1.1 Disinformation

Disinformation is false or misleading content deliberately produced and disseminated to deceive (Zhou and Zafarani, Reference Zhou and Zafarani2020). “Weaponized” disinformation – the focus of this monograph – is disinformation strategically deployed to achieve political or ideological ends (Benkler et al., Reference Benkler, Faris and Roberts2018; Krekó et al., Reference Krekó, Győri, Dunajeva, Janda, Kundra, Mesežnikov, Mesík, Szylar and Shekhovtsov2016). Intentionality separates disinformation from misinformation. While misinformation is false or misleading, it is not necessarily spread for the explicit purpose of deception. Disinformation may or may not be presented as news, distinguishing it from fake news, which has a news-like appearance. Disinformation is empirically verifiable as false, and thus distinct from rumors, which are characterized by ambiguous evidentiary standards, uncertain veracity, and social transmission dynamics (Berinsky, Reference Berinsky2017). And, while disinformation may build upon or incorporate elements of conspiracy theories, it is not necessarily predicated on explanations involving the secretive plots of powerful individuals or organizations (Douglas et al., Reference Douglas, Uscinski, Sutton, Cichocka, Nefes, Ang and Deravi2019).

Even following these criteria, classifying disinformation is challenging. Attributing specific disinformation narratives to Kremlin influence efforts or classifying an outlet as Kremlin-linked presents an even greater challenge. Verification typically occurs in two ways: (1) source verification – whereby all content from a non-credible source is considered disinformation; or (2) content verification – whereby fact-checkers evaluate individual content to adjudicate whether it is disinformation (Golovchenko, Reference Golovchenko2020, 4). I rely on both strategies, drawing on the expertise of third-party expert fact-checkers from the Czech and Slovak Republics (see Online Appendix A). These fact-checkers identify disinformation narratives, characterize disinformation-disseminating outlets, and conduct investigative work that traces narratives back to Russia or establishes connections between news outlets and the Kremlin. Nonetheless, much of the disinformation in CEE (and elsewhere) is not directly attributable to the Kremlin, but is nonetheless ideologically aligned with the Kremlin’s objectives. Thus, I use the term “Kremlin-linked” when evidence of direct connections to the Kremlin exists, and “pro-Kremlin” when narratives and media outlets are ideologically aligned with the Kremlin’s interests but lack verified connections.

1.2 Understanding Disinformation Consumption and Disinformation’s Effects on Public Opinion

I rely on theories of directionally motivated reasoning and dual-process cognition to explain individual-level disinformation consumption as well as disinformation’s effects on public opinion. Motivated reasoning theory suggests that individuals do not always seek to arrive at objectively correct or accurate conclusions. Rather, they may prefer “directional” conclusions (Kunda, Reference Kunda1990) – those that reaffirm preexisting beliefs, identities, or preferences (Druckman and McGrath, Reference Druckman and McGrath2019, 113). Directionally motivated reasoning can bias information processing at multiple stages: exposure, evaluation, and memory. Directional motivations lead individuals to preferentially seek out attitude-consistent media (Stroud, Reference Stroud2008). These motivations also lead individuals to judge congenial information as more credible and to recall confirming evidence more readily than disconfirming evidence (Taber and Lodge, Reference Taber and Lodge2006). In the context of disinformation, directionally motivated reasoning would suggest that individuals select into disinformation exposure and believe disinformation because it reaffirms preexisting attitudes and ideological commitments (Altay et al., Reference Altay, Berriche and Acerbi2023; Flynn et al., Reference Flynn, Nyhan and Reifler2017; Guess et al., Reference Guess, Nagler and Tucker2019).

Importantly, this phenomenon is not disinformation-specific. Media consumption more broadly is driven by self-selected exposure (Arendt et al., Reference Arendt, Northup and Camaj2019; Knobloch-Westerwick, Reference Knobloch-Westerwick2015). In turn, self-selected exposure leads to reinforcement effects, whereby exposure can maintain or strengthen preexisting attitudes (Arendt et al., Reference Arendt, Northup and Camaj2019, 465; Levendusky, Reference Levendusky2013, 613). Thus, media consumption tends to reinforce existing attitudes rather than persuade individuals to change them. However, given disinformation’s false nature, selection and belief may depend on individual-level cognition – specifically, the degree to which analytical reasoning can interrupt or override directionally motivated processing.

Dual-process cognition theory suggests that information processing occurs through both System 1 processing, which is fast, automatic, and intuitive, and System 2 processing, which is slow, analytical, and reflective. Motivated System 2 Reasoning (MS2R) accounts suggest that individuals with a greater propensity to engage in System 2 processing are also more likely to use these cognitive resources to arrive at attitude-consistent conclusions (Kahan, Reference Kahan2013; Knobloch-Westerwick et al., Reference Knobloch-Westerwick, Mothes and Polavin2020). In contrast, classical reasoning accounts contend that a greater propensity to engage in System 2 processing leads individuals to evaluate information more critically, making them less likely to form misperceptions (Pennycook and Rand, Reference Pennycook and Rand2019). As such, the two theories have different empirical implications for individual belief in disinformation. According to MS2R, analytical reasoning propensity amplifies – rather than attenuates – motivated reasoning. Consequently, among individuals whose ideological predispositions align with disinformation content, higher levels of analytical reasoning are expected to increase the likelihood of believing disinformation. Conversely, classical reasoning predicts that, regardless of attitudinal priors, individuals with a greater propensity for analytical reasoning will be less likely to believe disinformation.

