Abstract
Generative AI (GenAI) is rapidly establishing itself as a key topic in societal, political, and scholarly debates. From a news and information perspective, relatively little attention has been devoted to how GenAI is framed in reporting by influential news outlets, with most coming from globally recognized English-language titles. This study operationalizes a framing analysis of 516 articles published by a quality newspaper in Belgium between 2020 and 2024. Findings reveal that the launch of ChatGPT in late 2022 signified a notable shift in the quantity and ways in which GenAI was approached and framed, extending the scope beyond technological stories to focus more on regulatory and societal ramifications. We establish a typology of 12 different positive, neutral, and negative frames in how GenAI is positioned and discussed in reporting. Combined, they highlight the complex nature and practice of reporting on an emerging technology that is taking the world by storm, while also retaining traditional journalistic values and tenets intact. We extend the body of research on AI news framing beyond the English language and to quality news outlets specifically.
Keywords
Introduction
Generative AI (GenAI), previously defined as “a branch of [Artificial Intelligence] that can create new content such as texts, images, or audio that increasingly often cannot be distinguished anymore from human craftsmanship” (Feuerriegel et al., 2024, p. 123), is rapidly transforming major parts of societies, ranging from medicine (Fahrner et al., 2025) to education (Xia et al., 2024) and general human creativity and innovation processes (Sedkaoui & Benaichouba, 2024). Scholarship on its far-reaching ramifications on a host of different sectors is mushrooming fast. This also applies to the field of journalism, as GenAI potentially jeopardizes its highly institutionalized professional values and societal importance (Shi & Sun, 2024). Various scoping reviews (Ioscote et al., 2024; Schjøtt, 2025; Sonni et al., 2024) have discussed how newsrooms around the world experiment and subsequently implement AI tools into their daily work processes to automate writing and data analysis, whereas journalism researchers themselves increasingly embrace computational methods and multi-method study designs. Other empirical contributions have focused on the vast differences in news organizations as well as individual journalists’ and citizens’ attitudes toward GenAI used for the gathering, production, and presentation of news (Londoño-Proaño & Buele, 2025; Morosoli et al., 2025; Thomson et al., 2025).
Less scholarly attention thus far has been devoted to how journalists approach (generative) AI in their professional writing. This is a notable gap in scholarship as a better understanding of how journalists position GenAI can help us in better understanding how it is perceived by journalism's audience, ideally critically informed citizens in democratic societies who are perceptive of emerging technologies altering their lives and influenced by how they are presented to them (Brossard, 2013). Building on framing theory (Entman, 1993) and the narrow body of scholarly work that have already applied this to news coverage on AI (i.e., Chuan, 2023; Cools & Diakopoulos, 2024), this article presents a framing analysis of 516 news articles published in the Dutch-language Belgian quality newspaper De Standaard between May 2020 and April 2024. The time period distinguishes the period before and after the public launch of GenAI-powered chatbot ChatGPT's first version on November 30, 2022 as this signified a clear turning point in public awareness and perceptions toward (generative) AI (Xian et al., 2024). The framing analysis of this vast body of articles culminates in a classification of 12 types of journalistic framing toward GenAI, further differentiating negative, neutral, and positive frames. Hence, this study contributes to expanding the scholarly knowledge on how journalists position GenAI and communicate about it toward audiences in a non-English language market.
Literature review
The rise of GenAI in journalism research and practice
GenAI has rapidly emerged as one of the leading topics of interest for media researchers. This makes sense considering the fact that tools such as ChatGPT, Gemini, and CoPilot intensely rely on other pieces of (mainstream) media such as text documents, images, and videos for the creation of their own output. A notable research focus has been AI's crossmediality features, for instance its ability to convert text inputs to images using generative models (Vartiainen & Tedre, 2023). The ease with which this has become possible for billions of citizens around the world has led to the rise of so-called “AI slop,” or low quality media forms in text or audiovisual formats produced automatically at high volumes, yet generally with low values, that is, as noted for the creation of music and YouTube videos (Anderson & Niu, 2025; Roddy & Bridges, 2026).
