. The study showed that AI is giving journalists more power, freeing up journalists to work, creating better journalism. Further AI helped the public cope with a world of news overload and misinformation and to connect them in a convenient way to credible content that is relevant, useful, and stimulating for their lives (Beckett 2019: 89).

The purpose of this study is to examine the prospects utilizing the use of Artificial Intelligence automated services inside Arab newsrooms studying the case of Al Jazeera. Furthermore, the aim of the study is to examine how Al Jazeera are developing their own versions of robot automated systems for fact-checking. The study is based on interviews with the AlJazeera center for computing research, and AI convergence members inside the Al Jazeera newsroom.

On Applying Automation and AI to Practical Journalism from the Programmers’ Perspective
Kamil Ekštein, Jakub Sido, Miloslav Konopík, Ondřej Pražák

Department of Computer, Science and Engineering Faculty of Applied Sciences, University of West Bohemia

Artificial intelligence approaches have been affecting practical journalism with a gradually increasing intensity and as of now, a number of software applications are becoming integral parts of newsrooms and editorial offices. Their embrace, perception, and prehension by journalists have already been and are going to be discussed widely across various platforms. Nonetheless, there exists a perspective to consider these applications from, which may be fundamentally different. It is the perspective of the technical and natural science professionals (programmers, mathematicians, etc.) who are involved in the design and development of the above-said applications. Yet, the difference might be of certain interest as its investigation could provide the researchers from both the natural science and the humanities sides valuable insights and hopefully answers, too, why specific applications work or fail to work under concrete conditions.

This article is predominantly about the exposure of the technical section of a multidisciplinary team to the (sometimes cryptically) canonic world of journalism while working together on developing automated journalism tools. A vivid recent experience with an application that generates stock exchange news coverage is presented. The software has been already deployed in a practical daily task as it covers a relatively restricted topic domain.

Furthermore, a far more interesting task of generating short summaries of news series is discussed. It depends heavily on our recent research outcomes: We employed the new Czert model (https://arxiv.org/abs/2103.13031) that excels at capturing text semantics in the Czech language. Along with the increasing complexity of the mathematical techniques involved, the divergence of understanding between natural sciences and humanities grew, too. Examples of such terms and statements that are understood as perfectly clear and unambiguous by natural scientists and at the same time as obscurely shady and fuzzy by journalists (and generally social scientists) and vice versa are given. Another observation is described, too, that certain concepts and processes felt as crucial and worth time and effort were neglected as a result of an experience-induced misjudgment of the other section of the team.

Perceptions and attitudes towards AI-generated news: A thematic content analysis of AI and its impact on the craft of journalism as a provider of daily news

Daniela Frumusani, Valentina Marinescu, Silvia Branea,Ramona Marinescu

University of Bucharest, Romania

Bianca Fox

Nottingham Trent University, UK

Existing research on AI in journalism has focused more on understanding and explaining what AI means for the future of the profession, outlining the potential impact of AI in newsrooms or the changes in journalistic routine, and less on how the audience perceives the use of AI-assisted technology in news production and dissemination. This paper aims to refocus scholarly efforts on the audience and continues the line of inquiry on how the audience experience and perceive AI-generated news. Using the work of Clerwall (2014), Melin et al. (2018) and Wu (2019) as a framework, the paper makes a meaningful contribution to the literature by exploring the audience’s experience with predictive news and AI- generated news articles through the participants’ ‘personal experience story’ (Creswell, 2012). We will start our inquiry from the following research questions:
– How do news consumers experience the use of algorithms in determining their own personalized news agendas?
– How do news consumers perceive automated news?
– What is the audience’s attitude towards AI-driven tools being used in news production and distribution?

