Linguistic insights, media mechanisms and role of AI in dissemination and impact of disinformation

Dissemination of deceptive news designed to manipulate public opinion is not a new phenomenon in the western socio-political context. However, the dissemination of such information in the modern era has reached global proportions and unprecedented speed. This phenomenon has been driven by the pervasive use of digital platforms, which are increasingly becoming the primary sources of information for citizens. Although the prospect of a free information system without intermediaries may seem to epitomize the right to access information and the right to be adequately informed, events such as the US presidential election and the Brexit referendum have exposed the downside of what might be termed “information disorder” (Wardle and Derakhshan, 2017). This side has created a fertile ground for the proliferation of disinformation campaigns. The main obstacle to combating disinformation campaigns lies in their global nature. Through social media algorithms and automated bot networks, manipulative information often originates in one nation but reaches a global audience. Coordinated inauthentic behavior and state-sponsored efforts further complicate the problem by deliberately targeting social cohesion and public policy in other countries.

Because of the above, contrasting the dissemination of deceptive news has a crucial importance. The article, consistent with the strategies adopted by the EU to combat disinformation, focuses on disinformation campaigns within the digital ecosystem, where mainstream media can act as amplifiers. Using a multidisciplinary approach, this study traces the path of disinformation campaigns from their detection by linguistic cues of credibility to furtherance through the dissemination mechanisms, and finally, assessing their impact on the socio-political context.

In particular, the article investigates linguistic indicators of credibility through an examination of linguistic, lexical and semantic patterns that help identify disinformation. It identifies important markers of disinformation by analyzing textual components, such as titles, article content and emotional markers, and syntactic and semantic features, such as, for example, ambiguity, vagueness and the use of hedging words. The article also examines how structural components of social media, such as algorithm-driven engagement metrics and echo chambers, exacerbate public polarization and the spread of disinformation. Moreover, the threats to public trust, democracy and social cohesion posed by the increasing use of generative artificial intelligence (AI) to spread disinformation are also discussed in the article. Finally, the paper reflects on the impact of disinformation campaigns on public opinion and, consequently, on political decision-making processes, expanding the perspective toward a systemic understanding rather than a purely consequential one.

The aim of this study is to provide a comprehensive overview of the core components of disinformation and emphasize the point that successful countermeasures can be achieved through the concurrent promotion of platform accountability, media literacy and cross-sector collaboration.

The study has been carried out within the AI4Debunk [¹] project, founded by the European Commission, aiming at contrasting disinformation by providing citizens with a comprehensive set of fact-checking resources to navigate the digital media landscape more consciously and make informed decisions.

Before proceeding, it is essential to clarify that, within the context of this study, we will adopt the European Union’s definition of disinformation, which highlights its intentional and manipulative characteristic:

In recent years, linguistic analysis has significantly contributed to the recognition and prevention of disinformation campaigns. Although the phenomenon of disinformation presents a multifactorial complexity, the analysis of linguistic and lexical structures can help detect news created to mislead and manipulate public opinion. Different branches of linguistics, through various methodologies, have investigated the presence of specific linguistic indicators in deceptive news, which can be interpreted as “predictive deception cues” (Conroy et al., 2016).

A critical comparison of the literature that has focused on the linguistic analysis of unreliable textual news (articles and/or social network posts) will allow us to highlight those linguistic indicators that have proven to be statistically significant for the identification of deceptive news. It is essential to keep in mind that content indicators, but especially context indicators (Zhang et al., 2018), are characterized by high variability, as they rapidly evolve in concomitance with social changes and growing information saturation (Petrou et al., 2023). However, there are recurring linguistic patterns that can constitute a useful reference framework for reporting potentially unreliable news, even through automated systems. Nevertheless, such reports must always undergo a careful human evaluation to avoid hasty conclusions and ensure an accurate analysis.