The lack of quality or other redeeming features that AI-generated media outputs are frequently criticized for finds itself at odds with the norms, values, and expectations of the journalistic profession. In recent years, it has had to juggle losses in traditional advertising revenue often not successfully recuperated by the move to publishing and paying for news content online (Fletcher & Nielsen, 2017; Joris et al., 2025). The growing popularity of audiovisual-centric social media platforms such as Instagram and TikTok has caused fewer direct online user traffic to news platforms’ proprietary platforms (Newman et al., 2025; Wojcieszak et al., 2021) and to more novel types of journalism that is created, distributed, and consumed within said platforms directly (Hendrickx, 2021; Wirz & Zai, 2025). There, news publishers have to compete for users’ attention with all other forms of content available, which increasingly features that generated by AI. At the same time, news content is rarely actively recommended by social media platforms’ illustrious blackbox algorithms (Hagar & Diakopoulos, 2023), which in turn are also increasingly supported and thus influenced by AI (Bareither & Wirth, 2025).
Various studies from the past few decades have highlighted journalists’ general resistance toward changes to their typical work routines, be that brought on by digitization (Courtois et al., 2010; Robinson & Metzler, 2016; Ryfe, 2009), newsroom integration and synergies (García-Avilés et al., 2014; Hendrickx & Picone, 2020; Sehl et al., 2019) or, indeed, the growing importance of AI in newsrooms. Both at the meso level of media organizations and that of the micro level of individual journalists, multiple recent academic contributions have investigated this issue. For instance, de-Lima-Santos et al. (2024) analyzed 37 media-related AI guidelines from 17 different countries worldwide and find that there is still a need for guidelines on how to further develop AI more responsibly in a news and journalism context where human involvement remains prioritized in order to safeguard journalistic values and protect privacy.
In their interview-based study of Danish and Dutch journalists, Cools and Diakopoulos (2024) highlighted that the journalistic intuition or gut feeling played a pivotal role in shaping journalists’ perceptions toward GenAI in their work. Interviewees highlighted distinct threats: GenAI tools risk oversimplifying complex news issues, tend to hallucinate and confirm biases, which respondents linked to “the risk that the spread of inaccurate information can lead to reputational damage” (p. 12). More profoundly, respondents also expressed their fear of losing their own journalistic autonomy, which “can lead to homogenized content that lacks the individual creativity of writers” (p. 12). However, the same study also yielded the finding that journalists believe incorporating GenAI in their everyday work can make it more efficient and “can reduce time spent on routine tasks” (p. 12), while also offering opportunities for easier personalization and tailoring of content to specific target audiences and user preferences. Focusing on seven different Western nations, Thomson et al. (2025) interviewed 20 photo editors from 16 different publicly and privately owned media corporations about their own usage and perceptions of GenAI. Their overview of challenges is in many ways similar to that of Cools and Diakopoulos (2024). Here, respondents mainly mentioned the risk of mis- and disinformation, displacement of journalistic labor, copyright issues, difficulties to detect AI-generated visuals, and, again, risks to the reputation of their own news outlet and/or journalism as a profession coupled with biases brought on by algorithms (pp. 10–13).
Through a qualitative literature review of 61 articles published in media studies journals between 2010 and 2023, Schjøtt (2025) identified six recurring themes in emerging scholarship on AI. These include news’ ethical and political contexts as well as its economic conditions but also institutional and organizational logics and contexts of AI implementation through bourdieusian doxa and capital. Cultural and technological vantage points were also discerned in the literature analysis. Based on this, the author proposes to shift academic attention toward AI's material dimensions and epistemology and to focus less exhaustively on newsroom-centric works (p. 23). A clear example of the latter point is the study by Morosoli et al. (2025), which polled 1484 Dutch citizens’ attitudes and resistance toward the use of GenAI in contemporary journalism. Specifically zooming in on the trust and credibility changes GenAI brings to the journalistic profession, the study finds that citizens who see benefits toward the implementation of AI in journalism tend to display higher levels of trust both in AI-generated content and in the journalists who produced it. Inversely, lower trust in citizens’ own AI usage is linked to their refusal to engage with AI-generated content, including journalism. The authors rightfully stress the need for news publishers to be transparent at all times about where, how, and why GenAI to enhance citizens’ trust in news and journalism.