The paper adopts a futuristic approach and, in doing so, reflects on the audience’s perceived usefulness and ethical concerns raised by the use of AI tools in news production and distribution, before contextualizing the results and making recommendations for journalists. Results show that people have mixed feelings regarding the use of AI in news and, while they fully understand its benefits, there is a consensus that the widespread use of automated news will lead to yet more misinformation. Moreover, results show that the audience-journalism relationship is stronger than ever, people trusting journalists and appreciating their unique set of skills. Overall, this paper shows that if certain conditions are met, AI could lead to a better relationship between journalism and its audience.

Artificial Intelligence in Chinese Newsrooms

Joanne Kuai

Karlstad University, Sweden

Amid a universal challenge to news media in the digital age, how are Chinese newsrooms leveraging artificial intelligence (AI)? This paper will give an overview of the state-of-art of journalism innovation in China. It also gives a brief account of the developing stages of Chinese newsrooms’ adoption of AI, the motivating forces behind it and the implications on journalism and society.
The talk is informed by my years of research on the topic. Through several fieldtrips to China, I have conducted interviews with media professionals, AI tool developers, scholars, educators, and AI scientists. Additional data are acquired through media organizations’ internal
reference reports, publicly available documents and commercial database. The research contributes to the understanding of how Chinese newsrooms are implementing journalism innovation in China’s unique political and social context and add to the discussion of some common debates around AI+Journalism, such as AI’s impact on newsrooms’ workflows, technological challenges, ethical concerns and what is the gap between hype and reality.
By examining how AI is used in Chinese propaganda and censorship, it shows that the same technology can be used for with very different purposes to achieve the goals of different stakeholders. Analysis on China’s increasingly powerful news aggregator apps, led by Toutiao developed by ByteDance (also the parent company of TikTok) and its global expansion, shows how Chinese companies are riding the wave of China’s national strategy in developing AI to achieve the state’s goal of being an AI superpower in a decade.

From data journalism to artificial intelligence: A practical application of computer vision in news reporting
Mathias-Felipe de-Lima-Santos

School of Communication, University of Navarra, Spain

 Journalism is at a radical point of change that demands organizations to come up with new ideas and new formats for news reporting. In addition, the surge of data, sensors, and technological advances, notably in the mobile segment, brought immeasurable benefits to many fields of journalistic practice, data journalism in particular. In particular, the political watchdog reporting has been deeply evolved by the new scene that has emerged with data journalism (Carson and Farhall 2018). Thus, the datafication of society and the improved access to data and metadata bolster transparency goals of the news media industry, which are fundamental to assure the democratic functions of the society (Ritter 2014; Knight 2015; Lewis and Usher 2013). As technologies evolved, data journalists are turning to artificial intelligence (AI) technology to analyze these massive findings. 

Given the relative newness and complexity of implementing artificial intelligence in journalism, few areas managed to deploy tailored AI solutions in the media industry. In this sense, the practice has been a very Western-centric practice, engaging mostly Westerners actors in the data analysis and stories (Felle 2016; Stalph 2018; Lewis and Westlund 2015; Anderson 2019; Appelgren and Nygren 2014; Borges-Rey 2016; Coddington 2015; De Maeyer et al. 2015; Karlsen and Stavelin 2014). Few cases in the Global South have achieved great success, the Argentine newspaper La Nación is one of them. 

 Founded in 2011, La Nación Data has experimented with open data, civic journalism, and automated journalism since its foundation. Despite the lack of statistical data that affects most of the countries in the Global South, the data department innovated by engaging the audience in the news production. La Nación Data has an international reputation as well as internal recognition by innovative efforts that have led the news outlet to an award-winning position (Palomo, Teruel, and Blanco-Castilla 2019). Envisioning the future of journalism, the team is always seeking new ways to bring new forms and formats for the audience. 