The title is certainly one of the primary textual elements that is useful to examine when looking for linguistic spies. The change in the ways news is consumed, in particular the transition from traditional media to online platforms, has had a significant impact on reading habits. Today, reading information is characterized by greater speed and superficiality: users seem to have developed a growing tendency to read the title alone without further investigation of the content of the article (Reuters Institute for the Study of Journalism, 2024). This evolution has made the title an interesting text itself for analysis because it is knowingly used by the authors of unreliable content. Unlike those aimed at clickbait, typical of marketing, in the titles of deceptive news of a political or social nature, it is possible to find characteristics that reflect specific disinformation techniques, which are typical of the argumentative context in which they operate. Specifically, in this case, the title tends to be longer than that of authentic news, as it tries to condense the information necessary for a simple and easily understandable message without requiring further analysis. Consequently, the information density of disinformative titles is strategically designed to offer rapid and immediate access. Therefore, the title already includes everything the author wants the user to know, so the complete article does not usually provide more in-depth information (Horne and Adali, 2017; Revenko, 2020).

From a lexical point of view, disinformative titles are characterized by a small presence of function words and a higher concentration of keywords, mainly from the commonly used vocabulary. Moreover, disinformative titles have a significant frequency of names and verbal expressions, meaning a focus on individuals and actions, which allows creating a more direct connection with the reader and a sense of urgency. This structure is aligned with the elaboration likelihood model (ELM) (Petty and Cacioppo, 1986), a supplementary approach to persuasion that has been extensively studied in the area of persuasive communication. The ELM states that there are two ways to persuade someone: a central one that relies on a thorough and rational explanation of the message and a peripheral one that makes use of simple heuristics. Deceptive news tends to exploit the latter mechanism, playing on superficial and emotional signals to induce the reader to a rapid and unconsidered reaction. Another relevant element of analysis is the representativeness of the title compared to the content of the article. The studies by Zhang et al. (2018) and Carvalho et al. (2021) have highlighted that a title strictly faithful to the text of the article it refers to increases the probability that the news is authentic. Although representativeness is not the most decisive indicator of credibility, it obtained a significant relevance, suggesting that, in many cases, titles well correlated to the inner content can be used as a signal of the authenticity of the news itself.

Focusing on the texts of the articles or considering the news conveyed through a post, a highly relevant element useful to identify implausible news is its level of emotional intensity. In fact, as in the title, even in a larger portion of the text, Carvalho et al. (2021) study shows that alternative news uses emotive language three times more than authentic news. Using the sentiment analysis tool, another relevant aspect is the correlation between emotionally marked words and those with positive or negative connotations. The analyses show that texts of deceptive news tend to be more negatively oriented. The presence of strong emotional intensity in news is an important indicator of unreliability, as it differs greatly from the deontological principles of journalistic writing, which tend to be as neutral and objective as possible. Despite this, an automatic analysis of this indicator has not yet given entirely satisfactory results. Although the amount of emotional linguistic markers in the text may seem irrelevant, even a single sentence markedly negative can strongly influence the reader’s emotional perception (Aker et al., 2019).

The presence of emotional markers is therefore also consistent with the peripheral route to the persuasion model, which plays on emotional and superficial elements rather than elaborated and more thoughtful reasoning. Thus, another aspect to take into consideration is what we could define as “logos,” referring to the concept of credibility label similar to the nutritional one proposed in the work of Gollub et al. (2018) and Vincentius et al. (2018), but enriching it with other elements analyzed in the studies by Zhang et al. (2018) and Carvalho et al. (2021). By “logos,” it is meant the focus on all the persuasive features of a text whose presence or absence determines the credibility of an article. Specifically, internal inferences with logical arguments supported by scientific or single studies are signs of highly trustworthy articles. On the contrary, arguments that appeal to fear rather than rationality or that reduce complexity in a dichotomous form without the possibility of alternatives and encourage actions are significant indicators of low trustworthiness. These expedients, together with other less significant ones, have been classified as “fallacies” (Zhang et al., 2018; Carvalho et al., 2021).