Ideology matters for understanding individual-level information processing. As such, understanding which disinformation narratives are likely to resonate – and with whom – requires mapping the ideological cleavages that structure public opinion and political competition. Initially, in CEE, the transition from state socialism to democracy and market capitalism structured political competition around both a distributive economic dimension and an identity-based sociocultural dimension (Kitschelt, Reference Kitschelt1995). Through the early 2000s, the normative aspiration of “returning to Europe” and the material incentives of formal EU accession tempered the electoral viability of the most radical forms of nationalist conservatism and democratic backsliding (Haughton, Reference Haughton2014; Vachudova and Hooghe, Reference Vachudova and Hooghe2009). However, post-accession, sociocultural and geopolitical issues (re)emerged as the dominant axis of political competition (Vachudova, Reference Vachudova2021, 473). The expansion of cross-border economic exchange, increases in migration flows, deepening international integration, and mobilization of historically marginalized groups generated sociostructural changes with stark distributive and symbolic consequences (Kriesi, Reference Kriesi2012).

Citizens who gained from globalization and European integration – primarily the affluent, educated, and urban – embraced cosmopolitan values and inclusive definitions of national identity. Meanwhile, those experiencing economic dislocation or perceiving threats to traditional cultural norms increasingly rejected supranational governance and multiculturalism, demanding the reassertion of national sovereignty. Issue entrepreneurs amplified and politicized this emerging cleavage, compelling mainstream parties to take positions on questions of national identity, immigration, and sovereignty (Hobolt and De Vries, Reference Hobolt and De Vries2015). The result was a transformed political landscape where political competition increasingly centered on conflicts over identity, borders, and belonging. On one side of the divide, socially conservative, Eurosceptic, Russia-tolerant, anti-immigrant parties with largely left-leaning economic positions garner support from those disadvantaged by globalization (Lancaster, Reference Lancaster2022) or threatened by shifting group-based hierarchies (Buštíková, Reference Buštíková2019). On the other side, mainstream parties with centrist or socially liberal programs and pro-Euro-Atlantic commitments are supported by those with more inclusive, cosmopolitan national identities (Vachudova, Reference Vachudova2021, 473). Understanding these divisions helps explain: (1) the content of Kremlin-linked disinformation; (2) which countries are more or less susceptible to disinformation; and (3) which individuals are most likely to consume and believe disinformation narratives.

1.3 Monograph Layout

The remainder of this monograph is organized into seven sections. Section 2 situates disinformation within CEE’s ideological conflicts and details how motivated information processing and dual-process accounts of cognition explain individual reactions to disinformation. The section concludes by discussing the value of the Czech-Slovak comparison for analyzing Kremlin-linked disinformation. Section 2 uses time-series, cross-national data to analyze country-level predictors of disinformation “supply.” The analysis underscores that ideological polarization on the sociocultural dimension drives disinformation prevalence in CEE. Section 4 draws on sixty-eight in-depth interviews to illustrate how elite-level ideological conflicts over disinformation shape anti-disinformation policy in the Czech Republic and Slovakia. Section 5 examines disinformation demand, focusing on how widely individuals recognize, read, and trust disinformation media. Although disinformation outlets are less read and trusted than mainstream outlets, once ideological selection into news consumption is accounted for, readership rates and trust levels do not meaningfully differ between the two types of media. Section 6 focuses on evaluations of disinformation headlines, reaffirming the directionally motivated foundations of disinformation belief. This analysis also shows that analytical reasoning can buffer against misperceptions, but only among those not ideologically predisposed to agree with disinformation content. Section 7 evaluates disinformation’s effects on public opinion. Across six experiments and various attitudinal outcomes, disinformation either caused attitudinal backlash or had little effect. Section 8 discusses how findings contribute to academic and policy debates on disinformation.

2.1 Disinformation: A Weaponized Tool of Foreign Policy

A central objective of Russia’s foreign policy is to prevent the collective West from challenging Russia’s ability to assert its interests in its near abroad and further afield (Laruelle, Reference Laruelle, Kolstø and Blakkisrud2016; Lucas, Reference Lucas2014; Tsygankov, Reference Tsygankov2018). More specifically, Russia aims to challenge “U.S. and Western dominance of the international and democratic liberal order, restore Russia’s historic sphere of influence, and return Russia to a critical position as a coequal with the United States” (Conley et al., Reference Conley, Mina, Stefanov and Vladimirov2016, 4). The Kremlin not only seeks to hamper the ability of Euro-Atlantic institutions (i.e., the EU and NATO) to act against Russia, but it also attempts to portray Russia as the civilizational choice opposing the current, “Western,” liberal international order (Börzel and Zürn, Reference Börzel and Zürn2021, 295–296; Makarychev, Reference Makarychev2018).