Attempting to burst this bubble of incremental innovation, La Nación Data analyzed ways to use artificial intelligence (AI) techniques in journalism production. AI uses algorithms to enable machines to learn from experience, adjust to new inputs and perform human-like tasks. Amongst the several fields that AI is divided, computer vision (CV) is recognized by analyzing digital image processing. Contrary to what the name suggests, CV models cannot actually see the content of an image like a human eye (Marr 1982). Instead, they make use of mathematical algorithms to deduce what content is shown (Szeliski 2011). Using the computer to quickly classify and organize a plethora of images and videos, CV can accelerate the editing process and “enable journalists to source evidence for investigative pieces” (Marconi, Siegman, and Machine Journalist 2017). With that idea in mind, the La Nación Data team designed a project to map solar panels in Argentina, which later turned into a project to map the solar farms that were in full expansion in the country. 

Drawing upon literature on management, journalism, and computer science, I explore the hurdles inherent to executing a computer vision project that is in a rapid evolution in the machine learning space. In this study, through a mixed-method approach that combines both participant observation and interviews, I explain the hurdles and obstacles of deploying computer vision news projects, a subset of AI, in a leading Latin American news organization, the Argentine newspaper La Nación. Results highlight four broad difficulties to implementing AI, more specifically computer vision projects that involve the usage of satellite imagery: the lack of high-resolution imagery, the unavailability of technological infrastructure, the absence of qualified personnel to develop such codes, and, the lengthy and costly implementation process that requires significant investment. Thus, this study contributes to the scholarly work of journalism by shedding light on the emerging challenges and threats to deploy cutting-edge technological projects, mainly in the specific groups of Global South that already address other issues related to information technology. The study concludes with a discussion of the centrality of AI solutions in the hands of big tech corporations. 

Czech News Agency as a pioneer in the field of automated journalism

Radka Marková

Czech News Agency

Václav Moravec, Veronika Macková

Faculty of Social Sciences, Charles University, Prague

For the first time, the Czech News Agency automatically generated news articles along with its standard news service during the local and Senate elections in 2018. Thus, the national news agency became a pioneer in robotic journalism among the Czech media (Moravec et al., 2020). In 2020 Czech News Agency started using automatically generated texts in its weekly news reports on petrol prices and monthly statistics of traffic accidents. It used again as a tool for the election results (regional and Senate election). The main benefit of automation is an acceleration of the news service, limiting human errors and facilitating reporters’ work, thanks to which they could more focus on reactions and other exciting items from the elections. Newsrooms around the world are currently making progress in digital technology to help journalists with their work. There are areas of journalism that cannot yet automate and where a human journalist is still needed (Carlson, 2014).

The following main research questions were formulated:
RQ1: How has the Czech News Agency newsroom changed with artificial intelligence journalism?
RQ2: Should journalists worry about their jobs in the future?

In the qualitative content analysis, we focused on in-depth semi-structured interviews with twelve respondents and participatory observation in the Czech News Agency newsroom. Automated journalism speeds up work, makes no mistakes, but the role of editors is still irreplaceable. However, newsrooms will change as automatically generated texts increase, and editors and IT professionals will play a significant role.

Joining the bandwagon: Artificial Intelligence and the reconfiguration of journalism practises in Zimbabwe

