This research note explores the potential of artificial intelligence (AI) to support scalable training in science communication. Training scientists in science communication is increasingly being recognized as important, yet existing programs face scalability challenges and often prioritize one-way knowledge dissemination over dialogue. This study explores whether these challenges can be mitigated by integrating AI into the training process. We developed and empirically evaluated DiaLogic – an AI-based communication simulator. In all, 37 students engaged in two simulated dialogues and received AI-generated feedback between sessions. The results showed significant improvements in students’ performance. These findings highlight the promise of AI-based tools to support scalable science communication training.
Applying the influence of presumed media influence model into an autonomous vehicle (AV) context, this study examines how media viewers’ attention to content about risks and benefits of AVs affects their presumptions of influence of the content on others, perceptions of trustworthiness of the technology, and intention to ride in AVs. Through cross-lagged analyses of two-wave panel data (n wave1 = 1,306; n wave2 = 650), this study found causal evidence for the projection effect and impersonal impact. Notably, this study also found causal evidence for the consonance effect: viewers perceiving trustworthiness of AVs after developing intention to ride in AVs.
Despite increased scholarly attention to scientists’ role in policymaking, little is known about how scientists perceive and enact that role. We examine American scientists’ perspectives on and engagement with policymakers, recruitment into these roles, and possible influences of political ideology on these phenomena. We find that although both technical and advocacy-oriented communication were popular, scientists were more likely to engage in advocacy—possibly due to limited recruitment into technical roles relative to advocacy. Left-leaning scientists also endorsed a broader range of communication activities with policymakers, but political differences were less evident with respect to communication in practice.
Science storytelling can make evidence intelligible, memorable, and culturally resonant, yet persuasive narratives may still misrepresent how scientific knowledge is produced. This commentary argues that communicators need a standard beyond narrative trust: narrative integrity. The concept evaluates whether a story preserves science’s uncertainty, multicausality, collective labor, provisional conclusions, and capacity for revision. Drawing on research on transportation, framing, anecdotal bias, and processing fluency, we explain how narrative form can outrun evidence even when individual claims are accurate. We then propose six principles for designing stories that remain engaging while representing scientific inquiry honestly, transparently, and responsibly to diverse public audiences.
This study reconceptualizes information insufficiency as a goal-contingent construct within the Risk Information Seeking and Processing model. In multifaceted risk contexts, individuals confront health threats alongside challenges to deep-seated values and beliefs, producing different judgmental demands within a single risk event. Survey data from 858 Hong Kong adults show two distinct forms of information insufficiency, one oriented toward accurate safety judgments and another toward defending deep-seated values and beliefs. Each form is associated with distinct patterns of information seeking and processing. This study demonstrates how multiple goal-based information insufficiencies operate within the same individual in response to different judgmental demands.
Peer review serves as the primary “gatekeeping” mechanism in science communication, filtering validated knowledge from noise before it reaches the public sphere. This commentary argues that integrating Generative AI into this process acts as a structural disruptor, threatening to erode the “seal of quality” that underpins public trust in science. By analyzing the systemic feedback loop between AI-generated submissions (“supply-side flood”) and AI-assisted reviews (“demand-side automation”), we demonstrate how the ecosystem risks accelerating the “degradation of work” and creating “monocultures of knowing.” Using strategic foresight, we outline four scenarios for the future of scientific publishing, ranging from a collapse of trust to algorithmic governance. We conclude that to avoid a crisis of credibility, the community must pivot from “Algorithmic Aggregation” to “Contextual Stewardship.” We propose that human reviewers remain indispensable not for processing speed but for four uniquely human competencies: verifying ground truth, arbitrating ethics, communicating uncertainty, and curating the paradigm-shifting anomalies that automated systems reject as noise but are actually the driving force behind the progress.
Social media platforms have become important sites of science communication, with individual scholars increasingly creating content to engage public audiences with scientific knowledge. However, academic institutions have largely neglected to regard such science communication work through content creation as legitimate scholarly labour, instead continuing to focus primarily on the scholar as a producer of scientific knowledge. This commentary introduces the figure of the ‘scholar-creator’: scholars who use social media to translate, contextualise, and narrate scientific knowledge. It argues that scholar-creators should be formally recognised and supported, as they perform crucial work in making scientific knowledge matter outside the academy.
