The rise of conversational AI (CAI), powered by large language models, is transforming how individuals access and interact with digital information. However, these tools may inadvertently amplify existing digital inequalities. This study investigates whether differences in formal education are associated with CAI avoidance, i.e., the deliberate refusal or discontinuation of engaging with a CAI when the opportunity and demand arises. We leverage behavioral data from an online experiment (N = 1636) where participants were randomly assigned to one of three groups: a control group, a traditional online search task, or a CAI task (Perplexity AI). Task avoidance (operationalized as survey abandonment or providing unrelated responses during task assignment) was significantly higher in the CAI group (51 %) compared to the search (30.9 %) and control (16.8 %) groups, with the highest CAI avoidance among participants with lower education levels (similar to 74.4 %). Structural equation modeling based on the theoretical framework UTAUT2 and LASSO regressions reveal that education is strongly associated with CAI avoidance, even after accounting for various cognitive and affective predictors of technology adoption. These findings underscore education's central role in shaping AI adoption and the role of self-selection biases in AI-related research, stressing the need for inclusive design to ensure equitable access to emerging technologies.

Persistent inequalities in political knowledge are a central concern in political communication. We organize the mechanisms underlying the knowledge-gap literature by distinguishing between individual preconditions, structural features of the information environment, and topic characteristics. Within this framework, we note that self-directed information seeking, a prototypical form of intentional exposure, has received little attention despite its importance in navigating today's complex information environment. We conducted a field experiment in Germany combining randomized encouragements and passive browser tracking to examine how individuals with varying education levels acquire policy-specific knowledge through online search. Participants were randomly assigned to one of three conditions (verbal encouragement, financial encouragement, or control) to seek information on three salient policy topics differing in divisiveness and complexity (child support, energy transition, and cannabis legalization). We estimate both intention-to-treat (ITT) and local average treatment effects (LATE) of information seeking on post-search knowledge outcomes, with a focus on education and civic knowledge as moderators. While the interventions equalized information-seeking behavior, the results provide some support for the knowledge gap hypothesis: knowledge gains were concentrated among participants with higher education or baseline civic knowledge, who, according to our post-hoc exploratory analyses, appeared more effective at navigating search results. These findings indicate that a narrowing of knowledge inequalities goes beyond motivation: it calls for both individual-level interventions to strengthen citizens' skills and structural-level adaptations to foster more equitable learning environments.

As the exploration of digital behavioral data revolutionizes communication research, understanding the nuances of data collection methodologies becomes increasingly pertinent. This study focuses on one prominent data collection approach, web scraping;specifically, its application in the growing field of research relying on web browsing data. We investigate discrepancies between content obtained directly during user interaction with a website (in-situ) and content scraped using the URLs of participants' logged visits (ex-situ) with various time delays (0, 30, 60, and 90 days). We find substantial disparities between the methodologies, uncovering that errors are not uniformly distributed across news categories regardless of the classification method (domain, URL, or content analysis). These biases compromise the precision of measurements used in the existing literature. The ex-situ collection environment is the primary source of the discrepancies (33.8%), while the time delays in the scraping process play a smaller role (adding similar to 6.5% points in 90 days). Our research emphasizes the need for data collection methods that capture web content directly in the user's environment. However, acknowledging its complexities, we further explore strategies to mitigate biases in web-scraped browsing histories, offering recommendations for researchers who rely on this method and laying the groundwork for developing error-correction frameworks.

The proliferation of large language models (LLMs) can influence how historical narratives are disseminated and perceived. This study explores the implications of LLMs’ responses on the representation of mass atrocity memory, examining whether generative AI systems contribute to prosthetic memory, i.e., mediated experiences of historical events, or to what we term “prosthetic denial,” the AI-mediated erasure or distortion of atrocity memories. We argue that LLMs function as interfaces that can elicit prosthetic memories and, therefore, act as experiential sites for memory transmission, but also introduce risks of denialism, particularly when their outputs align with contested or revisionist narratives. To empirically assess these risks, we conducted a comparative audit of five LLMs—Claude, GPT, Llama, Mixtral, and Gemini—across four historical case studies: the Holodomor, the Holocaust, the Cambodian Genocide, and the genocide against the Tutsi in Rwanda. Each model was prompted with questions addressing common denialist claims in English and an alternative language relevant to each case (Ukrainian, German, Khmer, and French). Our findings reveal that while LLMs generally produce accurate responses for widely documented events like the Holocaust, significant inconsistencies and susceptibility to denialist framings are observed for more underrepresented cases like the Cambodian Genocide. The disparities highlight the influence of training data availability and the probabilistic nature of LLM responses on memory integrity. We conclude that while LLMs extend the concept of prosthetic memory, their unmoderated use risks reinforcing historical denialism, raising ethical concerns for (digital) memory preservation, and potentially challenging the advantageous role of technology associated with the original values of prosthetic memory.

