The study, published on February 9, 2026, centered on an online experiment in which adults in the United States were asked to use ChatGPT to evaluate climate-related claims. Rather than simply asking participants whether they trusted the chatbot, the researchers designed a two-track investigation. On one track, they measured how credible participants found both the original claims and the AI’s responses to them, using self-reported credibility judgments. On the other track, they applied computational text analysis to the actual conversation logs, dissecting the language of ChatGPT’s replies with a battery of language-model-based classifiers and link-level metadata.

That second track is what distinguishes the work from much of the earlier literature on AI and misinformation. Instead of treating a chatbot’s answer as an undifferentiated block of text, Tsang and Wang decomposed it into measurable linguistic properties: the valence, or emotional positivity, of the response; its formality; the degree of bias it expressed; how recent and current the information appeared; the presence of authority cues, such as references to experts or institutions; and its semantic richness, a measure of how much substantive, varied content the reply contained. Each of these dimensions was scored systematically across the corpus of conversations, allowing the team to test which features of AI language actually moved the needle on human perception.

The headline result is, in many ways, a humbling one for anyone who imagined that the mere authority of an AI system would sweep aside doubt. Credibility perceptions, the authors found, were driven primarily by individual differences among the people doing the asking. Participants who already accepted the scientific consensus on climate change rated both the claims and the chatbot’s responses as more credible. So did those who paid closer attention to climate news, those who had prior familiarity with ChatGPT, and, notably, younger participants. In other words, the audience arrived with predispositions, and those predispositions did much of the work in shaping what the conversation ultimately meant to them.

This pattern echoes a long-standing finding in communication research: people do not process information as blank slates. Motivated reasoning, selective attention, and prior knowledge all filter how a message lands. What is new here is that the same dynamics appear to govern encounters with a large language model, a technology often assumed to carry a kind of neutral, algorithmic authority. If a chatbot’s answer is filtered through the same psychological machinery as a cable news segment or a social media post, then the dream of AI as an automatic corrective to misinformation looks considerably more complicated.

Yet the study did find that the texture of ChatGPT’s language mattered, once individual differences were taken into account. Responses with positive valence and greater semantic richness were associated with higher perceived credibility. This suggests that when the chatbot offered answers that were substantively rich, varied, and framed in constructive rather than negative terms, participants found them more convincing. There is an intuitive logic to this: a reply that engages deeply with the substance of a claim, drawing on multiple facets of the evidence, gives a reader more to evaluate and more reason to feel informed, while a thin or dismissive reply may leave uncertainty unresolved.

The role of authority cues proved more surprising. When ChatGPT’s responses contained signals of authority, such as appeals to expert consensus or institutional sources, participants’ climate attitudes shifted slightly toward less extreme positions. The effect was modest, but its direction is intriguing. Rather than hardening views, the invocation of authoritative framing appeared to nudge people toward the middle ground. For a technology that is frequently criticized for hallucinating sources or overstating certainty, this finding hints that well-calibrated authority cues, deployed transparently, could play a constructive role in depolarizing contested topics.

The authors frame these results as support for a particular model of responsible climate communication: transparent, unbiased, and audience-tailored engagement. The emphasis on tailoring is significant. Because credibility judgments depended so heavily on who the user was, a one-size-fits-all chatbot response is unlikely to land the same way across a diverse public. A climate-attentive younger user with prior chatbot experience may find a detailed, evidence-dense reply persuasive, while a skeptical older user encountering the same text may remain unmoved. Designing conversational AI that adapts to these differences, without pandering or distorting the underlying science, emerges from the study as a central challenge for the field.

The implications extend beyond climate. Misinformation about vaccines, elections, and public health increasingly collides with AI assistants positioned as arbiters of fact. The PLOS Climate experiment offers a template for studying those collisions rigorously: capture the actual conversation logs, quantify their linguistic properties with computational classifiers, and link those properties to measured changes in human judgment. As large language models are embedded into search engines, messaging platforms, and voice assistants, this kind of evidence becomes essential for understanding not just what these systems say, but what their ways of saying it do to the people listening.