Bhekinkosi Jakobe Ncube, Lungile Tshuma
Department of Journalism, Film, and Television at the University of Johannesburg
This proposed paper seeks to examine and interrogate the growing significance, use, and adoption of algorithms, and artificial intelligence (AI) in the global South using selected newsrooms in Zimbabwe as case studies. This is against the realization that algorithms and artificial intelligence as examples of computational journalism are shaping public life and therefore there is a need to examine how artificial intelligence (AI) can be applied to contexts of human communication. The proposed paper is also against the background that artificial intelligence has the potential to transform every aspect of people’s lives and consequently, the paper seeks to investigate how AI is used by selected newsrooms in the country in gathering, packaging, and disseminating information. Focusing on the growing intersection of journalism and computation, that is, some of the forms of computational journalism such as artificial intelligence, the paper also explores the dynamics or new newsroom cultures that are emerging because of AI, and how they use of AI in newsrooms challenges or negates the concept of traditional journalism practice. The assumption here is that the advent of computational journalism forms such as artificial intelligence is transforming newsrooms. Using the sociology of news concept and deploying in-depth interviews and semi-structured interviews with journalists, online editors and editors of online sites, daily and weekly
newspapers from both private and state-controlled media in Zimbabwe, this study makes three main arguments related to the transformation of newsrooms occasioned by the use and adoption of AI in the global South. First, we argue that while the adaptation and adaption of AI are at its infant stage, its current use has changed the way in which journalists conduct their duties. Through AI, journalists’ newsgathering processes have been improved and made easier, as they can quickly identity breaking news, trending issues and also become aware of their audience choices. Second, closely related to the above, we argue that there have been shifts of power, roles, and responsibilities regarding new gatekeeper positions because of AI in the journalism production process. Thirdly, we are of the view that one of the key ethical issues, objectivity might be compromised.
AI in the newsroom: Defining journalism as an outsider
Stefanie Sirén-Heikel, Martin Kjellman

University of Helsinki

Carl-Gustav Lindén

University of Bergen

Newsrooms are increasingly looking to artificial intelligence as a driver of value––streamlining workflows, personalisation, story leads, and generating news stories––as examples of a continuously evolving tool kit. As these AI technologies are becoming increasingly relevant for newswork, the esoteric logics of journalism are pulled apart, classified, and converted into parameters applicable for quantified decision-making. In this process technologists from companies producing AI solutions are faced with questions such as how to define news, how to define themselves, and how to define their product. How these external actors situate themselves in the field of journalism, in particular in relation to producing journalistic content, is understudied. How these actors influence journalism and newswork is a subject that warrants attention: conversely, how does newswork and journalism influence the actors designing these AI systems?

Our study aims to add to the field of work examining AI in journalism. We focus on companies that provide solutions for automated news, texts based on structured data using natural language generation (NLG). Our material consists of expert interviews with interlocutors from nine companies, based in Europe and the US, that provide solutions for news automation. By asking our interlocutors to recount the implementation of their solutions in a newsroom we gain insights into how the implementation process unfolds. To understand this process, we apply institutional logics as a theoretical framework, gleaning information on what logics are at play when technologists and news organisations interact. Institutional logics is a useful framework for understanding how actors in separate institutional domains construct their organising principles, values, practices and beliefs, influencing identity and sensemaking (Belair-Gagnon et al., 2020; Thornton et al., 2012). By examining how our interlocutors talk about their product, journalism,and the interplay between the two, we can identify the building blocks that underpin the logics defining how these systems are designed and what their role is perceived to be.

Our preliminary findings indicate a demarcation between the logics of technologist and journalism: journalism is a separate domain, under increased financial pressure, and in need of transformation in order to survive. AI solutions are seen as part of this rational transformation, enabling news organisations to scale their output and move newsworkers from “menial” reporting towards “meaningful” reporting. However, at the same time, through the interaction with newswork, technologists recognise and assimilate practices relating to journalistic logics, such as accountability, validity and ethics. To a certain extent this is intentional, as news is seen as benchmarking their solutions, raising the bar for the quality of their products.

Operating in an increasingly hybrid domain, newswork exists in an intersection of several competing and overlapping logics (Lischka, 2020). As analytics has moved from marketing to the newsroom, business logics has blended into the logics of journalism. With the advent of AI tools for journalism, the logics of the technologist are added to the mix. Our study hopes to elaborate the understanding of how this process unfolds, exposing the logics of the technologist and exploring how they see their impact on newswork.