Russia uses various tactics in pursuit of these objectives. As one Slovak security expert remarked, “Russian influence is here, it has many pathways, and the reasons why it is here are numerous.”Footnote ⁶ Examples include ties with cultural organizations (particularly those commemorating World War II), paramilitary groups, and pro-Russian politicians, as well as coercive energy diplomacy, strategic economic investments, and cyber attacks (Conley et al., Reference Conley, Mina, Stefanov and Vladimirov2016; Krekó et al., Reference Krekó, Győri, Dunajeva, Janda, Kundra, Mesežnikov, Mesík, Szylar and Shekhovtsov2016; Shekhovtsov, Reference Shekhovtsov2017). While some tactics – like cyber attacks – directly undermine a state’s capacity to counter Russia, many strategies rely on empowering actors who are antagonistic to the EU and NATO, sympathetic to Russia, or both. By empowering these actors, Russia strengthens their ability to shape public debate on Euro-Atlantic commitments and the legitimacy of Russia’s actions. These tactics not only sow discord within states, but also hamper their ability to collectively coordinate on the international stage.

Of the many foreign-policy tools at Russia’s disposal, disinformation is perhaps the most frequently used (Van Herpen, Reference Van Herpen2015; Yablokov, Reference Yablokov2015). Weaponized disinformation relativizes truth for political or ideological ends – that is, creates narratives that offer plausible alternative interpretations of facts. The proliferation of alternative narratives fosters an inability or unwillingness to critically evaluate information and corrodes shared standards of truth (Paul and Matthews, Reference Paul, Matthews, Baines, O’Shaughnessy and Snow2016; Pomerantsev, Reference Pomerantsev2014). Simultaneously, disinformation narratives amplify contentious issues, deepening polarization and weakening belief in official government communications, mainstream media reporting, and other authoritative information sources. Practically, Kremlin disinformation undermines Western democratic institutions, legitimizes Russian actions, erodes faith in global governance, exacerbates domestic and international divisions, and enhances Russia’s global image (Bennett and Livingston, Reference Bennett and Livingston2018; Cižik, Reference Čižik2017). Disinformation efforts also benefit from domestic reproduction. Declining trust in mainstream media, the rise of social networks as primary news sources, financial incentives created by engagement-driven algorithms, and ideologically aligned – or opportunistically motivated – actors create an environment where false narratives can circulate widely without direct Kremlin intervention. Disinformation offers a low-cost, adaptable, and plausibly deniable way to sow discord and legitimize Russian interests abroad (Giles, Reference Giles2016, 36–40).

CEE is of particular interest for Russian foreign policy – and consequently a primary target of disinformation efforts. Russia perceives the region as a weak spot within both the EU and NATO, one that can be exploited to undermine these institutions from within (Juhász and Szicherle, Reference Juhász and Szicherle2017). There are multiple reasons for CEE’s vulnerability: historical and cultural proximity to Russia, low levels of media literacy and trust, under-developed anti-disinformation capabilities, high rates of conspiratorial thinking, and political elites who willingly legitimize Russian narratives (Laputska and Yeliseyeu, Reference Laputska and Yeliseyeu2024; Marinov and Popova, Reference Marinov and Popova2022; Štětka et al., Reference Štětka, Mazák and Vochocová2021). Variations in these factors help explain why some – though certainly not all – CEE states offer Russia a path of least resistance for creating discord within Euro-Atlantic institutions (Bokša, Reference Bokša2019; Reference Zgut, Charycka, Ivanovà, Stanek, Skalická, Pavelová, Losekoot Szymanska, Bystřická, Győri and Berkes2023). Disinformation is most effective when it focuses on topics or issues of high societal salience – that is, issues that command public attention and are subject to political contestation. Focusing on these issues ensures that disinformation narratives have resonance in public discourse and thus gain traction. As such, understanding disinformation in CEE requires understanding the region’s ideological landscape.

2.2 The Ideological Context of Disinformation

Central and Eastern Europe’s transition from state socialism was a concomitant transition to the market and to democracy. As such, in the 1990s, political contestation revolved not only around capitalism and redistribution, but also around state- and nation-building. While the exact nature of these debates differed across countries, they ultimately revolved around three competing visions: how to deal with communist legacies, who belongs to the national community, and the degree to which democratic institutions should protect individual (minority) rights (Pytlas, Reference Pytlas2015, 6). Consequently, political competition was not only structured around an economic dimension, but a sociocultural one as well (Kitschelt, Reference Kitschelt1995). Variations in communist regime type, federalist center-periphery arrangements, experiences with territorial loss, and ethnic demography determined the relative salience of sociocultural divisions across countries (Buštíková and Kitschelt, Reference Buštíková and Kitschelt2009; Minkenberg, Reference Minkenberg2017; Rovny, Reference Rovny2014). In countries where the boundaries of state and nation were clearly established and where religious-secular conflicts were absent, political competition centered primarily around the economic dimension. Where these issues were less settled, the sociocultural cleavage tended to predominate (Evans and Whitefield, Reference Evans and Whitefield1998, 117–121). Importantly, unlike in Western Europe, the economic and sociocultural cleavages tended to reinforce one another, pitting socioculturally liberal proponents of rapid market liberalization against conservative traditionalists favoring a stronger redistributive role for the state (Sitter, Reference Sitter2002, 436–437).