On the syntactic and semantic level, linguistic analysis has produced significant results, thanks to the use of methods typical of forensic linguistics, which are here integrated with other branches of linguistics to offer a broader and more systematic picture. From a syntactic point of view, one of the distinctive elements found in news with a low credibility rate is the strategic use of the theme-rheme structure, often aimed at creating sensational effects, rather than ensuring clarity and comprehensibility (Sousa-Silva, 2022). Another syntactic device is the use of relative clauses, used to describe ordinary people in detail, often to enrich the text with apparently credible information, but difficult to verify (Grieve and Woodfield, 2023). These clauses are combined with more frequent use of other subordinate clauses, again in contrast with the traditional journalistic approach, which tends to favor information density through nominal sentences that are primarily observed in reliable news. Deceptive news, on the other hand, distributes information over several sentences, giving only the illusion of an enriched text, but in reality, reducing its clarity and amount of information. Finally, an aspect observed in several studies is the extensive use of the passive voice, which allows manipulation of the visibility of the agent. In particular, the passive construction can be used for two main purposes: to emphasize or hide the agent, depending on the narrative. Specifically, when the verbal action is negatively connoted and the subject of the action is perceived as an opposing party, the agent is made explicit, to emphasize responsibility and encourage feelings of anger against it. On the contrary, the agent can be omitted to hide responsibility when it does not benefit the creator of the news, thus creating ambiguity and limiting the reader’s ability to verify the facts. This linguistic strategy, widely used in unreliable news, allows for manipulation of the interpretation of events, favoring specific interests and hiding or highlighting information selectively (Grieve and Woodfield, 2023; Mahyoob et al., 2021; Sousa-Silva, 2022).

From both lexical and semantic points of view, low credibility news exhibits a very evident vagueness and ambiguity. That is stated in the form of a list of indefinite words and a willfully vague creation of meaning. Text vagueness is also marked by several linguistic features, such as the excessive use of personal pronouns in place of specific nouns. Although pronouns are convenient elements for ensuring textual coherence and cohesion, their excessive use can also compromise the precision and clarity of the text, making the references to action and agent difficult to follow for the reader. Rashkin et al. (2017) also showed that less reliable news tends to contain a higher frequency of first- and second-person pronouns. Another linguistic indicator regarding the lexicon is the use of hedging words. These expressions, such as claims in a statement or downtoners, such as slightly, somewhat and a bit, serve to tone down the intensity of the statements, thus creating an effect of detachment or indecision of the writer (Grieve and Woodfield, 2023). Although the use of such terms is consistent with psychological theories, suggesting that deceivers tend to use indirect and non-definitive expressions (Burgoon et al., 2003), in subsequent studies by Zhang et al. (2018) and Carvalho et al. (2021), the annotators gave uncertainty signals a high credibility value. However, the credibility value assigned by the experts in the two studies is in line with deception theories. Therefore, lexical forms of uncertainty and attenuation do not represent definitive indicators of deception themselves. Rather, it is the management of such uncertainty that affects the overall perception of credibility. The uncertainty in inauthentic news is counterbalanced by the equally significant presence of evaluative and emphatic adverbs that intensify reality, whereas there is an almost total absence of adverbs of place and time, which are instead common in reliable news.

In line with the results of sentiment analysis, low credibility news is characterized by the common use of adjectives, which confer a highly subjective and evaluative approach. This style contrasts with traditional journalistic writing techniques, which, as already mentioned, encourage an impartial and objective narration. A further linguistic indicator is the intensive use of superlatives, compared to authentic news, which tends to use more comparatives, numerical data and precise statistics. These are characteristic elements of reliable journalistic writing because they serve to contextualize and verify information, reducing subjectivity and promoting an understanding based on concrete evidence. On the contrary, the prevalence of adjectives and superlatives highlights writing that aims to emphasize and manipulate reality, playing on sensationalism and emotion to capture the public’s attention (Revenko, 2020; Grieve and Woodfield, 2023; Sousa-Silva, 2022; Mahyoob et al., 2021).

Up until now, we have focused on the linguistic indicators, thanks to which disinformation can be identified, understood as false or misleading information, intentionally spread to deceive the public. However, the true power of disinformation cannot be fully understood without analyzing how it is spread. This is a phenomenon that, with the advent of digital platforms, has undergone radical transformations, to the point that in 2013, the World Economic Forum (2013) included it among global risks, equating it to threats such as terrorism. In that period, the full extent of the problem was not yet fully perceived, but it had already become clear that the optimistic vision of informational pluralism, anticipated initially by the advent of the internet and then by social networks, was crumbling.

