Sage Journals: Discover world-class research

Abstract

As part of a broader look at the impact of generative AI, this study investigated the emotional responses of journalists to the release of ChatGPT at the time of its launch. By analyzing nearly 1 million Tweets from journalists at major US news outlets, we tracked changes in emotion, tone, and sentiment before and after the introduction of ChatGPT in November 2022. Using various computational and natural language processing techniques to measure emotional shifts in response to ChatGPT’s release, we found an increase in positive emotion and a more favorable tone post-launch, suggesting initial optimism toward AI’s potential. This research underscores the pivotal role of journalists as interpreters of technological innovation and disruption, highlighting how their emotional reactions may shape public narratives around emerging technologies. The study contributes to understanding the intersection of journalism, emotion, and AI, offering insights into the broader societal impact of generative AI tools.

Keywords

artificial intelligence AI emotion journalism ChatGPT Twitter (now X)

With the introduction of any new technology, particularly one that appears to be disruptive or transformative in nature, there is a tendency for people to emphasize one of two competing hypotheses about the technology and its impact: dystopian versus utopian (Nowland et al., 2018; Valkenburg & Peter, 2007). At the risk of oversimplifying, we might say that one camp highlights, often in terms of anxiety, fear, and worry, the risks and potentially negative consequences of the technology. Meanwhile, another camp focuses more on the perceived benefits for individuals and society, often emphasizing hope, optimism, awe, and amazement.

Following this pattern, the launch and then rapid mainstream adoption of chatbots like ChatGPT has engendered all manner of hope as well as hysteria, astonishment as well as anxiety—with various detractors and supporters alike emphasizing, on the one hand, the potentially deleterious outcomes of large language models (LLMs) for jobs, education, and notions of truth and reality, and, on the other, the expansive possibilities for creativity, education, economic productivity, and more (e.g., see Mollick, 2024; Vrabič Dežman, 2024). While not the only LLM chatbot available, ChatGPT quickly became the most visible and attention-grabbing after its public release in late 2022. In less than a week after its launch, ChatGPT reached 1 million users, and by January 2023 it had an estimated 100 million active users, making it one of the fastest-adopted consumer applications in history.

If artificial intelligence (AI) in the broad sense refers to the use of machines to perform tasks typically requiring human intelligence, often by learning from experience, parsing natural language, recognizing patterns, and solving problems (Boden, 2018; Broussard, 2018; Mitchell, 2019), then ChatGPT and tools like it represent a specialized subset of AI called generative AI, so named because they involve the generation of new content—such as text, images, audio, video, or code—at unprecedented speed and scale. LLMs such as Claude 3.5 Sonnet from Anthropic or GPT-4o from OpenAI, for example, are advanced generative AI technologies that have been trained on huge volumes of text data that then allow users—such as people prompting ChatGPT—to generate human-like text on command. These developments in generative AI, in particular, have come to be seen as revolutionary, whether for good or ill, across many industries and domains. Many business executives talk about the technology’s potential to transform hiring (Kelly, 2023) and customer service (Das et al., 2023), even as the same tools and capabilities have elicited fears of replacement among workers, with some estimates of one in five jobs (or more) being affected by some form of displacement in the near term (Hatzius et al., 2023).

Ultimately, the actual impact of generative AI on jobs, industries, and daily life broadly may depend in large part on the stories that people tell themselves about disruptive technology, and such stories may reflect the tone—pro or con, hype or hysteria—that may be set by elite influencers in society (cf. Yin, 1999). Against this backdrop, this paper asks: Who ultimately shapes the formation and perpetuation of these narratives—and, therefore, why do they matter? We address this question by investigating how tastemakers like journalists responded to the mainstreaming of generative AI as indicated by emotion and tone expressed on social media.

Journalists offer a particularly useful case for several reasons. For many decades, research has demonstrated the power of the news media to set the agenda for public discussion (McCombs & Valenzuela, 2020), though journalists can be influential not only in drawing attention to certain topics over others but also in how they emphasize various qualities and characteristics of the people, issues, and ideas being discussed (Entman et al., 2009). Indeed, well-established is the distinct role of journalists as “sense-makers” for society (Singer, 1998). We see this in the way journalists help shape the terms of debate, establish narratives that persist over time (e.g., the so-called “first draft of history”), and offer framing devices that altogether matter for influencing public opinion formation (Lecheler & De Vreese, 2012)—not in an entirely deterministic way, but nevertheless in a way that has been shown to demonstrably matter (Entman et al., 2009). Even in the increasingly fragmented media landscape of the 21st century, in which news media must fight to claim a sufficient share of audience attention (Carlson et al., 2021), there remains substantial evidence that journalists still matter as tastemakers for society (Langer & Gruber, 2021; McCombs & Valenzuela, 2020; see also Schudson, 2018).

Importantly, the narratives constructed by journalists often function as “fictional expectations” (Beckert, 2016) used for decision-making: that is, the expected values or rewards can orient decision-making toward the attainment (or avoidance) of certain imagined futures. Thus, the prevailing narratives, to which journalists and the media at large serve as key contributors, not only impact current conceptions of these technologies but also influence ideas about and expectations from AI that are used as reference points in shaping future outcomes. These narratives, however, are not purely rational nor fully formed; rather, the emotional states of journalists themselves may serve as crucial elements in forming the imaginations that take shape and become collectively adopted in society. As such, the individual and collective emotions expressed by tastemakers like journalists—whether infused with optimism, skepticism, or alarm—contribute to the types of futures that are ultimately envisioned and conveyed through news narratives. Put another way, journalists are more than just reporters—they are instructors on how others should think and feel about the information being reported.

One way to tap into the imagined futures of journalists that has yet to be explored, particularly at any large scale, is to study the emotions of journalists. Emotion, or the discrete feeling states that represent reactions toward events (Barrett, 2017), can serve as a valuable marker for signaling our hopes and desires as well as our fears and anxieties—thereby offering a lens into how people feel about the future in a way that helps to render that future actualizable. Studying emotions, as customary in disciplines like communication science (Nabi, 2010) and psychology (Cacioppo & Gardner, 1999), thus provides deeper insights into the motivational forces driving journalistic perspectives and the narratives they create, which have consequences on public perception and discourse about the future.

This study examines markers of journalists’ emotions at scale, and it does so at a particularly apt moment: the periods before and after the introduction of ChatGPT. Our goal is to use this exploration of emotions to investigate the competing (and pervasive) hypotheses about the increasing role and influence of generative AI: Is the doom-and-gloom hypothesis of dystopian concern more prominent, perhaps because journalists fear how such tools, given their ability to interpret and generate language, might displace their work through automation? Or, is the optimistic hypothesis more prominent because journalists, as some evidence indicates thus far, see opportunities in generative AI to free them up by taking on mundane tasks that could be automated? In the past, emerging AI-based technologies have been met with skepticism from journalists. When news automation was introduced in 2013 by companies such as Narrative Science, journalists’ anxiety about potential job loss linked to “robot” journalism, the quality of automated news, and the implications of automation on journalistic norms became evident in stories they published (Carlson, 2015). Automation also sparked debates about journalistic authority, with one side recognizing the value of algorithms in producing evidence-based stories, and the other emphasizing high-quality news writing that can only be produced through human storytelling (Carlson, 2015). When deepfakes started circulating, journalists painted a dystopian future in their reports about such technologies and published speculative pieces about potential misuse by nefarious actors (Wahl-Jorgensen & Carlson, 2021).

In both instances, the news became the discursive field to construct future imaginaries about these technologies, positioning AI-based technologies as having negative implications in journalism. These cases also represent evolving temporal orientations of journalism, where discourses and speculations about the future, rather than the present facts, redefine what is considered “news” (Neiger, 2007). Indeed, contemporary journalism involves “conjectured futures” (Neiger, 2007), where worst-case scenarios are reported as potential, sometimes imminent, realities. While future orientations tend to vary by medium (Tenenboim-Weinblatt and Neiger, 2015), the emergence of platforms provides new venues for journalists to speculate about the future—blending professional and personal identities as they break the news in between personal commentaries.