The normative appeal and material incentives of returning to Europe – both in the form of EU accession and NATO membership – tempered the most radical forms of nationalist conservatism (Haughton, Reference Haughton2014; Vachudova and Hooghe, Reference Vachudova and Hooghe2009). Post-accession, however, divisions over sociocultural issues, commitments to international institutions, and liberal democracy would (re)emerge to structure political competition (Enyedi, Reference Enyedi2020, 374; Orenstein, Reference Orenstein2019, 126–127; Vachudova, Reference Vachudova2021, 474). The increasing salience of sociocultural and geopolitical issues was driven by a backlash to the distributive and symbolic consequences of globalization, increasing levels of migration, and the institutional advancement of minority rights (Guasti and Buštíková, Reference Guasti and Buštíková2023; Kriesi, Reference Kriesi2012). Increasing levels of free trade, the movement of people across borders, deepening international integration, and the advancement of gender, ethnic, and sexual minority rights led to economic and cultural concerns among members of dominant groups.

Over time, debates over sociocultural and geopolitical issues became increasingly intertwined with contestation over the nature of the regime. Voters, increasingly alienated from mainstream parties and threatened by sociostructural change, shifted their support to illiberal parties that fused cultural conservatism with economic populism (Binev, Reference Binev2023; Snegovaya, Reference Snegovaya2024). As populist challengers and radicalized mainstream parties grew in electoral appeal and won office, illiberal governments became increasingly “hostile to checks and balances, state neutrality, and the ability of the mass media and civil society to hold decision-makers accountable” (Enyedi, Reference Enyedi2023, 12). Democratic backsliding and growing threats to liberal institutions provided a powerful basis for civic counter-mobilization among pro-democracy actors (Guasti and Michal, Reference Guasti and Michal2025; Vachudova et al., Reference Vachudova, Dolenec and Fagan2024). These dynamics gradually crystallized into an autocracy–democracy cleavage, pitting parties that embrace illiberal majoritarianism against those defending pluralist institutions, judicial independence, and the rule of law. This regime dimension adds another layer onto existing ideological debates over sociocultural issues and geopolitics, with illiberal actors typically aligned with nationalist-conservative positions on identity and sovereignty, and pro-democratic forces generally oriented toward liberal-cosmopolitan values and Western institutional commitments.

In CEE, these ideological divisions operate against a backdrop of weak party-voter linkages and widespread distrust of politicians (Ceka, Reference Ceka2013). While partly a product of the region’s communist legacy (Pop-Eleches and Tucker, Reference Pop-Eleches and Tucker2011), the more proximate cause underpinning weak linkages and distrust is voter frustration with elite-level corruption (Engler, Reference Engler2016). This frustration creates opportunities for political entrepreneurs to use anti-establishment appeals and position themselves as outsiders challenging the corrupt status quo (Hanley and Sikk, Reference Hanley and Sikk2016; Haughton and Deegan-Krause, Reference Haughton and Deegan-Krause2015). Consequently, novelty becomes an electoral asset, while governing experience may signal complicity in corruption rather than competence. Thus, party birth, death, and re-branding are common, and electoral volatility remains high (Gyárfášová et al., Reference Gyárfášová, Hlatky and Slosiarik2024; Haughton and Deegan-Krause, Reference Haughton and Deegan-Krause2020). Despite this volatility, however, CEE political competition is ideologically structured (Rovny and Polk, Reference Rovny and Polk2017). Corruption is largely a valence issue, and thus populist appeals rooted in opposing corrupt elites can be found across the ideological spectrum, including at the center (Engler et al., Reference Engler, Pytlas and Deegan-Krause2019).

Just as Kremlin-linked disinformation capitalizes on the ideological salience of sociocultural and transnational issues, it also exploits widespread mistrust of established parties and political elites. When it comes to the former, disinformation narratives often focus on issues like migration, gender “ideology,” LGBTQ+ rights, and multiculturalism, depicting them as foreign ideas imposed on CEE by the EU or the “West.” As for the latter, narratives exploit widespread disillusionment with the post-communist transition by framing European integration and liberal democratic reforms as projects that benefited political and economic elites rather than ordinary citizens. Critically, widespread distrust of political elites and institutions creates a double vulnerability: It strengthens the credibility of these narratives while simultaneously limiting the government’s ability to effectively counter disinformation.