AI skills for a new generation of journalists in Western Balkan countries
Oliana Sula

University “Aleksander Moisiu” Durrës, Durrës, Albania

Artificial Intelligence (AI) is transforming the traditional journalism. One of the most important development has been the adaption and the growing awareness of AI in the newsrooms worldwide. As well of the past years different developments show the importance of AI in journalism such as the machine written articles, the role of AI tools in assisting journalists, AI has been as well used in curating content that users consume in different social media platforms creating a personalized news consumption experience. Recent studies show that there are challenges regarding the adoption of AI journalism worldwide that are connected with financial resources, cultural resistance, lack of strategic managerial insight and knowledge and skills. The “job” of the journalist is evolving as well requiring new skills related to AI. Even if the general perception is that AI is a threat to substitute the traditional job of the journalist, there is a need to rethink how AI will increase and reshape especially the tasks of future reporters and of future editors. In Western Balkans countries there are several gaps in the fields of AI and journalism, firstly there is a gap in the AI skills of journalists and on the other hand, there is a lack of awareness of organizations in matching skills and the necessities of AI equipment. This research explores the awareness about AI skills of journalism students in Albania that are supposed to be future generations of journalists. This study employs qualitative methods based on two focus groups realized with journalism students. The students were asked to explore some of the benefits and challenges of AI for journalism in the Albanian context together with the AI skills needed for the “job” of future generation journalists. The first group of skills is related to IT skills which are technical skills, data analytics, editing automated content, and sourcing and managing user generate content. This study concludes with some ethical considerations on AI skills and journalism.

Current state and future potential of local AI: An investigation of how local newspapers adopt AI along the news value chain

Bartosz Wilczek, Neil Thurman

LMU Munich

Natali Helberger

University of Amsterdam

Journalism has entered the age of Artificial Intelligence (AI) (Diakopoulos, 2019). However, previous research has focused on how national news media use AI. The use of AI by local news media, which are under particular pressure (Wahl-Jorgensen, 2019), has garnered little attention. Moreover, previous research has devoted limited attention to the conditions that might shape how the news media use AI. Therefore, this multi-case study (Eisenhardt, 1989) draws on the Theory of Diffusion of Innovations (Rogers, 2003) and explores how local newspapers in Germany are adopting AI for input, throughput, and output activities along the news value chain and what internal and external factors might facilitate the adoption of AI. The study is conducted in cooperation with the German School of Journalism.

The Theory of Diffusion of Innovations (Rogers, 2003) explains how internal and external factors shape the adoption of innovations by individual and collective actors. Based on this theory, we developed an initial framework that we validate and extend based on the findings of the exploratory study. Figure 1 presents the framework and related research questions.

This study applies an exploratory multi-case approach (Eisenhardt, 1989). 14 leading local newspapers were chosen based on the German online traffic database (IVW). In all the local newspapers, we interviewed Heads of Online Journalism (i.e., Online Editor-in-chief, Head of Digital) (N = 14). In 9 local newspapers, we were also able to interview Heads of Technology (i.e., Head of Data Analytics, Head of IT). The in-depth interviews took place between January and February 2021 (via Zoom).

Local newspapers are incrementally building up their repertoire of AI applications along the news value chain (RQ1). Regarding input activities, local newspapers started by adopting AI for issue monitoring and are now moving towards adopting AI for research and verification. Regarding throughput activities, they started by adopting AI for article production and are now moving towards adopting AI for content analysis and modification. However, output activities are the core growth area of AI adoption in local newspapers. After all, AI for engagement analysis and recommender systems is particularly closely related to local newspapers’ financial sustainability – e.g., through online subscriptions, which will become increasingly important in the future.

Moreover, local newspapers have scarce resources for AI innovation. Consequently, they are interested in support for AI innovation (RQ4). In terms of financial resources, they expect support particularly from the private sector (i.e., platforms), which provides available financial short-term solutions. Accordingly, local newspapers’ financial dependence on platforms might increase (Fanta & Dachwitz, 2020). In fact, through their funding schemes, platforms might not only influence if local newspapers will be able to (further) adopt AI but also how, i.e., whether AI will also follow ethical standards.