An examination of the emotions that are evident in journalists’ expressions online—and especially of how such emotions are manifest in the case of journalists who are most preoccupied with these issues of ChatGPT and AI (see Perreault et al., forthcoming)—can therefore offer a window into how publics broadly may come to form narratives about the imagined futures associated with these technologies. In effect: If journalists set the tone for how people may come to imagine these technologies and what they expect from them, then what kind of tone, positive or negative, is being established? And what might that reveal about the role of emotions in making judgments about the imagined futures of new technologies?

Literature Review and Contextual Background

AI and Imagined Futures

Since November 30, 2022, when the research firm and tech company OpenAI launched ChatGPT to much public fanfare, social media conversations about AI have been dominated by talk of ChatGPT’s role and influence (Maslej et al., 2023), intensifying broader public discussion about AI generally and generative AI in particular. The potential loss of human jobs due to AI stands out as a top concern among many Americans (Maslej et al., 2023), and the implications of AI for job security have become more pronounced, particularly given the demonstrated efficiency of ChatGPT and other generative AI products. This narrative is also evident in the media, where numerous articles have been published discussing jobs that could be replaced by AI after the launch of ChatGPT (e.g., Mitchell, 2023; Stahl, 2023).

News accounts such as these constitute part of how societies imagine the future in relation to technological developments. These imaginations, born out of people’s expectations of technologies, inform current practices and systems, which are enacted to attain or avoid such futures (Jasanoff & Kim, 2009). In other words, expectations about the future mobilize changes within the present, and tastemakers such as journalists serve as intermediaries not only in shaping these expectations, but also in producing new ones, as they possess significant influence over the narratives surrounding new, disruptive technologies (Brennen et al., 2022).

Long before the introduction of ChatGPT, future imaginations around AI had been circulating in society, and they fall into two parallel narratives of hope and fear (Cave & Dihal, 2019). The promise of longevity and immortality is one of these hopeful narratives, with news media reproducing this imaginary of an AI expected to make significant contributions to health care and medicine (Cave & Dihal, 2019; Wang et al., 2023). AI is also expected to make living conditions easier (Hancock et al., 2020), as everyday routine tasks are automated, essentially making AI serve as our assistants (Cave & Dihal, 2019). In this vision of AI, as humans are freed from doing mundane activities, AI helps people live more meaningful lives (Hautala & Heino, 2023). In addition, AI is expected to be a tool that can be effectively harnessed to drive innovations, augment economic and social capacity of humans, provide more opportunities for employment (Hautala & Heino, 2023; Wang et al., 2023), promote efficient systems and sustainability, and help nations advance in governance and labor productivity (Bareis & Katzenbach, 2022).

Inherent in these hopeful imaginaries are instabilities that can lead to dystopian outcomes, represented as narratives of fear. Opposite longevity and immortality is the possibility of inhumanity or the risk of losing human identity. Meanwhile, the ease of work afforded through AI is met with fears of obsolescence from work and being replaced by technologies. The uprising of machines is also anticipated due to the technology (Cave & Dihal, 2019), viewing the possibility of AI taking over humanity and controlling humans “in a matrix-like world” (Hautala & Heino, 2023, p. 5). Others imagine AI as a technology that people will use to take advantage of others and widen social inequalities, with corporations owning AI development as well as data, and humans becoming unemployed and further marginalized (Hautala & Heino, 2023). The rise of AI is also viewed as the end of humanity, and this is reflected in a 22-word statement signed by many scientists, public figures, and industry leaders which states that “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war” (Statement on AI Risk, 2023).

In both hopeful and fearful narratives, expectations about AI—and media or technology in general—are fueled by affect and emotion, but it is more pronounced in the dystopian rather than the utopian views of the technology (Nowland et al., 2018). Inaccurate portrayals and anthropomorphization of AI, conflating human autonomy with computer autonomy, and socio-technical blindness, or the failure to recognize how technologies always work in tandem with humans and institutions, are believed to contribute to AI anxiety, or the “fear and trepidation being expressed about out-of-control AI” (Johnson & Verdicchio, 2017, p. 2267; for further reflection on human–machine dynamics, see Guzman & Lewis, 2020). The role of journalists as tastemakers is further highlighted as they make these anxieties evident through their outputs (Brennen et al., 2022), and contribute further to the apprehensions and fear narratives revolving around seemingly uncontrollable technologies (Sartori & Bocca, 2023). Understanding the role of emotions on public perceptions of AI, and the significant role that journalists play in shaping these emotional narratives, underscores the importance of studying the intersection of emotions and journalism in the context of AI.

The Import of Affect and Emotion: A Psychology of Language Perspective

Classic social scientific theories of emotion suggest that there are six dominant emotional expressions (sadness, happiness, fear, anger, surprise, disgust) (Ekman et al., 1987), which represent feelings we have to socially constructed events (Barrett, 2017). The “goodness” or “badness” of something, also known as affect (Slovic et al., 2004), is related to emotion, but it is an independent concept. For the purposes of our work, we use emotion to mean discrete emotions from Ekman, Barrett, and others, while affect is a broader understanding of the positive or negative attributes of an object that includes (but is not limited to) discrete emotions.

The scientific study of emotion and affect is over a century old (James, 1884), with keen interests in how they are constructed (Barrett, 2017), regulated (Gross, 1998), spread (Kramer et al., 2014), how they guide decision-making (Nabi, 2003; Peters et al., 2006), and also associate with a range of social and psychological processes (Barrett et al., 2007; Cacioppo & Gardner, 1999). In many cases, emotion and affect are studied as antecedents or consequences of key psychological events. For example, people study emotions to understand how they guide health choices (Peters et al., 2006) and how emotion can be the result of trauma or upheavals (Galea et al., 2020; Markowitz, 2022; Seraj et al., 2021). We adopt a similar approach and attempt to understand emotion and affect as a consequence of technological innovation. The feelings people have about technology—reflecting either utopian or dystopian viewpoints—form a narrative about how the public should also think and feel about technology. This agenda-setting, trickle-down perspective is essential to assess at this paradigm-shifting moment with the rise of chatbots, LLMs, and generative AI. We take an innovative approach to this research agenda by tracking emotion in language patterns, at scale, and on social media.

The idea that language patterns can reveal important social and psychological information about people originates from a research tradition that suggests words serve as markers of attention and focus (Boyd & Markowitz, 2024; Boyd & Schwartz, 2021; Pennebaker, 2011). For example, word patterns have been linked to personality traits and individual differences (Ireland & Mehl, 2014; Newman et al., 2008; Pennebaker & King, 1999), well-being (Jaidka et al., 2020; Rude et al., 2004; Stirman & Pennebaker, 2001), societal discourse shifts in institutions (Jordan et al., 2019), and the main focus of the present article, emotion and affect (Doyle et al., 2021; Sauter, 2018; Vine et al., 2020). From this perspective, words indicate where or what people are attending to and where or what they are not attending to. A person who suggests “I hate technology” is focusing on a negative emotional state (e.g., using the word hate), but scholars in this tradition would not go so far as to suggest they felt negative. This words-as-attention model to understanding people through language is pervasive in the social sciences (Boyd & Markowitz, 2024; Boyd & Schwartz, 2021), with studies using words as a lens in humans, their psychological states, and internal processing of events.