The COVID-19 pandemic further intensified these dynamics. Across CEE, governments had to manage the health crisis and communicate public guidance amid low levels of public trust. Citizens were affected not only by the health risks of the virus but also by government measures such as lockdowns, mask mandates, and vaccination campaigns. This environment created fertile ground for conspiracies (Marinov and Popova, Reference Marinov and Popova2022, 237), which disinformation news sources exploited to spread disinformation about the pandemic’s origin and severity, and the legitimacy of government public-health efforts. Opposition nationalist and populist politicians amplified these narratives, capitalizing on public frustration to present themselves as defenders of national sovereignty and individual freedom. New anti-system movements, many of which would eventually coalesce into political parties, formed to protest perceived government overreach. Thus, the pandemic was an additional catalyst – one that intensified preexisting sociocultural divisions, mobilized new actors, and gave disinformation media novel content to exploit (Guasti, Reference Guasti2020).

Just as the pandemic was receding, Russia’s full-scale invasion of Ukraine and the resulting arrival of over six million Ukrainian refugees in Europe introduced new – and arguably existential – questions to CEE politics. Initially, there was broad support for both Ukraine’s war effort and for Ukrainian refugees; however, consensus did not last. As the war continued, populist radical parties, already sceptical of commitments to the EU and NATO, grew increasingly opposed to military aid for Ukraine and to resettlement aid for Ukrainian refugees (Hooghe et al., Reference Hooghe, Marks, Bakker, Jolly, Polk, Rovny, Steenbergen and Vachudova2024; Kamenova, Reference Kamenova2025). Public opinion shifted in a similarly sceptical direction (Hlatky, Reference Hlatky2026). Disinformation campaigns soon exploited these growing divisions, aiming to weaken Western support for Ukraine. Narratives depicting Ukraine as a failed state, governed by fascists who persecute Russian-speaking minorities, spread rapidly (Polegkyi, Reference Polegkyi2023, 100). More generally, Kremlin-linked disinformation increasingly focused on: undermining perceptions of Ukraine’s legitimate right to defend its own territory, eroding support for Western military and humanitarian aid, and justifying the Russian invasion (Erlich and Garner, Reference Erlich and Garner2023, 10). Simultaneously, Kremlin-linked messaging exploited public frustration over refugee resettlement, portraying refugees as a societal burden.

In sum, Kremlin-linked disinformation capitalizes on broader societal divisions and contentious political issues. This focus explains why certain individuals – namely sociocultural conservatives with anti-Western geopolitical commitments – may be ideologically predisposed to seek out disinformation narratives and find them credible.

2.3 Motivated Reasoning, Dual Processing, and Disinformation

Individuals seek out and believe disinformation when disinformation narratives resonate with their preexisting ideological commitments or political attitudes (Altay et al., Reference Altay, Berriche and Acerbi2023, 8; Batista Pereira et al., Reference Batista Pereira, Bueno, Nunes and Pavão2022, 2196; Guess et al., Reference Guess, Nagler and Tucker2019, 2; Vegetti and Mancosu, Reference Vegetti and Mancosu2020, 686; Zimmermann and Kohring, Reference Zimmermann and Kohring2020, 3). In other words, disinformation consumption and belief are a product of selective exposure and directionally motivated reasoning. According to motivated reasoning theory, information processing is not always guided by accuracy motivations: the desire to arrive at factually correct conclusions. Rather, individuals often process political information with directional aims – that is, for the purpose of arriving at specific conclusions. Individuals engage in directionally motivated reasoning to protect their identity and group standing, or to avoid the cognitive discomfort of revising firmly held beliefs (Druckman and McGrath, Reference Druckman and McGrath2019; Kunda, Reference Kunda1990; Taber and Lodge, Reference Taber and Lodge2006). As such, directional goals tend to override accuracy goals when political information implicates salient group identities or involves politicized, ideologically divisive issues – all features common to disinformation narratives. Moreover, directionally motivated reasoning leads individuals not only to seek out and accept information that confirms their preexisting attitudes, but also to discount or over-scrutinize information that challenges said attitudes. Thus, exposure to and belief in Kremlin-linked disinformation narratives should be systematically correlated with ideological agreement with disinformation content.

More generally, the dual-process theory of information processing suggests that information processing occurs through two systems. System 1 cognitive processing is fast, intuitive, and operates unconsciously. System 2 cognitive processing is analytical, reflective, and involves conscious reasoning, logic, and evaluation. Individuals vary in their tendency to rely on System 1 versus System 2 processing (Frederick, Reference Frederick2005), with important downstream implications for motivated reasoning. The MS2R account suggests System 2 processing – the tendency of an individual to engage in explicit, analytical cognitive reflection and deliberation when processing information – leads to stronger motivated reasoning (Kahan, Reference Kahan2013; Knobloch-Westerwick et al., Reference Knobloch-Westerwick, Mothes and Polavin2020). In other words, individuals who are predisposed to think analytically and reflectively about the information that they consume deploy these analytical capabilities to arrive at conclusions consistent with their preexisting beliefs. Thus, when disinformation accords with an individual’s ideological convictions, analytical reasoning can actually strengthen belief in false content. In contrast, “classical” accounts of reasoning suggest that analytical and deliberate reasoning (System 2) leads to more accurate judgments and conclusions than intuitive, autonomous, non-reflective reasoning (System 1) (Pennycook and Rand, Reference Pennycook and Rand2019, 40). Thus, the propensity to engage in analytical reasoning and the related deployment of cognitive resources should decrease an individual’s susceptibility to believing disinformation (Bago et al., Reference Bago, Rand and Pennycook2022; Erlich et al., Reference Erlich, Garner, Pennycook and Rand2023; Pennycook et al., Reference Pennycook, McPhetres, Bago and Rand2022; Tappin et al., Reference Tappin, Pennycook and Rand2020).