(Generative)AI and Its representation in news coverage
This study contributes a news content analysis on how GenAI is framed in one quality newspaper's coverage over a 4-year time span. It is one of the first studies of its kind to do so for GenAI specifically. That said, we draw on the rather narrow body of work that has previously assessed how artificial intelligence in general has been framed in news reporting. Perhaps rather unsurprisingly, a structured literature review on this topic revealed that most articles were published after 2010 and that they mostly focused on American and other leading English-language news outlets’ coverage (Chuan, 2023).
Collaborating with colleagues, the same author had previously carried out their own framing analysis and colleagues on how AI was approached in 399 articles published in key US newspapers (Los Angeles Times, New York Post, New York Times, USA Today, Washington Post) between 2009 and 2018 (Chuan et al., 2019). Business and economy were the most dominant topic in AI coverage during this time frame, with science and technology second. Ethics and moral issues emerged as an important topic throughout time, with both positive and negative valence depending on the article. Delving deeper into the New York Times specifically and focusing on articles published between 1986 and 2016, Fast and Horvitz (2017) “find that discussion of AI has increased sharply since 2009 and has been consistently more optimistic than pessimistic,” also noting that “the fear of loss of control of AI [has] been increasing in recent years” (p. 968). Another noteworthy contribution is that of Cools and Diakopoulos (2024) who use topic modeling and manual analysis to study 10,038 articles (one tenth of that for the manual component) published in the Washington Post and (again) the New York Times on AI and automation, here between 1985 and 2020. They find evidence for different positive and negative frames: The frames ‘Helping hand’, ‘Shortcoming’, and ‘Conflict’ were most prominent throughout the corpus. In the topics ‘Work’, ‘Health’ and ‘Sport’, the frame ‘Helping hand’ is prominent, which indicates a more positive approach to AI and automation. Within the topic of ‘Politics’, more dystopian frames were found, with ‘Conflict’ and ‘Frankenstein monster’ as the most prominent ones, for instance when attention is paid to the use of algorithms controlled by smart technology. (p. 18)
Finally, in what is to the best of our knowledge the only other scholarly work published at the time of writing that also investigates how GenAI specifically has been framed in news reporting, Xian et al. (2024) too combine topic modeling with qualitative analysis to assess a vast body of English-language news articles (N = 24,827). Amongst other things, the authors’ sentiment analysis yields the findings that more positive sentiments occur for business, corporate, and technology stories, yet more neutral to negative sentiments for articles focusing on GenAI's regulation and security.
To summarize, the limited body of similar works has been predominantly focused on English-language news outlets and their coverage of (Gen)AI. It has found evidence for positive, negative, and neutral frames in common usage as well as for their evolution over time as AI became more commonplace in societal and political debates. The launch of ChatGPT in late 2022 has been a key turning point in how said debates have been evolving in the past few years (Xian et al., 2024). Injecting these findings into our own study design outlined below, we first present the two guiding research questions:
What frames are used to discuss GenAI in news coverage of De Standaard between 2020 and 2024? How do the frames used to discuss GenAI in news coverage of De Standaard evolve over time between 2020 and 2024?
Data and method
Data set
To overcome the strong English-language bias in AI framing analyses, this study focuses on the news coverage of De Standaard (“The Standard” in English; henceforth abbreviated to DS), the biggest daily quality newspaper that has been published in Belgium's Dutch-speaking region Flanders since 1919. We focused on a case study of one quality newspaper due to its higher ascribed level of interventionism compared to tabloid outlets; as Bartholomé et al. (2018, p. 1693) argued, quality news titles “may want to give readers more than just the news and adapt an active approach, evaluating politicians and provide readers with interpretations and backgrounds.” This decision also helps us to keep the data set manageable and comparable by rather opting for a larger time span to better gauge differences in how GenAI is approached in reporting over time. DS is owned and published by the international media corporation Mediahuis and has previously been the topic of international academic scrutiny, for instance in its coverage of global journalism (Van Leuven & Berglez, 2016) and its contribution to news content diversity (Hendrickx & Ranaivoson, 2019).