What the study ultimately delivers is a measured dose of both caution and optimism. Caution, because the chatbot did not override prior belief; the psychological characteristics of users remained the dominant force shaping credibility judgments, meaning AI alone cannot close the gap between scientific consensus and public understanding. Optimism, because specific, designable features of AI language, including positive framing, semantic depth, and calibrated authority cues, showed measurable associations with more credible and less extreme outcomes. The conversation between humans and machines about climate change, it turns out, is a genuine conversation, one whose outcome depends on both sides of the exchange, and one that communicators, developers, and researchers now have better tools to study and to shape.

Subject of Research: How large language model responses influence public credibility judgments of climate-related claims

Article Title: Countering climate misinformation with large language models: Evidence from ChatGPT

Article References: Tsang, S. J., & Wang, D. (2026). Countering climate misinformation with large language models: Evidence from ChatGPT. PLOS Climate, 5(9), e0000930. https://doi.org/10.1371/journal.pclm.0000930

Image Credits: AI Generated

DOI: 10.1371/journal.pclm.0000930

Keywords: ChatGPT, climate misinformation, large language models, PLOS Climate, science communication, credibility, public opinion, computational text analysis, climate attitudes, AI chatbots, misinformation, audience effects

The study examined English-language newspaper sources from the United States and Europe up to December 31, 2024, assembling a final corpus of 43,027 articles, of which 9,684 came from the United States and 33,343 from Europe. The authors combined computational text analysis with interpretive examination of the corpus, a methodological pairing that allowed them to detect broad patterns of representation across a very large body of coverage while also reading individual articles closely for nuance. The scale of the dataset matters: with tens of thousands of articles spanning years of reporting, recurring patterns of framing and sourcing become statistically visible in ways that smaller samples cannot reveal.

What emerged from the analysis is what the researchers describe as a striking hierarchy of visibility and authority. Massimo Ragnedda, professor of media and communication at the University of Sharjah and the study’s senior author, said the findings reveal that the United States is predominantly positioned as a technological innovator and leader, while Europe is cast as a regulator and normative evaluator. In this symbolic geography, America supplies the vision of the AI future and Europe supplies the moral and legal guardrails, a division of narrative labor that mirrors but also reinforces real-world concentrations of technological and institutional power.

China’s position in the coverage is more complicated and more ambivalent. Although Beijing is a leading AI power alongside the United States, the authors found that it appears as a hyper-visible but ambivalent rival, associated in the reporting with surveillance, geopolitical competition, and strategic threat. In other words, China is never ignored—its AI ambitions receive enormous attention—but that attention is filtered through frames of suspicion and rivalry rather than through the celebratory innovation narratives routinely applied to American firms and researchers. Visibility, the study suggests, is not the same as favorable representation.

Most striking of all is who is missing. Countries in Africa and Latin America are largely absent from the narrative or positioned primarily as sites of implementation, risk, or developmental need. They appear, when they appear at all, not as authors of technological futures but as territories where other people’s technologies are deployed, where risks materialize, or where development gaps must be closed. The result is a global AI story told almost entirely from a narrow band of the English-speaking world, in which the majority of humanity figures as a backdrop rather than as a protagonist.

The researchers conceptualize this pattern as discursive stratification, a term meant to capture an unequal distribution of narrative authority in which some actors are repeatedly positioned as legitimate authors and interpreters of technological futures, while others appear mainly as recipients, sites of implementation, or objects of technological change. The metaphor of stratification is deliberate: like geological layers, the media coverage sorts nations into tiers of narrative standing, with the top layers granted the authority to imagine and define what artificial intelligence will become, and the lower layers assigned passive or problematic roles in that unfolding story.

This framing draws on recent political-economy scholarship that views AI not simply as an unprecedented technological innovation with far-reaching consequences for human life, but as a sphere beset by rivalries over power distribution, geopolitical competition, and infrastructural concentration. The study extends that perspective into the realm of communication, treating AI as an inequality regime in which communicative power, visibility, value, and opportunity are redistributed through unequal linguistic structures of control. On this account, the struggle over AI is fought not only in laboratories, data centers, and regulatory agencies, but in the language used to describe the technology in the world’s most influential newspapers.

Central to the argument is a redefinition of what news does. The researchers conceptualize news not as a mirror of technological change, but as a narrative infrastructure—a discursive and institutional system through which AI becomes publicly legible and through which some actors are positioned as innovators, regulators, threats, or absences. Just as physical infrastructure such as undersea cables and semiconductor supply chains determines who can access and deploy the technology, this narrative infrastructure determines who can meaningfully participate in the public conversation about it. The stories repeatedly told about AI, the authors argue, help shape perceptions of who leads, who regulates, who threatens, who benefits, and who remains largely unheard.