Directionally motivated reasoning also suggests that disinformation’s attitudinal effects depend on how narratives align with recipients’ preexisting attitudes. When disinformation reinforces existing attitudes, it may persuade – that is, cause individuals to move their attitudes in the direction of attitude-congruent information (Levendusky, Reference Levendusky2013). Conversely, exposure to counter-attitudinal disinformation may trigger defensive responses that produce backfire effects, moving attitudes in the opposite direction of the intended message. Though empirical evidence of attitudinal backlash is generally limited (Broockman and Kalla, Reference Broockman and Kalla2025; Guess and Coppock, Reference Guess and Coppock2020; Wood and Porter, Reference Wood and Porter2019), recent research suggests that polarization does occur when topics are salient and emotionally charged (Velez and Liu, Reference Velez and Liu2025, also see Levendusky, Reference Levendusky2013, 617) – characteristics that tend to define disinformation narratives. Finally, entrenched attitudes may prevent any form of attitudinal updating, especially when considering one-shot instances of disinformation exposure.

2.4 The Czech Republic and Slovakia in Comparative Perspective

While portions of this monograph explain disinformation supply by analyzing cross-national data from CEE, much of the empirical evidence comes from two cases: the Czech Republic and Slovakia. Comparing individual-level disinformation demand across the two countries provides analytical leverage for establishing scope conditions – that is, assessing the extent to which individual-level mechanisms operate consistently across contexts. The two countries vary substantially along multiple dimensions plausibly associated with how individuals engage with and respond to disinformation. If individual-level mechanisms are robust, similar patterns of motivated reasoning and selective exposure should emerge despite these structural differences. Conversely, if mechanisms are context-dependent, we should observe systematically different relationships between individual characteristics and disinformation consumption across the two cases. While an analysis of two cases has limited external validity, the Czech Republic and Slovakia represent sufficiently different contexts to provide indicative evidence about whether individual-level mechanisms of disinformation engagement operate similarly across the region.

Slovakia is deeply divided over both sociocultural and geopolitical issues (Mesežnikov and Gyárfášová, Reference Mesežnikov and Gyárfášová2018), and in recent years these divisions have taken on a regime-level dimension, with polarization increasingly extending to questions related to the rule of law and independent institutions. These divisions are reflected in an increasingly fragmented media landscape and hostile relationships between some politicians and many mainstream media outlets. Many of these same elites also openly advocate for cooperative ties with Russia and reinforce pro-Kremlin narratives (Lintner et al., Reference Lintner, Diviák, Nekardová, Lehotský and Vašecka2023). At the same time, while these elites rarely question Slovakia’s EU or NATO membership outright, they have shown little hesitation in politicizing both institutions for electoral gain. Elite endorsement of Russia, coupled with the de-legitimization of Euro-Atlantic institutions, lends credibility to disinformation outlets and narratives. Furthermore, institutional trust in government and mainstream media is low. Slovakia’s institutional infrastructure for countering disinformation also remains underdeveloped, and the country’s energy dependence on Russia further exacerbates vulnerability. Collectively, these conditions – polarization, elite endorsement of pro-Kremlin narratives, low institutional trust, and weak counter-disinformation capacity – position Slovakia as one of the most disinformation-vulnerable countries in CEE (Laputska and Yeliseyeu, Reference Laputska and Yeliseyeu2024).

In contrast, the Czech Republic is generally less polarized on both sociocultural and geopolitical issues. While migration and environmental policies are divisive, questions related to gender and sexuality rights are less salient and Czech society remains broadly sceptical of Russia. Although radical parties on both the left and the right have called for withdrawal from the EU and NATO, public support for both institutions remains relatively high and mainstream elites have largely reaffirmed the Czech Republic’s commitment to both institutions. While the 2025 parliamentary election raised questions related to the rule of law and liberal democracy, the Czech Republic has not experienced the erosion of independent institutions seen in Slovakia, and Czechs generally report higher levels of institutional trust than Slovaks. Importantly, Czech political elites and policymakers recognized disinformation as a national security challenge relatively early, and thus invested in counter-disinformation capabilities. Elite-level endorsements of disinformation outlets and narratives remain rare. Political stability and effective governance – at least relative to Slovakia – reinforce Czech resilience to foreign influence operations.