All articles published in DS between May 1, 2020 and April 30, 2024 were manually collected through Belga.press, an online repository of Belgian print and online newspaper articles accessible to journalists and researchers. This wide timeframe allows for a deeper and more nuanced analysis of how one reputable newspaper changed its perceptions and attitudes toward GenAI in its coverage. In line with Xian et al. (2024), we pinpoint ChatGPT's initial public release on November 30, 2022 as a turning point and benchmark in our analysis, comparing articles that were published before and after this time together but also separately to better establish changes and evolutions in frame usage. We followed Cools and Diakopoulos (2024) and discarded all articles with less than 500 characters including spaces as this allowed us to focus on longer articles that did not just offer short updates but rather more analytical insights and perspectives on GenAI, making it more interesting for our study design and frame retrieval. We only incorporated articles that were published in the print newspaper and/or on the news website of DS. Podcasts, social media posts, newsletters, and other forms of media were not considered as this would needlessly convolute our methodological framework. While we limit ourselves to text-based media, we do incorporate news articles along with analyses, opinion pieces, and editorials. Despite their differing journalistic purposes and aims, they all aid in painting a clearer picture in how GenAI is framed in DS’ reporting. This collection approach resulted in a total data set of 516 articles, which we make publicly available in this repository.
Framing analysis
In a first step, 48 articles of the total data set (the first one published during each month spanning our time frame of 4 years or 48 months) were selected for initial analysis. Close reading of all these articles culminated in an inductive typology of 12 different frames. Articles were frequently reread, coded, and annotated in an Excel file to discern recurring phrasings, concepts, and themes. The following variables were distinguished: article title, number of article characters (including spaces), year and month of publication, the article's general theme, and societal domain (inspired by Chuan et al., 2019), how AI is represented in the article and its applicable frame. These frames were based on the first analysis of 48 articles. The approach of selecting one article per month allowed for a better contextualization and visualization of changes in frame use over time.
Once this section of the analysis had been successfully applied and tested and the frames had been decided upon, the same procedure of close reading, annotating, and coding was applied to the remainder of the 516 articles comprising the total data set. This time, all articles were assigned one of the predetermined frames. Particular attention was devoted to differences in framing per societal domain. We were interested to see if political, economical, and technological coverage would approach and describe GenAI differently, especially as ChatGPT was ascribed progressively more societal applications. We look at how the different frames appeared more or less in specific domains and themes of news reporting, including highlighting articles that were found to not adhere to a specific frame in its coverage style. The results section will feature visuals of our findings that were created with the assistance of Julius AI, an advanced tool that uses machine learning to recognize patterns in datasets and generates visualizations that facilitate the interpretation of our data. We stress that this is the only instance of AI being leveraged to contribute to components of this manuscript.
The second author carried out the analyses under close supervision of the first author, who has prior experience with framing studies (Pakvis & Hendrickx, 2025). Robustness was guaranteed through multiple discussions between both authors and a third experienced scholar, offering overviews of and feedback on the chosen data method and how it was effectuated in this study. We opted for a fully manual method as a purely automated analysis would have not made it possible for us to explore nuances and underlying narratives in language use as well as more subtle differences in coverage over time (Karlsson & Sjøvaag, 2016). The results section will present article headlines and snippets, which are all personal translations to English from their Dutch-language originals by the authors.
Results
Figure 1 shows the overview of DS articles in our data set per month. We denote a spike in the first half of 2021, when the newspaper published a series of AI-related articles with titles such as “Can an algorithm decide about people?,” “Robots don’t wink,” and “GPT-3 writes a new Harry Potter chapter.” However, a clear trend emerges after the launch of ChatGPT, with the launch date marked vertically in green and all articles published since in red. Of the 516 articles in our data set, 421 or 81.6% were published after November 2022. March 2023 was the “busiest” month with 43 different articles, or more than one per day. This confirms that ChatGPT brought GenAI more into the mainstream by means of regular coverage in quality news media such as DS.
Trend of articles over time, highlighting ChatGPT era.
Table 1 displays all retrieved frames across the total data set of 516 articles following the initial analysis of a subset of 48 articles. Exactly a quarter of all articles does not have one discernible frame. Rather, they were found to be factual articles offering insight and analysis without obvious sentiments and frames. Examples include articles like “ChatGPT will summarize news articles from ‘Financial Times,’” “Pakistani former prime minister holds victory speech with artificial intelligence” and “Google launches AI model Gemini that ‘understands the world like we do.’” We denote three neutral, four outspokenly negative, and five outspokenly positive frames; combined they account for respectively 13.8%, 29.3%, and 32.0% of the total data set. Interestingly, the two most recurring frames are both positive (“AI as a handy tool” and “AI can be regulated”). This denotes a diverse and complex approach of AI in DS’ coverage, highlighting its advantages and risks without one narrative dominating.