The familiarity principle posits that acceptance increases with exposure, which has previously been shown with in vivo and simulated experiences with connected and autonomous vehicles (CAVs). We investigate the impact of a simulated video-based first-person drive on CAV acceptance, as well as the impact of information customization, with a particular focus on acceptance by older individuals and those with lower education. Findings from an online experiment with N=799 German residents reveal that the simulated experience improved acceptance across response variables such as intention to use and ease of use, particularly among older individuals. However, the opportunity to customize navigation information decreased acceptance of older individuals and those with university degrees and increased acceptance for younger individuals and those with lower educational levels.

Tag-Pag is an application designed to simplify the categorization of web pages, a task increasingly common for researchers who scrape web pages to analyze individuals' browsing patterns or train machine learning classifiers. Unlike existing tools that focus on annotating sections of text, Tag-Pag systematizes page-level annotations, allowing users to determine whether an entire document relates to one or multiple predefined topics. Tag-Pag offers an intuitive interface to configure the input web pages and annotation labels. It integrates libraries to extract content from the HTML and URL indicators to aid the annotation process. It provides direct access to both scraped and live versions of the web page. Our tool is designed to expedite the annotation process with features like quick navigation, label assignment, and export functionality, making it a versatile and efficient tool for various research applications. Tag-Pag is available at https://github.com/Pantonius/TagPag.

A major challenge of our time is reducing disparities in access to and effective use of digital technologies, with recent discussions highlighting the role of AI in exacerbating the digital divide. We examine user characteristics that predict usage of the AI-powered conversational agent ChatGPT. We combine behavioral and survey data in a web tracked sample of N = 1376 German citizens to investigate differences in ChatGPT activity (usage, visits, and adoption) during the first 11 months from the launch of the service (November 30, 2022). Guided by a model of technology acceptance (UTAUT-2), we examine the role of socio-demographics commonly associated with the digital divide in ChatGPT activity and explore further socio-political attributes identified via stability selection in Lasso regressions. We confirm that lower age and higher education affect ChatGPT usage, but do not find that gender or income do. We find full-time employment and more children to be barriers to ChatGPT activity. Using a variety of social media was positively associated with ChatGPT activity. In terms of political variables, political knowledge and political self-efficacy as well as some political behaviors such as voting, debating political issues online and offline and political action online were all associated with ChatGPT activity, with online political debating and political self-efficacy negatively so. Finally, need for cognition and communication skills such as writing, attending meetings, or giving presentations, were also associated with ChatGPT engagement, though chairing/organizing meetings was negatively associated. Our research informs efforts to address digital disparities and promote digital literacy among underserved populations by presenting implications, recommendations, and discussions on ethical and social issues of our findings.

Carbon footprint information via labels has raised interest as a tool to encourage pro-environmental behavior. We propose cognitive alternatives to the environmental status quo (Environmental cognitive alternatives; ECAs), the ability to imagine what a sustainable relationship with nature could look like, to improve the effectiveness of carbon labels. Using a discrete choice experiment with intervention and control groups, we investigate the effect of ECAs on low emission labeled, sustainable choices in a grocery shopping context. German participants (N = 150) were randomly assigned to three groups, activating either cognitive alternatives of a positive relationship with nature, or perceived environmental threat (PET), or nothing in a full control group. In the ECAs activation group, participants chose options with lower carbon emissions compared to the other two groups, and had stronger preferences on rating scales for these options. In the PET activation group, participants also had stronger preferences on rating scales than the control group, but this effect was not found for the choice of options. Activating ECAs might be a promising intervention for promoting sustainable choices, and carbon labeling could be helpful when paired with interventions that activate ECAs.

As citizens increasingly encounter political information in digital environments, understanding whether this engagement shapes their policy views has become a central concern. Drawing on dual-process theories of persuasion, we argue that motivational activation is an enabling condition for policy support change in high-choice online environments. We test this in a three-wave field experiment with German participants (n = 791) across three policy topics (basic child support, renewable energy transition, cannabis legalization), in which participants were randomly assigned to a control group, and two encouragement conditions: a verbal encouragement, or a monetary incentive tied to a knowledge test. Browsing behavior was passively tracked via digital trace data over a 20-hour window. We find that self-directed online information search produced changes in policy support for child support and cannabis legalization but not for the energy transition, with monetary incentives producing significant effects rather than verbal prompts. We discuss motivational salience, issue malleability, and search-environment quality as joint conditions under which political information engagement can produce detectable changes in policy support.