Drawing on this approach, a range of studies suggest emotion can serve as an integral marker of social and societal-level change and can be evaluated linguistically at scale. For example, prior work tracked the linguistic patterns of bloggers after the 9/11 terrorist attacks and observed a sharp increase in emotion immediately after the event, returning to baseline weeks later (Cohn et al., 2004). Other work has found that, compared to before the COVID-19 pandemic, academics wrote with a more emotional focus in their journal articles during the COVID-19 pandemic (Markowitz, 2022). Specifically, those who wrote about COVID-19 during the pandemic had the greatest rate of negative emotion compared to those writing before the pandemic or during the pandemic but about other topics. Altogether, dozens of studies suggest that by tracking how people represent emotion in language, we can understand important social and psychological information about communicators and how they perceive the world. We are among the first to apply this large-scale, computational approach to the understanding of how journalists write about new technology such as LLMs and chatbots.

Journalists, Disruptive Technologies, and Twitter (Now X)

Language patterns can unveil how journalists make sense of disruptive technologies, offering insights about their perceptions and expectations of technology plus how those attitudes could impact audiences. Previous research on journalists and their responses to disruptive technologies can provide an initial understanding of their expectations about new technology.

Journalists tend to hold varying expectations about disruptive technologies. As the internet became widely used in the early 2000s, and as new multimedia tools for reporting emerged alongside the development of the World Wide Web, such technologies were met with optimism by journalists, who imagined how new media might improve efficiency in collecting information and connecting with sources (Chadha & Wells, 2016; O’Sullivan & Heinonen, 2008; Pont-Sorribes et al., 2013). New technologies indeed have aided reporting: databases are used to help journalists anticipate and predict newsworthy events (Linden, 2017); automated journalism can convert structured data about topics like sports, weather, and finance into narrative articles with little human involvement (Diakopoulos, 2019); and social media platforms amplify news distribution and improve opportunities for audience interaction as well as sourcing (Lewis & Molyneux, 2018).

Despite these efficiency expectations, journalists also have expressed ample concern about how emerging technologies might degrade their profession. Journalists for many years have worried about a profusion of low-quality information online (O’Sullivan & Heinonen, 2008), do-it-yourself approaches to reporting that undermine the value of professionals (Posseti, 2009), and the likelihood that “robot journalists” (i.e., software) would produce stories with algorithmically introduced errors (Kim & Kim, 2018), highlighting the conflict between immediacy afforded by new technologies and accuracy prioritized in journalism (Lin & Lewis, 2022; Linden, 2017). More existentially, journalists have been alarmed that the growing use of automation in journalism could threaten their future employment, or at least make journalists become overly reliant on machines (Kim & Kim, 2018; Linden, 2017; Moran & Shaikh, 2022).

As disruptive technologies become integrated in newsrooms, journalists also anticipate an increase in their obligations—a “hamster wheel” of evermore work to be done (Usher, 2016; cf. Bélair-Gagnon et al., 2022, 2024). Some reporters resist by adhering to a “principle of continuity” (O’Sullivan & Heinonen, 2008, p. 367), sticking to conventions and old habits that have long worked for them (see also Powers & Vera-Zambrano, 2019). Other journalists recognize that new technologies could help advance their careers, but that such potential might only be realized if they have the training and resources to achieve proficiency (Powers & Vera-Zambrano, 2019), which is not easily accomplished in an era of cutbacks for many newsroom staffs (Ferrucci & Perreault, 2021).

Of all the technologies to emerge in recent decades, perhaps none has been more widely adopted by journalists than social media, used as an integral tool in news production and distribution. Platforms like Twitter, Facebook, and TikTok have played crucial roles in breaking news stories and disseminating them, and media organizations have strategically engaged the networked architectures and algorithmic design of such platforms to reach audiences where they are (Hermida, 2018; Lewis & Molyneux, 2018). Twitter, in particular, has emerged as journalists’ preferred social media platform, especially in much of the English-speaking West, because its continuous “stream of news, comments and analysis” makes it a handy, always-on source for news and information (Hermida, 2018, p. 4). Journalists have played an outsized role on Twitter, serving as a “real-time interpretive community” (Araiza et al., 2016, p. 310), blending personal and professional identities in the way they share and discuss the news online, and relying on each other and other users to make sense of events in real-time (see discussion in Mellado & Hermida, 2021). Twitter has also been used extensively by journalists as a de facto public sphere, as they draw on the platform to collect a “modern version of person-on-the-street interviews” (Lewis & Molyneux, 2018, p. 16). Indeed, journalists, in the United States especially, have become so reliant on Twitter that some have questioned what this dependence has done to cloud journalists’ sense of news judgment (McGregor & Molyneux, 2020).

In 2022, Elon Musk bought Twitter for $44 billion. At the time, “journalists around the world looked on in alarm,” as Musk, who had a fractious relationship with journalists, threatened to dismantle an existing blue-check verification system that favored news media (Gotfredsen, 2023). Many journalists (as well as academics; see Braun, 2024) reacted by threatening to leave the platform for other online communities. And while some did, an analysis in early 2023 of some 4,000 journalists from 19 US news outlets found that only a small fraction of journalists actually deactivated their accounts—and, on average overall, journalists were tweeting about 3% less after Musk’s takeover of the platform (Gotfredsen, 2023). In mid-2023, Musk renamed the service X, but colloquially many people still refer to the platform as Twitter.

Journalists continue to use Twitter (now X) not only for professional practice, but also for their personal agendas. By sharing behind-the-scenes information with their followers, they build stronger, more personal relationships with their audiences and may enhance their reach and reputation (Mellado & Alfaro, 2020; Mellado & Hermida, 2021; Molyneux, 2019). Because of their engagement on the platform, some journalists have even become political influencers, tailoring content to what their audiences want based on what they glean as relevant through interactions with their followers (Peres-Neto, 2022). Despite newsroom guidelines that reinforce nonpartisanship and objectivity, and despite the harassment and hostility that many reporters confront online (Davis Kempton & Connolly-Ahern, 2022), journalists have long used the platform to disclose personal information, share opinions about news stories, and socialize with others (Hermida, 2018; Lasorsa et al., 2012; Lee et al., 2016), making Twitter an ideal venue in which to see how journalists contribute to shaping narratives about AI.

The Current Paper

Taken together, we attempt to understand the relationship between journalists’ language patterns and emotions, before and after the launch of ChatGPT. This work is timely and important because it is presently unclear how tastemakers of technology thought and felt about such a consequential tool at the time of its rise into public consciousness and use. The case of journalists presents a previously underappreciated, but vital, dimension of this taste-making role in society. Thus, examining the reactions of journalists offers a window into their influence as key sense-makers for how people come to perceive and eventually use emerging technologies such as AI (see Brennen et al., 2022). We applied various computational and natural language processing techniques to consider how emotions were reflected and revealed over time during this critical moment in technological and generative AI history. Against this backdrop, we propose the following overarching research question:

RQ: What is the relationship between journalists’ reaction to advances in AI and the manifestation of emotion?

Method

Data Collection

To evaluate how journalists focused on emotion in their public disclosures before and after the ChatGPT launch, we gathered Twitter handles from a collection of journalists among 18 major news outlets in the United States (n = 4,071 unique accounts).¹ We used the Twitter lists from each source to access journalist Twitter handles.² Using the Academic Twitter API (Barrie et al., 2022), we extracted all Tweets 2 months before (n = 534,757 Tweets) and 2 months after (n = 424,623 Tweets) the launch of ChatGPT from each handle (November 30, 2022). Our database contained a total of 959,380 Tweets.

Automated Text Analysis

We used an automated text analysis tool, Linguistic Inquiry and Word Count (LIWC), to examine the rate of emotion across Tweets written before and after the ChatGPT launch (Pennebaker et al., 2022). LIWC is a gold-standard text analysis program that has been used extensively in the social sciences to evaluate social and psychological dynamics like emotion (Boyd & Schwartz, 2021; Pennebaker, 2011; Tausczik & Pennebaker, 2010). The program counts words as percentage of the total word count per text, identifying the degree to which words are found in its internal dictionary of social (e.g., words related to friends), psychological (e.g., words related to cognition, emotion), and part of speech categories (e.g., articles, prepositions). For example, the statement “I believe AI will be great for the world” contains nine words, and LIWC identifies the following categories, including but not limited to self-references (I; 11.11% of the total word count) and positive emotion terms (great; 11.11%). All language dimensions were drawn from the standard LIWC-22 dictionary, and each Tweet received a score (e.g., the percentage of each Tweet containing a verbal dimension of interest) across all measures of tone and emotion (see Table 1). Given the size and scale of the data, we did not pre-process the texts and they were therefore analyzed as they appeared online.