In the past, communication through traditional media, such as newspapers, radio and television, followed a one-way model characterized by a “one-to-many” (1 → n) relationship, where a few centralized senders diffused information to a large number of receivers. With the beginning of social media, this paradigm was radically transformed into a “many-to-many” (n ↔ n) model, where every user can be a producer and consumer of information. This decentralization of communication allowed more people to become sources of information themselves and to spread it without intermediaries, initially promoting the idea of the democratization of information. However, this process did not lead to the desired results. What dynamics contributed to making a process of informative pluralism a global threat? The answer lies in the logic governing new media distribution and in the way they reinforce our most natural cognitive processes, hence generating an unprecedented combination in the history of communication.

A crucial aspect to examine is the enormous amount of data daily generated by internet users, often unconsciously, through the acceptance of privacy policies that are rarely carefully read and offer varying levels of protection. The advent of Big Data has changed many domains, including marketing, allowing more investment in communications campaigns. Growing hybridization of politics and marketing, known as “political marketing,” has facilitated the use of Big Data in election campaigns and, by extension, disinformation campaigns (Bodó et al., 2017).

The analysis of Big Data allows the optimization of various types of campaigns (e.g. advertising, awareness, political or disinformation) by understanding the in-depth behaviors and emotions of certain audience segments. In the past, messages were distributed through channels that forced communication to a broad and undifferentiated audience, inevitably leading to wasted resources. The evolution of digital technologies made it possible to segment audiences into micro-targets, investing in a specific manner to reach only specific groups. This segmentation has significantly increased communicative effectiveness, as it allows messages to be tailored to the specific emotional triggers of each micro-target and to be spread through channels selected based on precise socio-demographic coordinates (Chester and Montgomery, 2017; Novelli, 2017). A significant example of this strategy is the Russian disinformation campaign during the war in Ukraine, where messages and distribution channels were chosen based on four distinct target groups, adapting the narrative to the specific characteristics of each (Tech Policy Press, 2023). Micro-targeting thus proves to be a powerful tool for identifying segments of the population interested in certain topics, analyzing their behaviors and delivering highly targeted messages. However, micro-targeting alone is not enough to explain the process of disinformation diffusion, which instead exploits organic mechanisms within digital platforms that play on our mental processes and lead users to become not only recipients of disinformation but also secondary promoters of disinformation through misinformation [²].

To understand how the structure of social networks profoundly influences user behavior, it is essential to consider the concept of “confirmation bias.” This cognitive bias causes our minds to reject information that conflicts with our value system or worldview while accepting and reinforcing news and information that confirms our beliefs. Confirmation bias is an economic cognitive process that facilitates quick decision-making and day-to-day processing. This phenomenon finds particularly fertile ground in virtual environments, where the structure based on cookies, fingerprinting and algorithms tends to create a space that reflects and amplifies the user’s expectations. The combination of this mechanism with the structures of digital platforms has created complex virtual environments, such as echo chambers, where people are exposed exclusively to information, opinions and points of view that confirm and reinforce their beliefs, primarily interacting with individuals who share those convictions (Quattrociocchi and Vicini, 2023).

Echo chambers have demonstrated that social networks tend to close users off in informational bubbles. Studies on user behavior have revealed that greater exposure and activity on social media do not lead to expanded interaction with different communities but rather to further isolation within one’s group (Schmidt et al., 2017). This isolation within bubbles, where only narratives shared by homophilic users circulate, feeds and consolidates polarization, which in turn manifests as the strengthening of pre-existing beliefs and as increasing division between groups with opposing views. Having confirmed the presence of echo chambers, their impact on the spread of disinformation becomes extremely interesting to understand. The model presented by Törnberg (2018) theorizes that echo chambers, which are defined as sets of users with polarized opinions and strongly connected within themselves, are more favorable to the spread of disinformation compared to unsegmented networks. In particular, research shows that news from these clusters is more spreadable due to the virality of these isolated groups. The observed effect is twofold. On the one hand, internal polarization increases virality up to a certain level (Pn = 0.6), making information more persuasive within the cluster, as the high connection between users reinforces the credibility and diffusion of the content. On the other hand, when polarization reaches a critical point, the further isolation of the cluster begins to limit dissemination outside of it, reducing overall virality. This dynamic further suggests that although some polarization has the potential to accelerate diffusion, excessive isolation may limit its global effectiveness. According to the model, echo chambers both overstate the diffusion within homogeneous clusters but also influence global dissemination by a “bandwagon” effect, whereupon the consensus established within the cluster influences external nodes. The “bandwagon” effect is a psychological phenomenon where an individual tends to adopt a given opinion or behavior following the increasing number of people adopting the same.