3.1 Ideological Polarization and Country-Level Disinformation Supply

For the first test of Hypothesis 1, I assess the relationship between party-system ideological polarization and disinformation supply. To operationalize disinformation supply, the dependent variable, I use V-Dem’s v2smforads measure. The measure asks country experts to evaluate the extent to which “foreign governments and their agents use social media to disseminate misleading viewpoints or false information to influence domestic politics” (Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman, Angiolillo, Bernhard, Cornell, Fish, Fox, Gastaldi, Gjerløw, Glynn, Good God, Grahn, Hicken, Kinzelbach, Marquardt, McMann, Mechkova, Neundorf, Paxton, Pemstein, von Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tannenberg, Tzelgov, Wang, Wiebrecht, Wig and Ziblatt2025, 337). Experts provide their assessments on a 0 (“extremely often”) to 4 (“never, or almost never”) scale. This measure is available annually for all CEE countries between 2000 and 2024. For regression analyses, I use the interval measure from V-Dem’s measurement model, which aggregates ratings across experts, taking into account uncertainty, disagreement, and measurement error. I reverse code the scale so that higher values indicate greater disinformation supply. The resulting scale ranges from −2.489 to 3.276, where 0 represents the sample mean across all country-years included in V-Dem’s dataset.

I measure ideological polarization at the party-system level. I do so for three reasons. First, party-system ideological polarization can be measured systematically across countries and over time using standardized expert surveys or manifesto data on party positions. Moreover, a party-system-level measure can capture polarization along multiple ideological dimensions – economic, sociocultural, and geopolitical – allowing comparison of which societal divides are most strongly associated with disinformation supply. Second, voters take cues from their preferred parties when forming opinions (Slothuus and Bisgaard, Reference Slothuus and Bisgaard2021), suggesting that the ideological positions of parties may shape mass polarization (Abramowitz and Saunders, Reference Abramowitz and Saunders2008; Moral and Best, Reference Moral and Best2023). As such, party-system polarization both reflects and drives the ideological divides that structure mass political attitudes. Third, party-system polarization distinguishes between latent and politically mobilized social divisions. Public opinion can be polarized on various issues, but only some are divisive enough to structure electoral competition. Party-system polarization thus identifies which issues have become high-stakes political battlegrounds worthy of targeted disinformation campaigns. To operationalize party-system polarization, I use the weighted standard deviation formula proposed by Dalton (Reference Dalton2008):

$P=\sqrt{\sum _{i=1}^N{v}_i{\left(x_i-{\bar{x}}_v\right)}^2}$(3.1)

where

$\begin{array}{rl}i& \text{is an individual party},\\ v_i& \text{is the vote share of party}i,\\ x_i& \text{is the ideological score of party}i,\\ {\bar{x}}_v& \text{is the vote-weighted mean ideological position of the party system},\\ N& \text{is the number of parties in the system}.\end{array}$

The index measures party-system ideological dispersion by calculating the vote-weighted standard deviation of party positions from the system’s mean. The index is calculated by taking the square root of the sum of squared distances between each party’s ideological position and the vote-weighted system average ideological position, with distances weighted by the party’s vote share. Consequently, polarization increases both as parties diverge further from the ideological center of the system and as electorally larger parties occupy more distant positions from the system mean.

I identify parties’ ideological positions using the Chapel Hill Expert Survey (CHES; Jolly et al., Reference Jolly, Bakker, Hooghe, Marks, Polk, Rovny, Steenbergen and Vachudova2022; Rovny et al., Reference Rovny, Polk, Bakker, Hooghe, Jolly, Marks, Steenbergen and Vachudova2025), an expert survey on political party positioning. CHES provides expert assessments of CEE parties for all elections between 2000 and 2024 (with the exception of Croatia, where parties were first evaluated for the 2011 election). To measure sociocultural polarization, I use expert assessments of party positions on the GAL/TAN dimension (0–10, “libertarian/postmaterialist”–“traditional/authoritarian”). I proxy for geopolitical positioning with parties’ positions toward EU integration (1–7, “strongly opposed”–“strongly in favor”). Finally, I measure polarization on economic issues with assessments of party positions on the traditional left-right economic dimension (0–10, “extreme left”–“extreme right”). For comparability, I rescale the European integration measure to 0–10. The polarization indices can range from 0 (all parties at identical positions) to 5 (two equally-sized parties at opposite extremes of the 0–10 scale). However, in the data, polarization ranges from 0.28 to 3.61, with standard deviations ranging from 0.58 to 0.67.

The unit of analysis is the country-year. However, CHES only provides data on party positions for country-election-years. To merge the V-Dem and CHES data, I first calculate polarization indices for each country-election-year; then, I forward-fill each country’s polarization values from each country-election-year through subsequent non-election years until the next election (when the polarization score is updated). I then merge these data with V-Dem’s country-year indicator of disinformation supply. The resulting dataset is an unbalanced panel of 240 country-year observations covering 11 CEE countries from their first available election in the CHES (between 2000 and 2002 for all countries except Croatia) through 2024.