Table 1.

Estimated Marginal Means in Linear Mixed Models.

Emotion dimension	Example	Before ChatGPT Launch		After ChatGPT Launch
Emotion dimension	Example	M	SE	M	SE	t	p	R² m	R ² c
Positive emotion (%)	good, love, happy	1.01	0.01	1.10	0.02	12.54	< .001	1.6E-04	4.7E-02
Negative emotion (%)	bad, hate, hurt	0.51	0.01	0.51	0.01	–0.80	.422	6.9E-07	6.7E-03
Positive tone (%)	accomplish, improve	3.88	0.03	4.07	0.03	14.94	< .001	2.3E-04	8.3E-02
Negative tone (%)	fright, plague	1.67	0.01	1.60	0.01	–9.29	< .001	9.2E-05	2.4E-02

Measures

We took a layered approach to evaluate the relationship between Time (before ChatGPT vs after ChatGPT) and emotion by evaluating different aspects of emotion: (1) discrete positive and negative emotions, and (2) positive and negative sentiment (tone). As prior work suggests, there are six discrete emotions (sadness, happiness, fear, anger, surprise, disgust) (Barrett, 2017), and LIWC has two separate categories to approximate discrete positive emotions (e.g., words such as amaze, awesome, and excellent) and discrete negative emotions (e.g., words such as agitate, suffer, and terrify). Sentiment, or tone, describes the general positive or negative feeling that a text may elicit (Preoţiuc-Pietro et al., 2016). This dimension is linked to, but still independent from, emotion. For example, the word birthday is a positively-valenced term and elicits a positive tone, but the word is not a discrete emotion. We therefore evaluated two discrete emotion categories (i.e., emo_pos and emo_neg in LIWC) and two sentiment or tone categories (i.e., tone_pos and tone_neg in LIWC).

Analytic Plan

Using the lme4 and lmerTest packages in R (Bates et al., 2015; Kuznetsova et al., 2020), we computed linear mixed models with a random intercept for Tweet writer to control for data non-independence. Four models were computed (one for each dependent variable), predicting emotion or sentiment from Time (before ChatGPT launch vs after ChatGPT launch).

Results

Estimated marginal means and effect sizes for each model are represented in Table 1. Overall trends in emotion and tone over time are provided in Figure 1 for illustration purposes. Descriptively, we also demonstrate how positive and negative emotion and tone were reflected in actual Tweets with the following examples (italicized words reflect the terms counted within each LIWC category):

(a) Positive emotion Tweet from Gus Garcia-Roberts: “AI is getting too good,”

(b) Positive tone Tweet from Pranav Dixit: “At least AI is fascinating and interesting and horrifying and exciting in all the ways crypto never was,”

(c) Negative emotion Tweet from CNN: “These artists found out their work was used to train AI. Now they’re furious [link],” and

(d) Negative tone Tweet from Yun-Hee Kim: “The year AI became eerily human [link].”

Figure 1.

Descriptive Trends in Emotion and Tone Over Time.

Positive and Negative Emotion

Writers focused on more positive emotion after the ChatGPT launch compared to before the ChatGPT launch (t = 12.54, p < .001). The relationship between time and negative emotion was not statistically significant (t = −0.80, p = .422).

Positive and Negative Tone

Writers had a more positive tone after the ChatGPT launch compared to before the ChatGPT launch (t = 14.94, p < .001). Writers also had a less negative tone after the ChatGPT launch compared to before the ChatGPT launch (t = −9.29, p < .001). Altogether, writers were more positive and less negative after the ChatGPT launch versus before the ChatGPT launch. It is also important to note that the prior effects in Table 1 were maintained after accounting for the overall daily happiness rating of Twitter using the Hedonometer (Dodds et al., 2011, 2015).

Alternative Explanations

One possible explanation for the prior effects is that journalists were reporting the general mood or interest level of the public, not their own mood or interest level as they learned about ChatGPT. We therefore evaluated the prior relationships between time and emotion after accounting for Google Trends data. Google Trends provides a metric of search term popularity by day, and we considered the popularity of the search term “chatgpt” in the United States over time.

After accounting for Google Trends data as a fixed effect in linear mixed models, writers still focused on more positive emotion after the ChatGPT launch compared to before the ChatGPT launch (t = 11.49, p < .001). Consistent with our prior results and compared to before the ChatGPT launch, writers had a more positive tone (t = 16.04, p < .001) but a less negative tone after the ChatGPT launch (t = −9.92, p < .001) upon controlling for Google Trends data.

A second alternative explanation for these effects is that the general increase in positivity and decrease in negativity over time is the result of context effects, namely that the holiday season was within the timeframe of interest. Perhaps people were generally more positive in December and January (compared to October and November) because of holidays, not ChatGPT. LIWC’s religion category contains holiday-specific words (e.g., Christmas, church, faith, temple) and therefore provides a useful proxy for holiday words. All relationships listed in Table 1 were maintained after controlling for holiday words via the religion index.

A final and crucial alternative explanation is the idea that Tweets focusing on AI and AI-related topics might be written differently than those without such terms. We therefore manually created a case-insensitive AI dictionary of 11 terms (i.e., chatgpt, chatbot, bot, artificial intelligence, ai, openai, gpt-3, gpt-3.5, llm, large language models, large language model), accounting for the presence or absence of AI in our models as a fixed effect (n = 3,006 Tweets contained at least one term, or 0.3% of all Tweets). All results reported in Table 1 were maintained. That is, Tweets contained more positive emotion (t = 12.72, p < .001) and positive tone (t = 15.15, p < .001) after the ChatGPT release compared to before. The relationship between time and negative emotion was not statistically significant after accounting for AI-related language (t = −0.74, p = .458), and there was a less negative tone after the ChatGPT release compared to before (t = −9.16, p < .001).

Discussion

Disruptive technologies such as AI give rise to competing narratives around their benefits and risks in society. The introduction of generative AI in the well-publicized form of ChatGPT in late 2022 is no exception, with contrasting discourses of hope and fear surrounding its implications for business, creative industries, education, and other sectors. Journalists, known for their ability to shape public imagination by offering the “first draft of history” on many noteworthy issues and events, perform an important role as tastemakers in the ongoing narratives around emerging technologies (Brennen et al., 2022). Our research adds to this understanding by examining journalists’ emotions before and after the public launch of ChatGPT. This approach aligns with and enhances the “emotional turn” in journalism studies (Wahl-Jorgensen, 2020) by exploring how journalists’ expressions on Twitter/X may reflect broader affective dynamics associated with technological change in journalism.

Our research offers at least three main contributions. First, it extends the study of technology and society by providing a large-scale empirical investigation into the emotional reactions of journalists on Twitter/X during a critical juncture of technological disruption. Second, it responds to the call for an “emotional turn” in journalism studies—the need to empirically and theoretically understand emotions and how they interact with norms, values, and practices in journalism (Wahl-Jorgensen & Pantti, 2021). Specifically, our findings contribute to discussions of how journalists may strategically or spontaneously display emotion in public discourse (see Wahl-Jorgensen, 2020); we go further in showing how such expressions may shape collective perceptions of technology. Third, by conducting an extensive analysis of language patterns on social media, we also offer a methodological approach for others to follow in identifying how influential tastemakers react to major technological disruptions.