This phenomenon of diffusion is not specific to disinformation but is inherent in the structure of social media and thus can be generalized to any information. However, the study titled Understanding engagement with U.S. (mis)information news sources on Facebook (Edelson et al., 2021) demonstrated, through the analysis of engagement metrics, that disinformative posts tend to generate higher engagement compared to content from reliable sources, with significant differences depending on political leanings. This phenomenon can be further understood in light of the percentage of emotional intensity present in disinformative news, already discussed earlier. News that provokes strong emotional reactions, such as anger and indignation, is amplified by social media algorithms, which prioritize content capable of altering and increasing user engagement (Tucker et al., 2018).

The logic behind this dynamic is that social media aim to keep users on their platforms to maximize profits. But why does it work? Once again, the answer lies outside digital platforms but consolidates within them. One aspect to consider is “peer pressure.” The research conducted by social psychologist Solomon Asch (Quattrociocchi and Vicini, 2023) demonstrated that the presence of other individuals, unlike an isolated context, significantly influences our behavior. Echo chambers, where not only like-minded people but also friends and family are present, amplify this social pressure, transforming it into a desire for conformity and approval within the group. The tools through which the user seeks and obtains the desired approval are engagement metrics, commonly used by every social network. Among these, “likes” are indicators of social approval that consolidate the sense of community and mutual support; comments reflect and amplify polarization, contributing to the construction of a collective identity based on homophily, with appreciative comments reinforcing the group and negative comments setting the group against outsiders in an identity dynamic of “us vs them”; finally, shares act as social validators, practically contributing to the spread of misinformation (Van Bavel et al., 2021).

Engagement metrics, and more particularly the sharing function, cause the bandwagon effect. This effect, extensively studied in marketing, also occurs in the dissemination of misinformation. As the exposure to a piece of news spreads, the acceptance process also amplifies, and hence, the intent to spread it leads to heightened exposure of other nodes in the network (Li et al., 2023). Even with much effort made in making solutions for disinformation workable, such as automated detection methods, research conducted by Nyhan and Reifer (2010) has pointed out that fact-checking not only does not change opinions but can cause a boomerang effect, the so-called “backfire effect.” This effect, particularly relevant among ideologically similar users, causes radicalization of views, strengthening original beliefs rather than modifying them.

On the other hand, more recent evidence from a study conducted by Costello et al. (2024) suggests that such polarization of positions could be overcome. As the authors assert, when people are engaged in organized debate that addresses the reasons behind their beliefs and provides counterarguments specially tailored to the beliefs, significant and lasting suppression of conspiratorial thinking is feasible, even in people with deeply entrenched positions.

Considering what has been discussed so far, one might wonder whether continuing to frame the research in dichotomous terms, such as true/false or right/wrong, rather than considering when and how it is used. While distinguishing between accurate and inaccurate information continues to be important, its success could nevertheless be contingent upon the context of communication and how far the messages resonate with the source of why individuals hold their beliefs.

Apart from the dissemination mechanisms mentioned above, spreading disinformation is facilitated by AI, particularly generative AI (e.g. deepfakes and AI-generated text) is being used to disseminate disinformation. There is growing evidence from academic studies, media reports and specialist analysis that shows how these technologies can be used to create and spread false or deceptive content (Newman et al., 2024; Manos et al., 2024; Gorwa and Garton Ash, 2024). Their ability to produce highly realistic text, images and videos increases disinformation by posing a relevant challenge to global information and communication ecosystems.