The revision histories of Wikipedia articles are a rich source of data about the interactions of editors with each other and with the content, yet they are not straightforward to mine or understand. We describe two tools for visual analytics that support this effort: (i) An interactive browser extension to study word authorship, age, and conflict dynamics, which provides an overlay on live Wikipedia articles; and (ii) a novel interactive Jupyter Notebook package that allows us to run analyses of editorial dynamics outof-the-box and is easily modifiable. Both leverage live data for any article on demand from several Web APIs, centering on our own WikiWho service, providing the most accurate mining of live word-level changes currently available. We show how these tools enable the exploration of the survival of content, productivity of editors, conflict dynamics, and other metrics through low-barrier interfaces while providing the opportunity for more quantitative investigations via access to the notebooks' underlying data structures.

Researchers rely on academic web search engines to find scientific sources, but search engine mechanisms may selectively present content that aligns with biases embedded in the queries. This study examines whether confirmation-biased queries prompted into Google Scholar and Semantic Scholar will yield skewed results. Six queries (topics across health and technology domains such as "vaccines" or "internet use") were analyzed for disparities in search results. We confirm that biased queries (targeting "benefits" or "risks") affect search results in line with the bias, with technology-related queries displaying more significant disparities. Overall, Semantic Scholar exhibited fewer disparities than Google Scholar. Topics rated as more polarizing did not consistently show more skewed results. Academic search results that perpetuate confirmation bias have strong implications for both researchers and citizens searching for evidence. More research is needed to explore how scientific inquiry and academic search engines interact.

This article evaluates the quality of data collection in individual-level desktop information tracking used in the social sciences and shows that the existing approaches face sampling issues, validity issues due to the lack of content-level data and their disregard of the variety of devices and long-tail consumption patterns as well as transparency and privacy issues. To overcome some of these problems, the article introduces a new academic tracking solution, WebTrack, an open source tracking tool maintained by a major European research institution. The design logic, the interfaces and the backend requirements for WebTrack, followed by a detailed examination of strengths and weaknesses of the tool, are discussed. Finally, using data from 1185 participants, the article empirically illustrates how an improvement in the data collection through WebTrack leads to new innovative shifts in the processing of tracking data. As WebTrack allows collecting the content people are exposed to on more than classical news platforms, we can strongly improve the detection of politics-related information consumption in tracking data with the application of automated content analysis compared to traditional approaches that rely on the list-based identification of news.

Anecdotal evidence suggests that the surge of populism and subsequent political polarization might make voters' political preferences more detectable from digital trace data. This potential scenario could expose voters to the risk of being targeted and easily influenced by political actors. This study investigates the linkage between over 19,000,000 website visits, tracked from 1,003 users in Germany, and their survey responses to explore whether website choices can accurately predict political attitudes across five dimensions: Immigration, democracy, issues (such as climate and the European Union), populism, and trust. Our findings indicate a limited ability to identify political attitudes from individuals' website visits. Our most effective machine learning algorithm predicted interest in politics and attitudes toward democracy but with dependency on model parameters. Although website categories exhibited suggestive patterns, they only marginally distinguished between individuals with anti- or pro-immigration attitudes, as well as those with populist or mainstream attitudes. This further confirm the reliability of surveys in measuring attitudes compared to digital trace data and, from a normative perspective, suggests that the potential to extract sensitive political information from online behavioral data, which could be utilized for microtargeting, remains limited.

Implicit and explicit gender biases in media representations of individuals have long existed. Women are less likely to be represented in gender-neutral media content (representation bias), and their face-to-body ratio in images is often lower (face-ism bias). In this article, we look at representativeness and face-ism in search engine image results. We systematically queried four search engines (Google, Bing, Baidu, Yandex) from three locations, using two browsers and in two waves, with gender-neutral (person, intelligent person) and gendered (woman, intelligent woman, man, intelligent man) terminology, accessing the top 100 image results. We employed automatic identification for the individual’s gender expression (female/male) and the calculation of the face-to-body ratio of individuals depicted. We find that, as in other forms of media, search engine images perpetuate biases to the detriment of women, confirming the existence of the representation and face-ism biases. In-depth algorithmic debiasing with a specific focus on gender bias is overdue.