The implications extend well beyond media criticism. Ragnedda notes that AI does not simply determine what we talk about; it also helps determine who is recognized as having the authority to speak about the technology and whose futures are considered worth imagining. If the public understanding of artificial intelligence is filtered through a narrow set of voices, then policy debates, investment decisions, and regulatory frameworks built on that understanding risk inheriting the same blind spots. The absence of African and Latin American perspectives, for example, means that questions of digital colonialism, labor exploitation in data annotation, and unequal access to AI benefits may be systematically underexplored in the outlets that most shape elite opinion.

The authors are nonetheless optimistic about the study’s contribution to the field. They write that it advances communication scholarship beyond descriptive analyses of representation toward a structural account of narrative authority in global media systems, showing how framing, sourcing hierarchies, and thematic emphasis combine to produce the symbolic geography of AI. Ragnedda distills the stakes into a single question: the issue is not only who builds AI, but who gets to define what AI means, and whose voice is missing. As artificial intelligence continues to reshape economies, security doctrines, and daily life, the study suggests that paying attention to the chain of dominance and dependence embedded in everyday news coverage may be as important as worrying about the technology itself.

Subject of Research: Discursive stratification and narrative authority in Western English-language news coverage of artificial intelligence

Article Title: Though seen as a threat to humanity, AI reporting in the West reflects a clear chain of dominance and dependence

Article References: Though seen as a threat to humanity, AI reporting in the West reflects a clear chain of dominance and dependence. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, media narratives, discursive stratification, narrative authority, news media, United States, Europe, China, Global South, computational text analysis, communication research, geopolitical competition

The study’s strength lies in its design. Rather than running a small laboratory experiment, the team exploited what statisticians call a natural experiment: Change.org introduced its ‘write with AI’ feature in some countries before others, creating a clean before-and-after boundary. The researchers collected roughly 1.5 million petitions and applied a difference-in-differences analysis, a causal inference technique that compares changes over time between groups that received the AI tool and groups that did not. This approach, widely used in economics and built on methods formalized by Callaway and Sant’Anna for settings with multiple time periods, allows researchers to isolate the effect of the intervention from broader trends that would have happened anyway. Because the rollout timing was determined by the platform rather than by users’ characteristics, the comparison approximates the rigor of a randomized controlled trial at population scale.

The first major finding concerns the texture of the text itself. After the AI tool became available, the lexical features of petitions shifted measurably at the platform level. Petitions grew longer and their language changed in ways consistent with the stylistic fingerprints of large language models: smoother phrasing, more standardized vocabulary, and altered readability characteristics. These are not trivial cosmetic shifts. Lexical measures such as vocabulary diversity and readability scores have long been used in computational text analysis to characterize how writing differs across authors, genres, and eras, and the fact that a single platform feature could move these metrics across an entire ecosystem of user-generated content is striking. In effect, the introduction of the tool bent the collective writing style of a global advocacy community.

But the second major finding is where the story turns. None of these stylistic improvements translated into better petition outcomes. Petitions written with access to the AI assistant did not attract more signatures, achieve their goals more often, or otherwise perform better than petitions written without it. For a tool presumably offered to help advocates succeed, that null result carries real weight. It echoes a growing body of evidence that generative AI can boost individual productivity in controlled settings, as shown in the well-known 2023 Science experiment by Noy and Zhang on professional writing tasks, while failing to deliver equivalent gains when success depends on persuading other humans in a noisy, competitive environment.

The researchers did not stop at the platform-level analysis. A persistent worry with observational studies is that the people who choose to use a new feature differ systematically from those who do not, contaminating any comparison. To address this, the team conducted a separate analysis of repeat petition writers: 4,611 users who had written at least one petition before the AI tool was introduced and another one after. This within-person comparison holds individual ability, motivation, and topic preferences roughly constant, isolating the effect of AI access on the same writers over time. The pattern that emerged was telling. On average, these returning writers produced longer petitions when they had access to AI, yet their second petitions fared worse in terms of outcomes than their earlier work. The very users most likely to benefit from accumulated platform experience saw no advantage, and by some measures a disadvantage, from the writing assistant.