Despite the numerous advantages of generative AI in different sectors, like health care, finance and education, the role it plays in the spread of disinformation is worrying. The use of generative AI models stimulates the creation of fake media that can no longer be easily distinguished from real media. This reality distortion disrupts democratic processes, manipulates public perception and erodes trust in the information ecosystem. The technological innovation brought about by the use of AI has improved the scale and efficiency of disinformation campaigns. An analysis of the role of AI in the spread of disinformation is important in assessing the appropriate response that ethical coalition approaches to public education need to take regarding the new challenges posed by AI. As asserted by Shoaib et al. (2023), there is a need to organize cyber awareness campaigns on the application of generative AI to promote mass awareness and sustain resilience by equipping citizens with the power of critically analyzing the changing mechanisms of online deception. This is also underlined by Pauwels (2024), who testifies to the worthiness of collective efforts by civil society, media, government and global actors. These actors can actively contribute to reducing the damage that disinformation causes to society by strengthening social resilience policies.

A crucial aspect to consider is that people acting as amplifiers of disinformation (e.g. propagandists and disinformation-for-hire companies) will be greatly empowered by the advances in AI technology. In fact, AI may reduce the cost of content production by up to 70%, allowing the creation of videos and deepfake content that were previously considered too expensive (Musser, 2023). Despite this, AI may improve the cost-efficiency of existing disinformation techniques, such as bot and deepfake video creation, without drastically increasing their extensive capabilities (Musser, 2023). The lower cost of content production can increase the number of participants in disinformation campaigns and can create new forms of disinformation, such as audio deepfakes and chatbots, although their effectiveness is unproven (Bristow, 2023). Furthermore, AI enables tailoring disinformation content to specific audiences, allowing strategies to be refined based on audience reactions, social media moderation policies and tailored demographic strategies.

The dissemination of disinformation has often been seen as a weakness in threat actors’ strategy. According to Simon et al. (2023), content creation is insufficient, since content must attract attention and reach a target audience. Current experiences, as Yang and Menczer (2023) point out, show that AI-driven bots that mimic human behavior can build and interact with audiences for easier content propagation, solving this challenge. Such bots, while upon initial consideration they might not be thought to have interactivity, can indeed compel viewers through hashtag bombing or bulk content creation to generate additional public interest (Schwieter and Gandhi, 2024). AI use also allows for an unprecedented level of targeting, making disinformation more efficient by strategically crafted text, images and videos aimed at specific groups (Donovan et al., 2020). Although evidence to support the idea of hyper-tailoring has yet to be presented, the political implications are far-reaching, particularly for distinctive communities or individuals who are vulnerable to publicly available or hacked data (Simchon et al., 2024).

One of the main consequences of the use of AI in disinformation campaigns is threatening societal trust in information and democracy (Altay and Acerbi, 2023; Jungherr and Rauchfleisch, 2024). As argued by Kreps and Doug (2023), AI-generated disinformation has been shown to deepen existing societal divisions, particularly in politics and public trust. This may cause an overall disruption of confidence in the reliability of information that can further escalate the level of anxiety within the democratic system, in addition to disrupting public trust (Jungherr and Rauchfleisch, 2024).

The production of AI-generated disinformation has notably increased in recent years. A company called NewsGuard [³], which assesses the credibility and transparency of news and information websites, has tracked AI-generated disinformation and has produced several reports. In June 2024, one of their audits showed that AI chatbots within Russian-linked websites made up fake accounts in 32% of cases, while falsely stating that these sources were real. In July 2024, NewsGuard’s Monthly AI Misinformation Monitor found these same chatbots spread disinformation 30% of the time, debunked 41% of their false claims and answered nothing 29% of the time. Additionally, in January 2025, a chatbot within DeepSeek [⁴] was audited, with the report concluding that it had only a 17% accuracy rate and repeated false claims 30% of the time. Due to such disinformation, these AI models do not guarantee the trustworthiness of their replies, which creates a breeding ground for disinformation. These trends indicate the multifaceted and proliferative nature of AI-generated disinformation. Governments and civil society should work with technology companies to establish ethical guidelines and sound policy recommendations to avoid such harm from AI-generated disinformation. In addition, promoting critical thinking skills and digital literacy is also crucial to empower people to counter the increasing wave of manipulation in the information age.

In light of what has been discussed so far, AI has the potential to become one of the most powerful tools for innovation and progress in various fields. However, its rapid development is putting a strain on the risks of keeping information reliable and trustworthy. Mitigating these risks is critical to ensuring the veracity of information and the democratic principles that depend on it.