The revision histories of Wikipedia articles are a rich source of data about the interactions of editors with each other and with the content, yet they are not straightforward to mine or understand. We describe two tools for visual analytics that support this effort: (i) An interactive browser extension to study word authorship, age, and conflict dynamics, which provides an overlay on live Wikipedia articles; and (ii) a novel interactive Jupyter Notebook package that allows us to run analyses of editorial dynamics out-of-the-box and is easily modifiable. Both leverage live data for any article on demand from several Web APIs, centering on our own WikiWho service, providing the most accurate mining of live word-level changes currently available. We show how these tools enable the exploration of the survival of content, productivity of editors, conflict dynamics, and other metrics through low-barrier interfaces while providing the opportunity for more quantitative investigations via access to the notebooks’ underlying data structures.

While individuals’ trust in search engine results is well-supported, little is known about their preferences when selecting news. We use web-tracked behavioral data across a 2-month period (280 participants) and we analyze three competing factors, two algorithmic (ranking and representativeness) and one psychological (familiarity), that could influence the selection of search results. We use news engagement as a proxy for familiarity and investigate news articles presented on Google search pages ( n = 1221). We find a significant effect of algorithmic factors but not of familiarity. We find that ranking plays a lesser role for news compared to non-news, suggesting a more careful decision-making process. We confirm that Google Search drives individuals to unfamiliar sources, and find that it increases the diversity of the political audience of news sources. We tackle the challenge of measuring social science theories in contexts shaped by algorithms, demonstrating their leverage over the behaviors of individuals.

The conspiracy theory that the US 2020 presidential election was fraudulent - the Big Lie - remained a prominent part of the media agenda months after the election. Whether and how search engines prioritized news stories that sought to thoroughly debunk the claims, provide a simple negation, or support the conspiracy is crucial for understanding information exposure on the topic. We investigate how search engines provided news on this conspiracy by conducting a large-scale algorithm audit evaluating differences between three search engines (Google, DuckDuckGo, and Bing), across three locations (Ohio, California, and the UK), and using eleven search queries. Results show that simply denying the conspiracy is the largest debunking strategy across all search engines. While Google has a strong mainstreaming effect on articles explicitly focused on the Big Lie - providing thorough debunks and alternative explanations - DuckDuckGo and Bing display, depending on the location, a large share of articles either supporting the conspiracy or failing to debunk it. Lastly, we find that niche ideologically driven search queries (e.g., "sharpie marker ballots Arizona") do not lead to more conspiracy-supportive material. Instead, content supporting the conspiracy is largely a product of broader ideology-agnostic search queries (e.g., "voter fraud 2020").

Researchers in the political and social sciences often rely on classification models to analyze trends in information consumption by examining browsing histories of millions of webpages. Automated scalable methods are necessary due to the impracticality of manual labeling. In this paper, we model the detection of topic-related content as a binary classification task and compare the accuracy of fine-tuned pre-trained encoder models against in-context learning strategies. Using only a few hundred annotated data points per topic, we detect content related to three German policies in a database of scraped webpages. We compare multilingual and monolingual models, as well as zero and few-shot approaches, and investigate the impact of negative sampling strategies and the combination of URL content-based features. Our results show that a small sample of annotated data is sufficient to train an effective classifier. Fine-tuning encoder-based models yields better results than in-context learning. Classifiers using both URL content-based features perform best, while using URLs alone provides adequate results when content is unavailable.

Given that Facebook is still the most widely used social networking site in the world, its influence on democratic processes is under constant scrutiny. Academics have put a special focus on Facebook's role in inhibiting or enhancing citizens' news exposure. Recent studies using digital behavioral data have analyzed the prevalence and effects of "Facebook news referrals". Using a web tracking tool that captures general browsing behavior as well as public posts seen on Facebook, this paper lays the groundwork for the field by assessing the validity of previously proposed operationalizations. We validate news referrals by investigating whether different measures actually reflect exposure to a news URL a user saw on Facebook. We furthermore assess the effects of news referrals on central outcomes in extant literature, contingent on different operationalizations. The results show that the most precise measure of news referrals are referral IDs attached as parameters to news URLs by Facebook. Still, the substantive findings are broadly comparable across approaches, lending further credibility to published research, despite measurement errors. The paper demonstrates the need for academics to constantly innovate in order to measure citizens' online behavior in an ecologically valid manner.