Perhaps the most consequential finding is the increase in homogeneity. Using dynamic difference-in-differences estimates, the researchers showed that platform-wide homogeneity among petitions rose after the AI tool’s introduction. In plain terms, petitions began to sound more like one another. This result aligns with a rapidly expanding literature on what some researchers call generative monoculture. A 2024 study in Science Advances by Doshi and Hauser found that generative AI enhanced individual creativity while reducing the collective diversity of novel content. Work presented at the 2025 CHI Conference by Agarwal, Naaman, and Vashistha showed that AI writing suggestions can homogenize text toward Western stylistic conventions and erode cultural nuance. Padmakumar and He asked directly whether writing with language models reduces content diversity and found evidence that it does. The Change.org study extends these laboratory and small-scale findings to a genuine production environment used by millions.

Why should homogenization matter for something as seemingly benign as petition language? Advocacy platforms depend on differentiation. A petition competes for attention in a crowded feed, and its distinctive voice, urgency, and local specificity are part of what makes a reader stop and sign. When an algorithmic assistant nudges thousands of writers toward the same polished register, the collective signal degrades even if each individual text improves. Researchers have raised parallel concerns in other domains: a 2026 Nature analysis by Hao and colleagues found that AI tools expanded individual scientists’ output while contracting the focus of science as a whole, and work on Stack Overflow suggested that large language models may threaten the value of community-generated digital public goods. The pattern that emerges across these studies is consistent: individual-level gains, community-level costs.

The null effect on outcomes also invites a deeper question about what petition success actually depends on. Prior research on e-petitions, including studies of linguistic cues and multidimensional time-series predictors of success, has suggested that factors such as topic salience, timing, social networks, and media coverage often dwarf the influence of prose quality. A beautifully written petition about an issue nobody cares about will still fail; a clumsy petition riding a wave of public outrage may succeed regardless. The Change.org results are consistent with this view. If outcomes are driven primarily by the underlying demand for a cause rather than by the craft of the appeal, then a tool that polishes craft without touching demand should indeed leave outcomes unchanged. The study cannot rule out subtler possibilities, such as offsetting effects in which longer, more polished text helps in some respects and hurts in others, but the headline conclusion stands: no measurable outcome benefit at scale.

There are also social and perceptual dimensions that the authors situate within a broader research program. Studies have shown that human heuristics for detecting AI-generated language are unreliable, that there can be a social evaluation penalty for using AI, and that perceptions of AI-mediated communication affect trustworthiness and perceived authenticity. For petition signers, the suspicion that a heartfelt appeal was machine-generated could matter as much as the text itself. Related work on the ‘AI ghostwriter effect’ shows that users often fail to feel ownership of AI-generated text while still claiming authorship, raising questions about authenticity in civic expression. A platform that encourages AI-assisted advocacy may therefore be trading not only diversity of style but also the perceived sincerity that makes grassroots appeals compelling.

The practical implications reach well beyond one platform. Companies including Amazon and Meta have embedded generative writing tools into seller listings and social posts, making the Change.org experiment a preview of dynamics likely to unfold across the internet. The study’s message for platform designers is that embedding AI writing assistance is not a neutral upgrade: it reshapes the linguistic character of an entire content ecosystem and can introduce unintended consequences such as homogenization without delivering the hoped-for performance gains. For researchers, the work demonstrates the power of natural experiments and difference-in-differences designs for evaluating AI interventions in the wild, complementing randomized trials that capture only narrow populations and short horizons. And for the millions of people who turn to online platforms to press for change, the findings carry a quietly empowering lesson: the cause, the community, and the timing still matter more than the polish of the prose. The data and code behind the analysis have been released through the Open Science Framework, allowing other researchers to reproduce the results and probe further into how generative AI is quietly rewriting the way the public speaks online.

Subject of Research: Causal effects of an in-platform AI writing tool on petition content and outcomes on Change.org

Article Title: Introducing AI to an online petition platform changed outputs but not outcomes

Article References: Corpus, I., Gilbert, E., Koenecke, A., & Naaman, M. (2026). Introducing AI to an online petition platform changed outputs but not outcomes. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02580-8

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02580-8

Keywords: generative AI, Change.org, online petitions, difference-in-differences, content homogenization, large language models, computational text analysis, Nature Human Behaviour, platform design, AI writing tools, natural experiment, digital activism