How disinformation campaigns spread in digital dynamics raises the legitimate question of whether, and to what extent, these processes have evident consequences in the real world. This question is not easy to answer, since the immediate cause–effect relations are not always easily traceable (as in the case of elections) and, perhaps, they are not even the main object to investigate to obtain an answer. The solution may lie in observing the consequences of disinformation campaigns as part of a systemic process, rather than sequentially. In this sense, it is useful to draw on studies in this field from different perspectives and disciplines, which “are often entirely complementary windows into a much larger process” (Ruths, 2019).

As illustrated in the previous paragraph, the field of cognitive science has shifted its focus from the idea of the rational man towards a model of man who uses heuristics to simplify decision-making problems and reduce cognitive load. This approach perfectly aligns with the structure of social media, which exploits the confirmation bias mechanisms, inducing individuals to search and find information that confirms their pre-existing beliefs, while tending to ignore contrary ones (Geschke et al., 2018). As a result, truth loses importance, giving space to a fragmented reality, which has been defined as the “post-truth era.” Each person’s “truth” develops within their echo chamber, fostering a polarization of opinions. This polarization is not created by social networks but is amplified by them, transforming society into a place of constant tension between poles, which do not open to comparison but strengthen their identity of belonging to the group in an “us vs them” mechanism (Morlino, 2021).

As several studies have argued, the hostility of online interactions is a key factor in explaining the escalation of affective polarization. Specifically, partisan criticisms intensify the in-group’s identity climate, leading to a more pronounced distance from the out-group and fostering distorted perceptions of its members (Iyengar et al., 2012; Lelkes, 2016).

Such affective polarization is a tool that certainly has not gone unnoticed by certain segments of society. In a context where the collection of information about behaviors and needs has become less difficult (see the Cambridge Analytica case) and where the absence of mainstream media allows everyone to become an information source, it is not surprising that previously marginalized entities have now accessed large communication spaces and are using them to spread distrust in institutions and traditional media, which are considered the cause of their marginalization (OECD, 2024). As highlighted in the study by Jennings et al. (2021) about the consequences of the proliferation of disinformation during the COVID-19 pandemic, distrust toward the political system and its members often goes with feelings of insecurity, contempt, fear and anger that can translate into defiance, support for populist challenge. As Pomerantsev (2020) points out, it is therefore not surprising that:

Therefore, another possible consequence of polarization, observed by Svolik (2019), is how it could damage the fundamental principles of democracy. Social media, by their nature, are neither democratic nor anti-democratic; they are tools useful to those who know how to use them. Just as in authoritarian regimes oppressed groups have used social media to promote revolt movements – as happened during the Arab Spring, although with different outcomes in various countries – in the same way, in some democratic contexts, antidemocratic forces can use access to data to influence and polarize public opinion, building loyalty through divisive dynamics and orienting it toward their value system (Tucker et al., 2017). Svolik (2019) observes that, in a polarized context, voters tend to justify or ignore the authoritarian behaviors of the leaders of their political camp to maintain power, putting party affiliation before the protection of democratic values. Affective polarization, as highlighted by Iyengar and Westwood (2015), accentuates the importance of emotions and party loyalty, reducing the relevance of facts and rationality in electoral choices. The daily process of polarization affects the role of centrist voters, who play a crucial role in the balance of political forces (Svolik, 2019). The result of this dynamic is an increasingly vulnerable democracy. When polarized citizens accept the concentration of power and the violation of democratic norms in the name of their party, the boundaries between democracy and authoritarianism become increasingly blurred.

In this context, it is relevant to note that, despite the European Union’s concerns about disinformation campaigns from abroad, the vulnerability of democracy often originates within its very core (D’Andrea et al., 2025). Russian disinformation related to the war in Ukraine represents a symbolic case. Unlike other destabilization initiatives, in this case, the campaign did not achieve the desired results, given the unified position of the European Union. The Russian apparatus organized its strategy considering three main targets: the Russian population and the post-Soviet population, the Ukrainian government and Ukrainians and the international community, with a specific focus on NATO and the European Union. Focusing on the latter target, disinformation initially focused on presenting Russia as a defensive rather than an offensive state, justifying the invasion as a reaction to NATO expansion. However, having to face the firm European reaction, which included sanctions and military supplies to Ukraine, the Russian narrative quickly adapted. It started to highlight the consequences of the war for the West, raising socio-economic concerns. Central issues include Europe’s dependence on Russian gas and the food crisis, to spread discontent in the European public and created divisions between government institutions and the citizenry (Giusti, 2023).