Ultimately, it matters to study these prominent reactions because they contribute to shaping public narratives about technologies and their impact—whether pro or con, hopeful or fearful. In turn, the stories that people tell themselves about technology influence not only how emergent technologies are understood in the present, but also how they are imagined for the future, with corresponding consequences for individual and collective decisions made around matters such as acceptance, adoption, and regulation (cf. Brennen et al., 2022).

Contributions to the Study of Journalism and Emotion

Our findings build on Wahl-Jorgensen’s (2020) argument that emotions in journalism have long been overlooked because of journalism’s long-standing commitment to objectivity, which tends to make journalists’ emotional work largely invisible. While previous research has examined emotions in newsrooms, our study moved beyond traditional settings to consider digital platforms, where journalists increasingly display emotion in a more visible and networked way. The positive emotional valence toward ChatGPT documented in this study suggests that journalists’ responses were not merely passive or accidental reflections of external technological change but were part of a broader affective engagement with digital innovation. This may accord with Wahl-Jorgensen’s (2020) concept of the “strategic ritual of emotionality,” which describes how journalists use emotions in their reporting—often in the form of personalized storytelling and anecdotal leads in “human interest stories”—to make their work compelling and relatable. Such analyses of the relationship among journalists, public discourse, and emotional expressions have been limited not only to traditional forms of news texts and environments but also to mostly small case studies; our research thus offers a first-of-its-kind examination of emotions at scale.

While our findings document that journalists met the development of ChatGPT with positivity, it is useful to ask why journalists reacted in this way. Wahl-Jorgensen (2020) argues for moving beyond describing emotions to theorizing their role in professional identity and technological adaptation. The distinct optimism we observed in journalists’ early reactions to ChatGPT may be indicative of the broader hype cycles associated with new technology, wherein initial enthusiasm often precedes more critical assessments that come later (Bourne, 2024). This pattern suggests that journalists may have initially seen ChatGPT as a possible solution to persistently frustrating professional challenges, such as the digital-driven demands to produce more with fewer resources, rather than as an immediate threat to job security.

This positivity, however, is rather noteworthy because it was far from certain. In fact, it may be rather surprising given the recent history of disruptive technologies in journalism, where a succession of innovations—from the internet to social media to the smartphone—have largely undercut the traditional business models for journalism, leading to the layoffs of thousands of reporters and editors, and at the same time have amplified the “digital demands” on journalists, forcing those still in the profession to do more and more with fewer and fewer resources (e.g., see Bélair-Gagnon et al., 2022, 2024; Lewis & Molyneux, 2018). As a result of the growing anxiety, stress, and burnout that many journalists report, many are simply calling it quits (Mathews et al., 2023). So, for journalists to express such positive emotion about the introduction of generative AI—a technology that some observers initially feared could replace human writers, amplifying the challenges facing an already troubled profession like journalism—offers a surprising twist. Perhaps it says something about the number of journalists who might have seen in ChatGPT a tool that could take on their grunt work, providing a respite from the incessant “hamster wheel” of expectations surrounding digital publishing (Usher, 2016).

Imagined Futures of Labor-Saving Technologies

Indeed, historically, people’s reactions to technologies believed to be “labor-saving”—such as semi-automatic technologies like type-cast printing and the loom—were initially positive in some cases, particularly when such tools improved industries and elevated standards of living (Carlopio, 1988). The positive emotions we document in this study may also signal an initial inclination toward the hopeful imaginaries associated with AI, both AI generally and generative AI particularly, which relates to how such technologies can be used for such labor-saving functions—making work more efficient, freeing journalists from manual tasks in reporting, and augmenting economic and social capacity broadly (Cave & Dihal, 2019; Hautala & Heino, 2023). These positive emotions also run parallel to how journalists initially viewed the internet in the early 2000s (Chadha & Wells, 2016), even if those feelings may have changed as digitally driven burnout increased (see examples in Bélair-Gagnon et al., 2024).

It must be acknowledged that as new and disruptive technologies become fully integrated into workflows, they may exacerbate existing unjust economic, social, and cultural conditions, leading to moral panics often associated with new technology (Carlopio, 1988). Historically, the mismanagement of such technologies has led to negative perceptions and even prompted workers to destroy the machines and tools they deemed threatening (Carlopio, 1988). In the recent past, the emergence of new technologies has contributed to increased expectations for journalists to possess the necessary skills to use these technologies as well as the actual obligation to use them (Bélair-Gagnon et al., 2022; Lewis & Molyneux, 2018). But the real, negative effects of new technologies on the bases of these expectations is likely to be felt belatedly rather than immediately, more so in hindsight than in the moment, much like the negative impacts of news digitalization. In other words, the negative emotions around generative AI may come later in the everyday discourse of journalists as the technology is widely used in newsrooms, which warrants a further examination of journalists’ emotions down the line.

The Value of Natural Language Processing

With these considerations in mind, it is also worth noting that our computational work offered major empirical advantages compared to other projects that might be interested in understanding the opinions of journalists in response to technological innovation. We used natural language from nearly 1 million social media posts to identify how thousands of journalists thought and felt about a technology in the moment that it was introduced. This metaphorical microscope, in terms of scale (e.g., the number of people and posts in the analyses) and scope (e.g., the longitudinal nature of the work), is often difficult to capture for most studies that tend to be cross-sectional in nature or contain only a small number of journalists as participants. Our work has taken advantage of such computational social scientific approaches to understand psychological information about journalists through their own words—specifically, how they considered, made sense of, and felt about a new technology with uncertain prospects for their own future and profession. We advocate for scholars (and newsrooms) interested in understanding journalists’ opinions on technology to consider using such natural language processing techniques in their own work as well, as they offer a unique window into opinions that are immediate, unadulterated, and psychologically rich.

Limitations and Future Directions

We acknowledge that Twitter (now X) is only one of many spaces where journalists might express their reactions to a critical incident such as this one, and we also acknowledge that it is impossible to know precisely if and how public perceptions may have been influenced by journalists’ emotional valence in this instance or any other. However, our study captures a longitudinal snapshot of the broader range of platforms and venues where journalists’ emotions are displayed, and it offers a valuable look at the central question of this research: how the expression of emotion online offers a window into psychological reactions to AI. We are, of course, limited to correlational and not causal claims in this work due to the nature of the field study that was conducted. Despite this, Twitter’s role in public discourse is well-established in scholarly work, and it remains a preferred platform for journalists to break news and share personal opinions. Future research could explore how this phenomenon unfolds across other platforms, such as Facebook, to provide more comprehensive insights into journalists’ reactions as a taste-making community and the integration of these reactions into the news narratives that unfold in the weeks, months, and even years thereafter.

We were also limited to data from journalists associated with large and mainstream media entities. This was purposeful given our tastemaker and sense-maker framing, and it reflects the reality that journalists at major national news organizations tend to be mimicked by journalists at regional and local media outlets (e.g., see the concept of intermedia agenda-setting; McCombs & Valenzuela, 2020). However, future work might use journalists in smaller markets or at more niche publications to identify how our results compare. Finally, we collected these data shortly before the Twitter API became more expensive to use to conduct academic research, which, unfortunately, may complicate future work of this kind. Access to social media data is important for academe and industry alike (Rathje, 2024)—a conversation we believe is just in its infancy and deserves attention as collectively we seek to address some of society’s most pressing questions about human emotion, perception, and behavior.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Shirley Papé Chair in Emerging Media, an endowed chair professorship in the School of Journalism and Communication at the University of Oregon.

ORCID iDs

Seth C. Lewis

David M. Markowitz

Jon Benedik A. Bunquin

Notes

Author Biographies

Seth C. Lewis (PhD, University of Texas at Austin) is the Shirley Papé Chair in Emerging Media in the School of Journalism and Communication at the University of Oregon. His research interests include the shifting dynamics of news and journalism as well as artificial intelligence, human-machine communication, and the social implications of emerging technologies broadly.