Previous research on climate change consensus messaging has mostly taken place in controlled lab settings. In this field experiment, we engaged U.S. residents (N = 158) in brief doorstep conversations on climate change. Research assistants read a script about the scientific consensus (treatment) or basic facts about climate change (control) and then provided participants with a magnet containing the same information. The consensus message had a significant positive effect on consensus estimates (beta = 0.45) and belief in climate change (beta = 0.41), but not on other downstream attitudes or behavior. These results mostly align with theory and have implications for consensus messaging.
The current research observed historical public discourse surrounding the issue of open science given its contemporary salience across scientific disciplines. The Twitter API was used to collect the population of 1,723,169 open science-related tweets published from Twitter’s inception through September 2022. A latent factor Dirichlet multinomial mixture (LF-DMM) model was used to analyze textual tweet content. Stepwise segmented compositional regression identified evolutionary trends and revolutionary shifts in the discourse over time. Seven themes emerged, and results demonstrated nuanced trajectories in the representation and prevalence of themes occurring in public discourse about open science over time.
Existing research on intermedia agenda-setting within the hybrid media system has primarily focused on political communication. This study extends prior work by examining how a scientific controversy circulated across legacy and social media platforms. Specifically, it investigates the debate over hydroxychloroquine (HCQ) as a potential COVID-19 treatment and explores whether, and how, legacy media and social media set each other’s agendas. We conducted a cross-media and cross-platform analysis using data from legacy media (broadcast, cable news, and national newspapers; N = 2,276) and social media (Twitter, N = 416,087; Facebook, N = 28,566). Combining time-series analysis with large language model-assisted framing analysis, the results reveal a complex relationship across platforms: overall, coverage of HCQ did not exhibit consistent agenda-setting effects between legacy and social media. However, framing patterns diverged significantly. Legacy media emphasized Conflict and Public-Risk frames, while social media discourse was dominated by polarized Conflict and Economic-Consequences frames. These findings contribute to the literature on networked agenda-setting in the hybrid media system by providing empirical evidence from science communication, extending the framework beyond its predominant focus on political communication.
This preregistered study tests whether a one-item self-categorization question can reproduce Sch & auml;fer et al.'s four-segment typology of science communication audiences: "Sciencephiles," "Critically Interested," "Passive Supporters," and "Disengaged." Using survey data from 3,272 visitors to CERN's Science Gateway, we compare classifications from the single-item measure with those from the established 10-item instrument. Results show high precision but moderate recall in identifying Sciencephiles, but poor accuracy for other segments. Overall, the one-item measure is not interchangeable with the multi-item instrument in this setting; it may serve as a brief screener for Sciencephiles when brevity is required, and false positives must be minimized.
This guest editorial commentary introduces a special issue of Science Communication on artificial intelligence's (AI) dual role as both a tool for and a topic of science communication. It argues that this dual role redistributes communicative and epistemic agency in ways that unsettle established assumptions about authorship, credibility, and trust. The commentary reviews seven studies spanning practitioner use, audience cognition, and misinformation correction, then develops two theoretical questions: how AI reshapes the social infrastructure of trust in science, and what science communication means when scientific discovery itself is partially delegated to AI.
To address the challenge of scalability in fact-checking, we report two parallel experiments examining which message features may boost communicative engagement with AI-facilitated observed correction of misinformation related to human papillomavirus (HPV) vaccination. Study 1 (N = 1,789) found that expert endorsement enhanced AI-generated visual exemplars' effects on engagement. Study 2 (N = 1,765) showed that the advantage of adopting an empathetic tone for AI (vs. human) fact-checkers was dependent on participants' baseline perceived anthropomorphism in a nonlinear pattern. These findings highlight the importance of strategically selecting motivationally relevant messages and design features to improve the scalability of AI-powered observed correction.
