Veranstaltungsprogramm

Conspiracy theories have become a consequential force in democratic politics, increasingly shaping public discourse across digital and institutional media. Once limited to fringe outlets, narratives such as the “Great Replacement” and “New World Order” now appear in mainstream political rhetoric and media reporting. These narratives are especially central to far-right ideological projects, where they serve to mobilize ethno-nationalist identities and portray political conflict as a struggle against global conspiracies. Far-right hyperpartisan media such as Breitbart, InfoWars, and The Daily Caller have played a central role in circulating these ideas, often outside the bounds of traditional journalistic norms. However, legacy media institutions are not immune to these dynamics. Situated within hybrid media systems, they may engage with conspiratorial narratives through mechanisms of convergence, elite cueing, or topical overlap.

This paper analyzes how two illustrative conspiracy theories—the “New World Order” and the “Great Replacement / White Genocide”—are reported and diffused across far-right hyperpartisan and legacy U.S. news media between 2011 and 2021. Using a dataset of approximately 22,500 articles from eight media outlets, we combine time-series analysis, multiscale topic modeling, and large language model-based classification to examine when, what, and how conspiracy theories are featured.

We find that conspiracy-related content appears earlier and more frequently in far-right media, but increases across all sources over time. Convergence in topic coverage is often driven by elite cues (e.g., Trump-era controversies) or major events (e.g., COVID-19), though different media logics shape how these topics are framed. Far-right outlets tend to embed conspiracy theories in narrative form, while legacy media more often present counter-narratives that distance themselves from core conspiratorial claims.

The study contributes to scholarship on digital journalism, media polarization, and the diffusion of conspiratorial content. It highlights how networked media ecologies facilitate convergence between ideologically distinct media spheres and underscores the importance of narrative form in shaping public legitimacy.

This study analyses political communication and coherent comments during the 2025 German federal election campaign. Using computational text analysis of politicians’ videos and associated comment sections, we explore how thematic and affective dynamics vary across parties. The findings reveal strong affective polarization: fringe parties such as the AfD and Die Linke attract largely supportive and emotionally positive comment publics, while centrist actors face predominantly critical and hostile responses. These patterns suggest that TikTok fosters affective counterpublics that amplify ideological divisions and emotional resonance within digital political discourse.

Political elites increasingly use moral language to influence voting decisions by signaling their moral commitments. However, whether politicians’ moral language reflects their moral values remains unknown. To this end, we developed a research program combining candidate surveys and annotations of these candidates’ campaign communication. We collected survey data on the moral foundations of political candidates (N=949) running for office in four German state elections as well as these candidates’ Meta posts (N=17,646 observations). We conducted crowd-sourced annotations of the posts (N=320 coders) to fine-tune a transformer model that predicts moral labels for the entire corpus. Results show that candidates tend to express the moral concerns they personally endorse in their campaign communication. The study also finds variation in the alignment across the political spectrum, both at the individual and the party level. Parties on the political left and extreme political right show significantly greater alignment than centrist and liberal-conservative parties.

Introduction

Polarization is a widely debated social phenomenon. Affective polarization is considered particularly problematic for democratic discourse and societies' ability to address crises such as climate change because it hinders open debate, entrenches positions, and can foster destructive divisions (Esau et al., [in review]). While the polarization of textual content in online media has been studied extensively, little research has examined the interplay between exposure to biased media coverage and attitudes towards social groups. Despite the intense media coverage of disruptive climate movements, research on affective polarization in the climate context remains scarce (van Eck, 2024). Climate movements are therefore a key case study for investigating affective polarization and the social cues it triggers. Therefore, this study asks: To what extent does the biased portrayal of climate movements in the news media influence audience affective polarization towards them?