David M. Markowitz (PhD, Stanford University) is an associate professor in the Department of Communication at Michigan State University. His research interests include studying language patterns from natural databases to infer what people are thinking, feeling, and experiencing psychologically.

Jon Benedik A. Bunquin (MA, University of the Philippines) is a doctoral candidate in the School of Journalism and Communication at the University of Oregon and an assistant professor in the Department of Communication Research at the University of the Philippines. His research interests include using quantitative and computational methods to explore science and risk communication in digital media, focusing on the implications for public understanding and engagement with science and technology.

References

Araiza

J. A.

Sturm

H. A.

Istek

Bock

M. A.

(2016). Hands up, don’t shoot, whose side are you on? Journalists tweeting the Ferguson protests. Cultural Studies ↔ Critical Methodologies, 16(3), 305–312. https://doi.org/10.1177/1532708616634834

Bareis

Katzenbach

(2022). Talking AI into being: The narratives and imaginaries of national AI Strategies and their performative politics. Science, Technology, & Human Values, 47(5), 855–881. https://doi.org/10.1177/01622439211030007

Barrett

L. F.

(2017). How emotions are made: The secret life of the brain. Houghton Mifflin Harcourt Publishing Company.

Barrett

L. F.

Mesquita

Ochsner

K. N.

Gross

J. J.

(2007). The experience of emotion. Annual Review of Psychology, 58(1), 373–403. https://doi.org/10.1146/annurev.psych.58.110405.085709

Barrie

J. C.

Chan

Rico

König

Davidson

(2022). academictwitteR: Access the Twitter Academic Research Product Track V2 API Endpoint (0.3.1). https://CRAN.R-project.org/package=academictwitteR

Bates

Mächler

Bolker

B. M.

Walker

S. C.

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1). https://doi.org/10.18637/jss.v067.i01

Beckert

(2016). Imagined Futures: Fictional Expectations and Capitalist Dynamics. Harvard University Press.

Bélair-Gagnon

Bossio

Holton

A. E.

Molyneux

(2022). Disconnection: How measured separations from journalistic norms and labor can help sustain journalism. Social Media + Society, 8(1), 20563051221077217.

Bélair-Gagnon

Holton

Deuze

Mellado

(Eds.). (2024). Happiness in journalism. Routledge.

10.

Boden

M. A.

(2018). Artificial intelligence: A very short introduction. Oxford University Press.

11.

Bourne

(2024). AI hype, promotional culture, and affective capitalism. AI and Ethics, 1–13.

12.

Boyd

R. L.

Markowitz

D. M.

(2024). Verbal behavior and the future of social science. American Psychologist. Advance online publication. https://doi.org/10.1037/amp0001319

13.

Boyd

R. L.

Schwartz

H. A.

(2021). Natural language analysis and the psychology of verbal behavior: The past, present, and future states of the field. Journal of Language and Social Psychology, 40(1), 21–41. https://doi.org/10.1177/0261927X20967028

14.

Braun

(2024). Journalism, media research, and Mastodon: Notes on the future. Digital Journalism, 12(4), 557–564.

15.

Brennen

J. S.

Howard

P. N.

Nielsen

R. K.

(2022). What to expect when you’re expecting robots: Futures, expectations, and pseudo-artificial general intelligence in UK news. Journalism, 23(1), 22–38. https://doi.org/10.1177/1464884920947535

16.

Broussard

(2018). Artificial unintelligence: How computers misunderstand the world. MIT Press.

17.

Cacioppo

J. T.

Gardner

W. L.

(1999). Emotion. Annual Review of Psychology, 50(1), 191–214. https://doi.org/10.1146/annurev.psych.50.1.191

18.

Carlopio

(1988). A history of social psychological reactions to new technology. Journal of Occupational Psychology, 61(1), 67–77. https://doi.org/10.1111/j.2044-8325.1988.tb00272.x

19.

Carlson

(2015). The Robotic Reporter: Automated journalism and the redefinition of labor, compositional forms, and journalistic authority. Digital Journalism, 3(3), 416–431. https://doi.org/10.1080/21670811.2014.976412

20.

Carlson

Robinson

Lewis

S. C.

(2021). News after Trump: Journalism’s crisis of relevance in a changed media culture. Oxford University Press.

21.

Cave

Dihal

(2019). Hopes and fears for intelligent machines in fiction and reality. Nature Machine Intelligence, 1(2), Article 2. https://doi.org/10.1038/s42256-019-0020-9

22.

Chadha

Wells

(2016). Journalistic responses to technological innovation in newsrooms. Digital Journalism, 4(8), 1020–1035. https://doi.org/10.1080/21670811.2015.1123100

23.

Cohn

M. A.

Mehl

M. R.

Pennebaker

J. W.

(2004). Linguistic markers of psychological change surrounding September 11, 2001. Psychological Science, 15(10), 687–693. https://doi.org/10.2307/40064028

24.

Das

A. C.

Phalin

Patidar

I. L.

Gomes

Sawhney

Thomas

(2023). AI customer service for higher customer engagement. https://www.mckinsey.com/capabilities/operations/our-insights/the-next-frontier-of-customer-engagement-ai-enabled-customer-service

25.

Davis Kempton

Connolly-Ahern

. (2022). “Who’s going to be a creep today?” Understanding the social media experiences of women broadcast journalists. Social Media + Society, 8(2), 20563051221108410.

26.

Diakopoulos

(2019). Automating the news: How algorithms are rewriting the media. Harvard University Press.

27.

Dodds

P. S.

Clark

E. M.

Desu

Frank

M. R.

Reagan

A. J.

Williams

J. R.

Mitchell

Harris

K. D.

Kloumann

I. M.

Bagrow

J. P.

Megerdoomian

McMahon

M. T.

Tivnan

B. F.

Danforth

C. M.

(2015). Human language reveals a universal positivity bias. Proceedings of the National Academy of Sciences, 112(8), 2389–2394. https://doi.org/10.1073/pnas.1411678112

28.

Dodds

P. S.

Harris

K. D.

Kloumann

I. M.

Bliss

C. A.

Danforth

C. M.

(2011). Temporal patterns of happiness and information in a global social network: Hedonometrics and Twitter. PLOS ONE, 6(12), e26752. https://doi.org/10.1371/journal.pone.0026752

29.

Doyle

C. M.

Gendron

Lindquist

K. A.

(2021). Language is a unique context for emotion perception. Affective Science, 2(2), 171–177. https://doi.org/10.1007/s42761-020-00025-7

30.

Ekman

Friesen

W. V.

O’Sullivan

Chan

Diacoyanni-Tarlatzis

Heider

Krause

LeCompte

W. A.

Pitcairn

Ricci-Bitti

P. E.

Scherer

Tomita

Tzavaras

(1987). Universals and cultural differences in the judgments of facial expressions of emotion. Journal of Personality and Social Psychology, 53(4), 712–717. https://doi.org/10.1037/0022-3514.53.4.712

31.

Entman

R. M.

Matthes

Pellicano

(2009). Nature, sources, and effects of news framing. In The handbook of journalism studies (pp. 195–210). Routledge.

32.

Ferrucci

Perreault

(2021). The liability of newness: Journalism, innovation and the issue of core competencies. Journalism Studies, 22(11), 1436–1449. https://doi.org/10.1080/1461670X.2021.1916777

33.

Galea

Merchant

R. M.

Lurie

(2020). The mental health consequences of COVID-19 and physical distancing: The need for prevention and early intervention. JAMA Internal Medicine, 180(6), 817–818. https://doi.org/10.1001/jamainternmed.2020.1562

34.

Gotfredsen

S. G.

(2023). Journalists Remain on Twitter, but Tweet Slightly Less. Columbia Journalism Review. https://www.cjr.org/tow_center/journalists-remain-on-twitter-but-tweet-slightly-less

35.