Frames in Data Set and Frequency.
ChatGPT's launch not only led to a higher volume of articles published, they also became more diverse in terms of themes and domains. For the 2020–2022 period, marked in blue in Figure 1, news mainly discusses technological developments (with such headlines as “Will you be smarter than your smartphone?” and “Eye scan can predict risk for heart attack”). From 2023 onwards we denote a broadening in scope and topics, also investigating GenAI's social, economic, and cultural dimensions.
Contrary to our own expectations, Figure 2 shows that the ratio of neutral, positive, and negative sentiments in articles remained roughly the same throughout the entire time period. We do gradually find a decrease in neutral frames to a more balanced view, particularly between 2021 and 2022. But the distinctions between positively and negatively loaded articles in particular has remained surprisingly stable. This again signifies that a quality newspaper such as DS uses its journalistic responsibility to actively strive for balanced and diverse reporting in its coverage to give its readers a nuanced image on the topic. As society and corporations emerged as the most recurring domains of interest for news articles throughout 2023 and onwards, followed by culture and politics, the sentiment dimensions remained surprisingly stable.
Sentiment distribution by year.
The remainder of this section goes into more detail about all 12 retrieved frames, in the order of frequency as retrieved in the data set and presented in Table 1. For most frames, a text excerpt from the data set is shared to further elucidate their meaning and operationalization in reporting.
AI as a handy tool
Sentiment: positive
Appearances in data set: 62 (12.0%)
Most recurring domains: culture, society, leisure
This frame focuses on AI as a useful aid that can contribute to innovation and productivity and can make the world better. AI is praised for increasing efficiency and solving problems that can make everyday life easier. The city [of Antwerp] wants to explore possibilities to detect and recognise vehicles’ type and age. This way, cars that are still following the rules can be automatically granted access [to the low-emission zone] without sending a fine first. That could take away a lot of frustration. (From the article headlined “
AI can be regulated
Sentiment: positive
Appearances in data set: 46 (8.9%)
Most recurring domains: politics, society
This frame focuses on positive steps toward regulation and (ethical) guidelines to keep AI under control. Ethics, lawmaking, and privacy are recurring themes. In those three years, the text—and the world—has changed. The lightning-fast developments in Generative AI since late 2022 (when ChatGPT was launched) necessitated the European institutions to adjust their course considerably. Among other things, this led to the European Parliament voting for an entirely different version of the AI Act than the European Commission and European Council. (From the article headlined “
AI threatens human identity
Sentiment: negative
Appearances in data set: 40 (7.8%)
Most recurring domains: media, society, culture
This frame focuses on AI threatening creativity, authorship, copyright, privacy, and ethics as cornerstones of human identity. Articles often discuss the impact of AI on the creative sector and its ability to undermine authenticity and originality, for instance through deepfake video and audio content. This further risks undermining trust in the truth ascribed to media content collectively with possibilities of abuse, identity theft, and spreading dis- and misinformation. Over 12% of Belgian youths admit knowing apps that generate fake nude pictures using artificial intelligence. Over a quarter of those have already tried doing it. Almost one out of four does so to take revenge. A new, worrying kind of sexual violence is on the rise. (From the article headlined “
AI magnifies (societal) problems
Sentiment: negative
Appearances in data set: 40 (7.8%)
Most recurring domain: society
This frame focuses on AI's ability to exacerbate societal issues such as injustice, privacy, and discrimination. As AI is trained on human materials, this means it can also learn misogynistic, racist, or other inflammatory information and consider and spread it as truth. AI is also used as a war weapon and a contributor to the climate crisis, next to potentially harming people's physical and/or mental health. A Belgian man, a father of a young family, has ended his life after long conversations with a chatbot (…) DS tested the same technology and ascertained that it can prompt suicide. (From the article headlined “
AI as a battle field
Sentiment: neutral
Appearances in data set: 38 (7.4%)
Most recurring domain: companies