Theoretical Approach and Hypotheses

Potential media effects on affective polarization are closely linked to social identity theory (Tajfel & Turner, 2004). Intergroup evaluation may occur and trigger distinct social identity cues. Empirical research suggests that legitimacy, morality, and emotion influence people's perception of social groups and may therefore contribute to affective polarization. We therefore use these three dimensions as the analytical core to operationalize positively or negatively biased media exposure in our study. Based on these three dimensions, we hypothesize that participants with extreme initial attitudes remain stable when exposed to biased content, due to reinforcement and backfire effects (H1); neutral participants shift towards more negative attitudes when exposed to negative portrayals, due to persuasion effects and ambivalence reduction (H2a); neutral participants shift towards more positive attitudes when exposed to positive portrayals (H2b); participants with extreme initial attitudes may moderate their views when confronted with neutral content, as cognitive dissonance arises without reinforcing social identity cues (H3).

Method

Methodologically, we combined two main data sources to address the research question: In a four-wave panel study, about 1.800 participants were surveyed regarding their attitudes towards Fridays for Future (FFF) and Die Letzte Generation (LG), while their online news exposure was tracked in parallel. To capture attitude changes over time, we developed a new measure that identifies consistent trends across waves. Further, we performed an automated content analysis of more than 6.000 scraped news articles the participants were exposed to. Using GPT4.1, we automatically coded news content along the three dimensions of legitimacy, morality, and emotion, each represented by a validating and an invalidating category. News articles were included if they referred directly to FFF or LG. On the participant level, exposure variables were constructed: the absolute number of negative, positive, or overall biased portrayals, as well as their share within total exposure. These measures were then used for descriptive and inferential tests. Regression models controlling for demographics tested the stated hypotheses.

Results

Of the stated hypotheses, only H1 found weak support: participants with strongly negative initial attitudes consumed significantly more negative news content about FFF and/or LG in absolute terms. However, this effect disappeared once we controlled for overall news consumption. Descriptively, several noteworthy patterns emerged: The news coverage was more biased in case of LG compared to FFF, particularly in negative. Survey data mirrored this pattern: significantly more participants expressed strongly negative attitudes towards LG than towards FFF. When divided by initial attitude, participants showed significantly different attitudinal changes. Negative participants were especially stable in their views, while positive and neutral participants tended to shift towards more negative attitudes over time.

Discussion and Outlook

The results show LG was portrayed more negatively and with greater bias than FFF, which is consistent with earlier research pointing out that journalism often struggles to portray disruptive social movements fairly (Dablander et al., 2025). This may help to explain why public attitudes towards LG were also more negative and why links between news exposure and attitude change were more visible in this case. However, the majority of hypotheses were not supported. One likely explanation is that most participants did not exhibit significant attitudinal changes during the panel period. Additionally, average exposure levels were very low, with participants reading an average of fewer than five articles. Under these conditions, substantial effects of media bias were unlikely to manifest. Nevertheless, the descriptive results remain insightful. They suggest that once formed, attitudes towards climate movements are relatively stable and resistant to change. Negative portrayals, especially those depicting movements as threats, appear to influence social cues more strongly than positive portrayals. Furthermore, media coverage appears to mirror individuals' initial attitudes, suggesting that opinions may have been formed before the start of the panel period. Therefore, our results should not be interpreted as evidence that the news media play no role in shaping public opinion. On the contrary, they point to the potential importance of early media framing, particularly the choice of words and depiction of threats, in the early stages of opinion formation. Once attitudes are formed, they appear resistant to news media influence, suggesting that polarization may already be underway.

The rapid adoption of generative artificial intelligence (GenAI) is transforming how citizens seek and interpret political information, raising new questions about human autonomy in democratic opinion formation. This paper conceptualizes the reconfiguration of human political autonomy in the age of GenAI through a two-dimensional theoretical framework integrating internalist-procedural and externalist-relational perspectives. The internalist dimension concerns individuals’ capacity for reflective endorsement of political beliefs, while the externalist dimension addresses the informational environments enabling or constraining such reflection and political autonomy. GenAI’s persuasive, opaque, and hyper-personalized outputs threaten both dimensions by introducing biased or factually inaccurate information and by fostering patterns of unreflective reliance and informational isolation. Yet, GenAI may also enhance autonomy by prompting self-examination, exposing users to diverse viewpoints, and providing explanation and context. By conceptualizing these dynamics, the paper establishes a foundation for assessing when reliance on GenAI constitutes a loss of autonomy or a legitimate form of cognitive support, an urgent issue for democratic resilience.