These narratives are not isolated but are part of a long-term strategy aimed at destabilizing the western socio-political order. However, foreign disinformation campaigns exploit pre-existing frictions, such as ideological divisions, economic tensions and cultural conflicts, aggravating them through anti-globalist, anti-liberal, anti-elite and anti-immigration rhetoric, themes that easily take root among populist forces, conservative groups and Eurosceptics. Over 17,000 cases of pro-Kremlin disinformation documented by the EU vs Disinfo [⁵] database demonstrate how these campaigns are a pervasive phenomenon that daily undermines the integrity of western democratic systems. As Giusti (2023) points out:

In other words, foreign disinformation campaigns are not the source of problems, but amplify pre-existing discomforts, taking advantage of the vulnerability that a polarizing society creates.

Strategies against disinformation campaigns hold a paradox: if not managed with extreme caution, such strategies can in turn become tools of manipulation, aimed at undermining trust in democratic institutions. In the complex dynamics of the global information ecosystem, in which authoritarian regimes operate with fewer constraints than democratic systems, a disproportionate or poorly calibrated response by the European Union could effectively contain foreign interference on the one hand, but on the other hand, it could provide material to reinforce narratives aimed at manipulating public opinion, fostering a growing distrust toward the institutions themselves. These narratives could, for example, exploit the perception of an alleged “attack against freedom of speech” to suggest that the institutions themselves act in a repressive or anti-democratic manner. In this way, the effort to protect democratic systems could paradoxically lead against democracy itself, undermining its foundations and making it vulnerable to the very forces it intends to fight.

How can we, then, effectively fight disinformation in the post-truth era, when the search for objective truth has been replaced by the need to affirm one’s own beliefs? This question does not have a single answer but requires a systemic and coordinated approach. It is essential to involve all the implicated parts: national and supranational institutions, private platform operators, traditional media and citizens themselves. Only a collective action, oriented toward transparency and the protection of fundamental democratic principles, will be able to stop the progress of disinformation and safeguard the solidity and cohesion of democratic systems.

Foreign interference and disinformation in European democratic processes present complex challenges that are rooted both in technological progress and in inherent socio-political vulnerabilities. This study concentrated on three central themes: the detection of disinformative content by linguistic patterns; the mechanisms driving disinformation campaigns, pushed by an online ecosystem that amplifies cognitive biases, polarization and emotional manipulation; and, finally, the impact of these phenomena on public opinion and thus democratic policy-making.

The study highlighted the multidimensional complexity of disinformation campaigns, necessitating multidisciplinary approaches. Linguistics provides valuable tools for identifying recurrent patterns characteristic of disinformation. However, the effectiveness of these tools relies on a broader understanding of the context in which such campaigns operate, particularly the digital dynamics that exploit and intensify human cognitive mechanisms. With the change in the media landscape, driven by the proliferation of digital platforms, the media environment has transformed into one where echo chambers and information fragmentation have become widespread. That has made it easier for disinformation to spread and exacerbated social polarization and radicalism. Moreover, the increasing capability of AI to produce very impactful content – from deepfakes to customized fake news – threatens democratic societies. In addition, these phenomena directly affect the political debate and decision-making, as testified by the rise of populism and the decline of trust in institutions.

Countermeasures in such cases are necessary, which go beyond the detection and debunking of disinformation. A systemic approach is required, in which all actors, from the government authorities to digital platform holders, professional media outlets and above all, the general public, must be involved. The only way to mitigate the effects of disinformation, to strengthen democratic resilience and to provide the grounds for a well-informed, participatory public discussion is through a coordinated collaboration founded in media literacy and transparency activities.

Funding: This research was funded by the European Commission under the Project AI4Debunk, grant number 101135757.

Declaration of conflicting interests: The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.