This frame focuses on AI as an important weapon in the tech war, with big corporations attempting to retain their positions. Corporate fights, competition, and the question who will emerge as the most dominant player are recurring. After Elon Musk's complaint against OpenAI—on the basis that it allegedly betrayed its own non-profit principles—the company fights back by publishing old emails from the Tesla CEO. They reveal that Musk too believed OpeAI needed extra funds. (From the article headlined “
AI is unstoppable
Sentiment: negative
Appearances in data set: 37 (7.2%)
Most recurring domain: society
This frame focuses on the potentially dangerous, irreversible changes spurred by AI, especially considering the lack of effective regulation. This frame is abounded with dystopian and pessimistic future scenarios featuring new power structures prioritizing tech companies without no room for objective truths. The uncertainty and the scope of the damage that can be done if we build machines that are smarter than we are can easily be abused (…) with catastrophical consequences. (From the article headlined “
AI is strongly developed, but humans remain necessary
Sentiment: positive
Appearances in data set: 36 (7.0%)
Most recurring domains: society, media
This frame focuses on AI as a powerful technology that still needs human intelligence and creativity as it only makes predictions based on data that cannot generate its own ideas. Collaborations between man and machine and how they can lead to creativity and innovation are recurring themes. Working with AI in the newsroom will always be a balance exercise between what helps us and you as our reader and what contains too many risks that could damage journalistic core values such as trust, originality and independence. (…) In this debate, every DS editor has a voice, as a journalist but also as a human being. (From the article headlined “
AI pushes humans aside
Sentiment: negative
Appearances in data set: 24 (4.6%)
Most recurring domains: companies, society
This frame focuses on how AI threatens human labor and social justice, particularly when corporations start using AI for maximizing profits, evoking ethical concerns about governments’ and companies’ responsibilities toward employees. This frame echoes concerns about job losses and AI-powered systems making decisions based on predictions and patterns. Marxist terminology about an economic revolution and new production means upending the current ways of producing goods and services are recurring. History teaches us that revenues of technological revolutions rarely reach workers. (…) We know who gets the bulk of the profit: the same people who also benefited from the previous productivity revolution. (From the article headlined “
AI improves society
Sentiment: positive
Appearances in data set: 21 (4.1%)
Most recurring domain: society
This frame focuses on AI as a contributor to societal development, such as improved health care, education, welfare, innovation, and sustainability. This optimistic frame portrays AI as fundamentally better, improving citizenry's life quality. What to do with a variety of half used ingredients? We asked our assistant cook. When opening the tool you can choose to generate a recipe for a starter, main course or dessert. You can also let the tool determine what type of dish goes best with the leftovers you put in. (From the article headlined
AI is in its infancy
Sentiment: neutral
Appearances in data set: 17 (3.3%)
Most recurring domains: companies, society
This frame focuses on AI as a technology still in development with relatively little details known. Recurring themes are innovation, uncertainty about the future and experiments. This frame critically positions AI as not yet a threat, but heeding its possible impact. AI will replace jobs slower than expected, particularly because of the start-up costs. Fine-tuning the technology to make it perform tasks as good as humans is still very costly. (From the article headlined “
AI in Flanders/Belgium/Europe
Sentiment: neutral
Appearances in data set: 16 (3.1%)
Most recurring domains: politics, companies
This frame focuses on the development and impact of AI within a Flemish, Belgian, or European context. Recurring themes include local innovation, policy, and unique challenges and opportunities for AI in these specific geographic contexts. Often these include news stories about people or companies that take part in the AI revolution and help to put Flanders, Belgium or Europe in the public eye. Emmanuel Macron wants France and Europe to be the frontrunners of the new AI revolution. The French president is willing to spend millions of euros on it. This way, he wants to create two to three global players and to ensure that Europe doesn’t miss out on the AI train. (From the article headlined “
Are big tech corporations trustworthy?