Gross

J. J.

(1998). The emerging field of emotion regulation: An integrative review. Review of. General Psychology, 2, 271–299. https://doi.org/10.1037/1089-2680.2.3.271

36.

Guzman

A. L.

Lewis

S. C.

(2020). Artificial intelligence and communication: A human–machine communication research agenda. New Media & Society, 22(1), 70–86.

37.

Hancock

J. T.

Naaman

Levy

(2020). AI-mediated communication: Definition, research agenda, and ethical considerations. Journal of Computer-Mediated Communication, 25(1), 89–100.

38.

Hatzius

Briggs

Kodnani

Pierdomenico

(2023). The potentially large effects of artificial intelligence on economic growth (Economics Research). Goldman Sachs.

39.

Hautala

Heino

(2023). Spectrum of AI futures imaginaries by AI practitioners in Finland and Singapore: The unimagined speed of AI progress. Futures, 153, 103247. https://doi.org/10.1016/j.futures.2023.103247

40.

Hermida

(2018). Social media and journalism. In The SAGE handbook of social media (pp. 497–511). Sage. https://doi.org/10.4135/9781473984066

41.

Ireland

M. E.

Mehl

(2014). Natural language use as a marker of personality. In Holtgraves

T. M.

(Ed.), The Oxford handbook of language and social psychology (pp. 201–218). Oxford University Press.

42.

Jaidka

Giorgi

Schwartz

H. A.

Kern

M. L.

Ungar

L. H.

Eichstaedt

J. C.

(2020). Estimating geographic subjective well-being from Twitter: A comparison of dictionary and data-driven language methods. Proceedings of the National Academy of Sciences, 117(19), 10165–10171. https://doi.org/10.1073/pnas.1906364117

43.

James

(1884). What is an emotion? Mind, 9(34), 188–205.

44.

Jasanoff

Kim

S.-H.

(2009). Containing the atom: Sociotechnical imaginaries and nuclear power in the United States and South Korea. Minerva, 47(2), 119–146. https://doi.org/10.1007/s11024-009-9124-4

45.

Johnson

D. G.

Verdicchio

(2017). Reframing AI discourse. Minds and Machines, 27(4), 575–590. https://doi.org/10.1007/s11023-017-9417-6

46.

Jordan

K. N.

Sterling

Pennebaker

J. W.

Boyd

R. L.

(2019). Examining long-term trends in politics and culture through language of political leaders and cultural institutions. Proceedings of the National Academy of Sciences of the United States of America, 116(9), 3476–3481. https://doi.org/10.1073/pnas.1811987116

47.

Kelly

(2023, September 13). How AI is a game changer for recruiting and retention at Salesforce and other companies. Forbes. https://www.forbes.com/sites/jackkelly/2023/09/13/how-ai-is-a-game-changer-for-recruiting-and-retention-at-salesforce-and-other-companies/

48.

Kim

(2018). Newspaper journalists’ attitudes towards robot journalism. Telematics and Informatics, 35(2), 340–357. https://doi.org/10.1016/j.tele.2017.12.009

49.

Kramer

A. D. I.

Guillory

J. E.

Hancock

J. T.

(2014). Experimental evidence of massive-scale emotional contagion through social networks. Proceedings of the National Academy of Sciences, 111(24), 8788–8790. https://doi.org/10.1073/pnas.1320040111

50.

Kuznetsova

Brockhoff

P. B.

Christensen

R. H. B.

Jensen

S. P.

(2020). lmerTest: Tests in linear mixed effects models (3.1-3). https://CRAN.R-project.org/package=lmerTest

51.

Langer

A. I.

Gruber

J. B.

(2021). Political agenda setting in the hybrid media system: Why legacy media still matter a great deal. The International Journal of Press/Politics, 26(2), 313–340. https://doi.org/10.1177/1940161220925023

52.

Lasorsa

D. L.

Lewis

S. C.

Holton

A. E.

(2012). Normalizing Twitter. Journalism Studies, 13(1), 19–36. https://doi.org/10.1080/1461670X.2011.571825

53.

Lecheler

De Vreese

C. H.

(2012). News framing and public opinion: A mediation analysis of framing effects on political attitudes. Journalism & Mass Communication Quarterly, 89(2), 185–204.

54.

Lee

N. Y.

Kim

(2016). Tweeting public affairs or personal affairs? Journalists’ tweets, interactivity, and ideology. Journalism, 17(7), 845–864. https://doi.org/10.1177/1464884915585954

55.

Lewis

S. C.

Molyneux

(2018). A decade of research on social media and journalism: Assumptions, blind spots, and a way forward. Media and Communication, 6(4), 11–23.

56.

Lin

Lewis

S. C.

(2022). The one thing journalistic AI just might do for democracy. Digital Journalism, 10(10), 1627–1649.

57.

Linden

C.-G.

(2017). Decades of automation in the newsroom. Digital Journalism, 5(2), 123–140. https://doi.org/10.1080/21670811.2016.1160791

58.

Markowitz

D. M.

(2022). Psychological trauma and emotional upheaval as revealed in academic writing: The case of COVID-19. Cognition and Emotion, 36(1), 9–22. https://doi.org/10.1080/02699931.2021.2022602

59.

Maslej

Fattorini

Brynjolfsson

Ligett

Lyons

Manyika

Ngo

Niebles

J. C.

Parli

Shoham

Wald

Clark

Perrault

(2023). The AI Index 2023 Annual Report. AI Index Steering Committee, Institute for Human-Centered AI, Stanford University. https://aiindex.stanford.edu/report/

60.

Mathews

Bélair-Gagnon

Carlson

(2023). “Why I quit journalism”: Former journalists’ advice giving as a way to regain control. Journalism, 24(1), 62–77.

61.

McCombs

Valenzuela

(2020). Setting the agenda: Mass media and public opinion. John Wiley & Sons.

62.

McGregor

S. C.

Molyneux

(2020). Twitter’s influence on news judgment: An experiment among journalists. Journalism, 21(5), 597–613.

63.

Mellado

Alfaro

(2020). Platforms, journalists and their digital selves. Digital Journalism, 8(10), 1258–1279. https://doi.org/10.1080/21670811.2020.1817763

64.

Mellado

Hermida

(2021). The promoter, celebrity, and joker roles in journalists’ social media performance. Social Media + Society, 7(1), 2056305121990643.

65.

Mitchell

(2023, January 25). ChatGPT could make these jobs obsolete. https://nypost.com/2023/01/25/chat-gpt-could-make-these-jobs-obsolete/

66.

Mitchell

(2019). Artificial intelligence: A guide for thinking humans. Pelican.

67.

Mollick

(2024). Co-Intelligence. Random House UK.

68.

Molyneux

(2019). A personalized self-image: Gender and branding practices among journalists. Social Media + Society, 5(3), 2056305119872950.

69.

Moran

R. E.

Shaikh

S. J.

(2022). Robots in the news and newsrooms: Unpacking meta-journalistic discourse on the use of artificial intelligence in journalism. Digital Journalism, 10(10), 1756–1774. https://doi.org/10.1080/21670811.2022.2085129

70.

Nabi

R. L.

(2003). Exploring the framing effects of emotion: Do discrete emotions differentially influence information accessibility, information seeking, and policy preference? Communication Research, 30(2), 224–247. https://doi.org/10.1177/0093650202250881

71.

Nabi

R. L.

(2010). The case for emphasizing discrete emotions in communication research. Communication Monographs, 77(2), 153–159.

72.

Neiger

(2007). Media oracles: The cultural significance and political import of news referring to future events. Journalism, 8(3), 309–321. https://doi.org/10.1177/1464884907076464

73.

Newman

M. L.

Groom

C. J.

Handelman

L. D.

Pennebaker

J. W.

(2008). Gender differences in language use: An analysis of 14,000 text samples. Discourse Processes, 45(3), 211–236. https://doi.org/10.1080/01638530802073712

74.