This study investigates whether hostile media perception extends to AI chatbots when correcting geopolitically framed climate misinformation. A 2 (correction source: foreign vs. domestic AI) & times; 2 (evidence disclosure: present vs. absent) + 1 (control) experiment was conducted among Chinese participants (N = 998). Results show that AI corrections effectively reduced misinformation beliefs. However, foreign AI corrections were perceived as more biased and were less effective than domestic AI corrections. Disclosing the evidence underlying AI corrections reduced perceived bias and improved correction effectiveness. The study also examined how national narcissism and belief in machine heuristics moderated these effects.
As social media becomes central to environmental communication, understanding what drives widespread sharing is essential. This scoping review analyzes 82 peer-reviewed studies examining whether emotional environmental content on social media contributes to virality. It maps research trends, methodologies, and theoretical approaches used to study emotion in online environmental communication. Findings show a sharp increase in publications since 2020, with quantitative computational analyses of Twitter (now X) dominating the literature. Most studies relied on dimensional or sentiment-based emotion measures rather than discrete emotions. Theory-driven and experimental studies were rare. Overall, the review highlights gaps in theory, methodological diversity, and platform coverage.
Despite the pervasiveness of misinformation, it remains unclear how repeated exposure to falsity shapes veracity judgments. We conducted an online experiment (N = 499) in which participants were exposed to false and true health and science information presented either consecutively or in a balanced mix. Consecutive exposure to false information increased the likelihood of judging subsequent information as false while simultaneously improving judgment accuracy and confidence-weighted accuracy. These findings suggest that shifts in judgment orientation induced by repeated exposure to false information reflect adaptive epistemic vigilance, highlighting how informational contexts shape veracity judgments under conditions of triggered evaluation.
Diffusion of innovations theory proposes that opinion leaders can enhance the success of science communication campaigns. Such strategies hinge on an assumption that target communities include well-connected and trusted members who need only be identified and recruited. However, sparse or difficult to engage communication networks may not meet this assumption, making opinion leader-based interventions ineffective or impossible. This project reviews two studies involving rural populations where attempts to identify opinion leaders failed. We leverage these examples to highlight barriers to community-based campaigns, theoretical implications for diffusion scholarship, and suggested strategies for communication scholars and practitioners encountering sparse or fragmented networks.
This study investigates racial biases in AI-generated occupational images across models developed in the United States and China. Situated at the intersection of human-AI communication and postcolonial theory, we conceptualize generative AI as an active participant in science communication that shapes visual knowledge and racial representation within a global postcolonial order. Constructing a dataset of 9,600 images generated by four models (GPT-4o, Llama 3, Wanx2.0, and Wenxin 3.5), we examine three levels of racial biases-representational bias, positional bias, and racialized meaning bias-using a mixed-methods approach. Findings show that White individuals are overrepresented, granted spatial dominance, and encoded through aesthetic and symbolic conventions that racialize non-White bodies. We reveal how generative AI reproduces global racial hierarchies under algorithmic neutrality, advancing a cross-national auditing framework and contributing to decolonial science communication by foregrounding a human-centered AI perspective.
The present study empirically tests video production tactics for TikTok- and Instagram-like videos that can be employed to achieve specific, short-term communication objectives. Using the Strategic (Science) Communication as Planned Behavior (SCPB) model as a framework, a between-subjects experiment varies the race/ethnicity of the speaker, the tone of the video, and the inclusion of references to pop culture to examine how these tactics are related to curiosity, enjoyment, and motivations to follow similar content online. We use a representative sample of U.S. adults with oversampling of underrepresented STEM groups. Findings and implications are discussed.
In algorithmically curated environments where science videos reach audiences primarily through incidental exposure rather than active seeking, understanding viewers' perceived effects provides insights for communicating science effectively. We developed and validated a multidimensional scale measuring perceived effects of watching science videos through three sequential studies. Study 1 used an open-ended survey to generate items (N = 150), Study 2 employed a close-ended survey for factor identification (N = 309), and Study 3 tested criterion validity (N = 286). We identified five dimensions of perceived effects: information acquisition, practical application, enjoyment, time-passing, and social interaction. These dimensions show varied associations with content exposure and perceptions of science.