Sentiment: negative
Appearances in data set: 10 (1.9%)
Most recurring domains: companies, society
This frame focuses on big tech corporations involved in AI and casts doubts about their ethics and trustworthiness. Recurring themes are corporate ethics, power, and transparency. This frame is critical toward AI companies and questions their underlying agendas and potential power in the technological revolution. There were concerns at Microsoft. Their ethical council warned that their own chatbot could lead to a current of disinformation on Facebook. That ethical council was shut down. (From the article headlined “
Discussion and conclusions
Citizens around the world increasingly embrace and incorporate GenAI in their daily lives, by using ChatGPT for recommendations of generating images and videos for publication purposes. The technology is marked by multimodality and multifunctionality (Ronge et al., 2025), with a lot of its impact still unknown and difficult to estimate at the time of writing. When it comes to the journalism industry specifically, scholarship has thus far only scratched the surface of how it is affected by GenAI, which has the power to firmly place journalism at an intriguing crossroads in the next few years to come. The first batch of emerging contributions have assessed the behavior, approaches, and tactics of news organizations and individual journalists toward integrating AI in their work routines (Cools & Diakopoulos, 2024; Thomson et al., 2025), highlighting the various opportunities and threats of doing so for journalistic labor, integrity, and trust. This is echoed by citizens’ thoughts and feelings about engaging with AI-assisted journalistic content (Morosoli et al., 2025). But from the vantage point that citizens are influenced by how news media as gatekeepers approach and frame new technologies such as GenAI (Brossard, 2013), limited academic attention has been devoted to how this occurs in practice. A few studies have assessed the reporting of leading English-language news outlets on this topic, often using longitudinal study designs that combine automated and manual coding (Chuan et al., 2019; Cools & Diakopoulos, 2024; Fast & Horvitz, 2017). In doing so, they fail to incorporate the recent accessibility wave of GenAI specifically and neglect news content that is published in languages other than English. To this end, this paper has proposed a framing analysis of 516 articles published by the Dutch-language Belgian quality newspaper De Standaard (DS) between 2020 and 2024, in line with Xian et al. (2024) using the launch of ChatGPT in late November 2022 as a benchmark in the temporal dimension of the analysis.
Our analysis too yielded an overview of 12 distinct frames recurring in coverage, with an almost even split between positive and negative frames (Table 1). This ratio remained surprisingly stable throughout the time period (Figure 2), even though we clearly denote a peak in articles published in the direct aftermath of ChatGPT's launch and onwards (Figure 1). This finding indicates that journalistic attention increased as GenAI became more readily accessible and easier to use, causing its relevance to spike. In turn, this fostered a variety of new angles to discuss ethical and societal dilemmas, with articles covering such topics as labor, privacy, copyright, and AI's influence on culture and creativity. We argue that the frames used to approach and discuss AI in DS evolved along with the rapidly changing context wherein it became more and more intertwined with citizens’ everyday lives. This is an interesting finding in line of previous research on this particular quality newspaper, which had laid bare that its newsroom workers were more resistant toward approaching digitalization and user metrics in their daily work flows compared to colleagues from other news outlets operating under the same media corporation (Hendrickx et al., 2021).
While we cannot claim that our findings are automatically transferable to other news outlets or regions, it is relevant that there are significant overlaps with the very narrow body of similar works. The plurality in frame sentiments and types as well as the increasing number of dimensions and themes over time that was established in previous studies (Chuan et al., 2019; Cools & Diakopoulos, 2024; Fast & Horvitz, 2017) also ring true to our study. Specifically, the rise of ChatGPT and other accessible GenAI tools will very likely continue to increase the societal, and thus journalistic, relevance of this technology. More framing analyses from other, particularly non-Western markets are needed to continue to make scholarly sound statements on how GenAI is approached and discussed in news coverage, particularly considering its influential role in setting societal and political debates. Our narrow focus of one quality newspaper from Belgium can be considered as a shortcoming to our study, but it rather invites other scholars to replicate our method to other types of news outlets from around the world. The case study we presented corroborated yet also extended previous findings by incorporating the advent of GenAI. Time will have to tell if it all turns out to be just a fad that will be over soon or truly the transformational power that will continue to alter citizens’ lives in the coming years, for better and potentially also for worse. At this time, we subscribe to the view of Londoño-Proaño and Buele (2025), who rightfully note that AI “has proven to be a key tool in the modernization of journalism” yet it “does not replace the essence of the journalist as a narrator, analyst and interpreter” (p. 6). Should we end up living in a world wherein GenAI and humans co-exist peacefully, it remains of paramount importance to continuously remind ourselves that a technology such as GenAI in its current form cannot surpass but rather only assist humans.
Footnotes
Acknowledgments
The authors wish to thank Prof. Astrid Vandendaele for her guidance and feedback.
Ethical statement
This research is based on news coverage on GenAI. Therefore, no human subjects were involved.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data availability
The authors’ data set of news articles is available in this repository.