Nowland

Necka

E. A.

Cacioppo

J. T.

(2018). Loneliness and social internet use: Pathways to reconnection in a digital world? Perspectives on Psychological Science, 13(1), 70–87. https://doi.org/10.1177/1745691617713052

75.

O’Sullivan

Heinonen

(2008). Old values, new media. Journalism Practice, 2(3), 357–371. https://doi.org/10.1080/17512780802281081

76.

Pennebaker

J. W.

(2011). The secret life of pronouns: What our words say about us. Bloomsbury Press.

77.

Pennebaker

J. W.

Boyd

R. L.

Booth

R. J.

Ashokkumar

Francis

M. E.

(2022). Linguistic inquiry and word count: LIWC-22. Pennebaker Conglomerates. https://www.liwc.app

78.

Pennebaker

J. W.

King

L. A.

(1999). Linguistic styles: Language use as an individual difference. Journal of Personality and Social Psychology, 6, 1296–1312. https://doi.org/10.1037/0022-3514.77.6.1296

79.

Peres-Neto

(2022). Journalist-Twitterers as political influencers in Brazil: Narratives and disputes towards a new intermediary model. Media and Communication, 10(3), 28–38.

80.

Perreault

G. P.

Lewis

S. C.

Ely

(forthcoming). ChatGPT, Generative AI, and an Epistemic Opportunity for Journalistic Authority. Digital Journalism.

81.

Peters

Lipkus

Diefenbach

M. A.

(2006). The functions of affect in health communications and in the construction of health preferences. Journal of Communication, 56, S140–S162. https://doi.org/10.1111/j.1460-2466.2006.00287.x

82.

Pont-Sorribes

Rovira

S. C.

Di Bonito

(2013). Challenges and opportunities for science journalists in adopting new technologies: The case of Spain. Journal of Science Communication, 12(3), A05. https://doi.org/10.22323/2.12030205

83.

Posseti

(2009, June 16). Twitter’s difficult gift to journalism. New Matilda. https://newmatilda.com/2009/06/16/twitters-difficult-gift-journalism/

84.

Powers

Vera-Zambrano

(2019). Endure, invest, ignore: How French and American journalists react to economic constraints and technological transformations. Journal of Communication, 69(3), 320–343. https://doi.org/10.1093/joc/jqz015

85.

Preoţiuc-Pietro

Schwartz

H. A.

Park

Eichstaedt

Kern

Ungar

Shulman

(2016). Modelling valence and arousal in Facebook posts. Proceedings of the 7th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, 9–15. https://doi.org/10.18653/v1/W16-0404

86.

Rathje

(2024). To tackle social-media harms, mandate data access for researchers. Nature, 633(8028), 36.

87.

Rude

Gortner

E.-M.

Pennebaker

(2004). Language use of depressed and depression-vulnerable college students. Cognition and Emotion, 18(8), 1121–1133. https://doi.org/10.1080/02699930441000030

88.

Sartori

Bocca

(2023). Minding the gap(s): Public perceptions of AI and socio-technical imaginaries. AI & SOCIETY, 38(2), 443–458. https://doi.org/10.1007/s00146-022-01422-1

89.

Sauter

D. A.

(2018). Is there a role for language in emotion perception? Emotion Review, 10(2), 111–115. https://doi.org/10.1177/1754073917693924

90.

Schudson

(2018). Why journalism still matters. John Wiley & Sons.

91.

Seraj

Blackburn

K. G.

Pennebaker

J. W.

(2021). Language left behind on social media exposes the emotional and cognitive costs of a romantic breakup. Proceedings of the National Academy of Sciences, 118(7), e2017154118. https://doi.org/10.1073/pnas.2017154118

92.

Singer

J. B.

(1998). Online Journalists: Foundations for Research into Their Changing Roles. Journal of Computer-Mediated Communication, 4(1), JCMC412. https://doi.org/10.1111/j.1083-6101.1998.tb00088.x

93.

Slovic

Finucane

M. L.

Peters

MacGregor

D. G.

(2004). Risk as analysis and risk as feelings: Some thoughts about affect, reason, risk, and rationality. Risk Analysis, 24(2), 311–322. https://doi.org/10.1111/j.0272-4332.2004.00433.x

94.

Stahl

(2023, March 3). Will ChatGPT Replace Your Job? Forbes. https://www.forbes.com/sites/ashleystahl/2023/03/03/will-chatgpt-replace-your-job/

95.

Statement on AI Risk. (2023). Center for AI safety. https://www.safe.ai/statement-on-ai-risk

96.

Stirman

S. W.

Pennebaker

J. W.

(2001). Word use in the poetry of suicidal and nonsuicidal poets. Psychosomatic Medicine, 63, 517–522.

97.

Tausczik

Y. R.

Pennebaker

J. W.

(2010). The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29(1), 24–54. https://doi.org/10.1177/0261927X09351676

98.

Tenenboim-Weinblatt

Neiger

(2015). Print is future, online is past: Cross-media analysis of temporal orientations in the news. Communication Research, 42(8), 1047–1067. https://doi.org/10.1177/0093650214558260

99.

Usher

(2016). The constancy of immediacy: From printing press to digital age. In Alexander

J. C.

Breese

E. B.

Luengo

(Eds.), The crisis of journalism reconsidered: Democratic culture, professional codes, digital future (pp. 170–189). Cambridge University Press.

100.

Valkenburg

P. M.

Peter

(2007). Online communication and adolescent well-being: Testing the stimulation versus the displacement hypothesis. Journal of Computer-Mediated Communication, 12(4), 1169–1182.

101.

Vine

Boyd

R. L.

Pennebaker

J. W.

(2020). Natural emotion vocabularies as windows on distress and well-being. Nature Communications, 11(1), 1–9. https://doi.org/10.1038/s41467-020-18349-0

102.

Vrabič Dežman

. (2024). Promising the future, encoding the past: AI hype and public media imagery. AI and Ethics. https://doi.org/10.1007/s43681-024-00474-x

103.

Wahl-Jorgensen

(2020). An emotional turn in journalism studies? Digital Journalism, 8(2), 175–194.

104.

Wahl-Jorgensen

Carlson

(2021). Conjecturing Fearful futures: Journalistic discourses on deepfakes. Journalism Practice, 15(6), 803–820. https://doi.org/10.1080/17512786.2021.1908838

105.

Wahl-Jorgensen

Pantti

(2021). Introduction: The emotional turn in journalism. Journalism, 22(5), 1147–1154. https://doi.org/10.1177/1464884920985704

106.

Wang

Downey

Yang

(2023). AI anxiety? Comparing the sociotechnical imaginaries of artificial intelligence in UK, Chinese and Indian newspapers. Global Media and China, 20594364231196547. https://doi.org/10.1177/20594364231196547

107.

Yin

(1999). Elite opinion and media diffusion: Exploring environmental attitudes. Harvard International Journal of Press/Politics, 4(3), 62–86.

Journalists,Emotions,and the Introduction of Generative AI Chatbots: A Large-Scale Analysis of Tweets Before and After the Launch of ChatGPT

Abstract

Keywords

Literature Review and Contextual Background

AI and Imagined Futures

The Import of Affect and Emotion: A Psychology of Language Perspective

Journalists, Disruptive Technologies, and Twitter (Now X)

The Current Paper

Method

Data Collection

Automated Text Analysis

Measures

Analytic Plan

Results

Positive and Negative Emotion

Positive and Negative Tone

Alternative Explanations

Discussion

Contributions to the Study of Journalism and Emotion

Imagined Futures of Labor-Saving Technologies

The Value of Natural Language Processing

Limitations and Future Directions

Footnotes

Declaration of Conflicting Interests

Funding

ORCID iDs

Notes

Author Biographies

References