To estimate correlations, I use a linear model with country fixed effects, which control for time-invariant country characteristics. Associations are estimated from within-country variation in polarization over time. I estimate all models with jackknife-based cluster-robust (CR3) standard errors. All right-hand-side variables are lagged by one year to mitigate simultaneity bias. Across different model specifications, I control for V-Dem’s freedom of expression and alternative sources of information index, V-Dem’s rule of law index, annualized changes in GDP/capita, inflation, and unemployment, proportionality (using the Gallagher least squares index), the effective number of parliamentary parties, the share of total domestic energy consumption supplied by Russian fossil fuel imports, and ethnic majority group size. Online Appendix A discusses the choice of control variables in greater detail.

3.1.1 Results

Table 1 presents results. Model 1 only includes the three polarization indices, while Models 2–5 sequentially add control variables. Across all models, sociocultural polarization has a statistically significant association with disinformation supply. Based on the estimates of Model 5, a one-standard-deviation (0.611) increase in sociocultural polarization is associated with a 0.140 increase in disinformation supply – which amounts to around ten percent of the outcome’s standard deviation. In other words, the association is fairly substantial in magnitude. There is limited statistical evidence that economic and geopolitical polarization are associated with disinformation supply.

Table 1Party-system ideological polarization and country-level disinformation supply

Unstandardized coefficient estimates with cluster-robust standard errors in parentheses; * $p<0.05$, ** $p<0.01$

To assess robustness, I reestimate all models using two alternative polarization measures from V-Dem. The first measure (v2cacamps) captures the extent to which differences in political opinion affect personal relationships, while the second (v2smpolsoc) measures whether differences of opinion exist on a few or multiple issues and the severity of these differences. As such, these measures more closely capture affective and mass polarization than the party-system measures used in the preceding analysis. However, they are not dimension-specific, and thus cannot identify which divisions drive polarization. Online Appendix A presents results, which confirm a positive association between both measures and disinformation supply.

3.2 Ideological Polarization and Individual-Level Disinformation Exposure

I also test Hypothesis 1 with individual-level survey data from Eurobarometer. Doing so helps triangulate – determine whether aggregate-level results replicate at a lower unit of analysis with a different measure of disinformation supply. Flash Eurobarometer 464, fielded in 2018, asked respondents to evaluate how often they “come across news or information” that is “false” or “misrepresent[s] reality.” Responses were measured on a four-point scale ranging from “every day or almost every day” to “seldom or never.” Flash Eurobarometer 2832 assessed disinformation exposure by asking respondents to indicate how often they have been “exposed to disinformation and fake news over the last 7 days.” Respondents answered on a five-point scale, ranging from “very often” to “never.” To create a common scale, I collapse “every day or almost every day” and “at least once a week” from the 2018 survey and “very often” and “often” from the 2022 survey into one response category; “several times a month” (2018) and “sometimes” (2022) into another; and “seldom or never” (2018), “rarely” (2022), and “never” (2022) into a third. While this introduces measurement imprecision, it enables consistent analysis across survey years. In the resulting three-point scale ($M=2.34$; $SD=0.79$), higher values indicate greater disinformation exposure.

The polarization indices remain the primary predictors of interest. However, I decompose the measures into between-country means and within-country deviations. This decomposition distinguishes cross-sectional differences in polarization across countries from temporal changes within countries, allowing for the estimation of both between- and within-country effects while accounting for unobserved country-level heterogeneity. I control for the same lagged country-level covariates used in the previous analysis, with the exception of majority group size, which is largely time invariant between 2018 and 2022. I also include year fixed effects (2018 vs. 2022) to account for survey-year-specific shocks. At the individual level, I control for trust in mainstream media and frequency of social media use, two important confounders driving both polarization and disinformation exposure (Tokita et al., Reference Tokita, Aslett and Godel2024). Additional individual-level controls include age, gender, education, and locality size. Online Appendix A presents full item wording, response scales, and descriptive statistics. Given the hierarchical data structure, I estimate associations using multilevel models with country-level random intercepts.

3.2.1 Results

Table 2 presents results. Model 1 presents the results of a multilevel model estimated with only country-level covariates, while Model 2 adds all individual-level controls. Although the intraclass correlations for both models are low, indicating limited residual country-level clustering after controlling for country-level covariates, the multilevel specification allows for the decomposition of within- and between-country variation in polarization and accounts for the nested structure of the data.

Table 2Party-system ideological polarization and individual-level disinformation exposure

Unstandardized coefficient estimates with standard errors in parentheses; * $p<0.05$, **$p<0.01$; Full results in Online Appendix A.

Both within-country changes and between-country differences in sociocultural polarization are positively associated with disinformation exposure. Per Model 2, association sizes are substantive – amounting to between 22 percent (within) and 33 percent (between) of disinformation exposure’s standard deviation. Both between- and within-country differences in economic polarization are negatively associated with disinformation exposure, with between-country differences substantively more consequential than within-country changes. Unlike in the country-level results, geopolitical polarization exhibits a negative over-time, within-country association with disinformation exposure, though the association is relatively modest, amounting to 10 percent of the outcome’s standard deviation.

More generally, within-country variation in polarization is the more consequential predictor of disinformation exposure. Across both models, all within-country polarization measures are statistically significant. However, within-country increases in sociocultural polarization are associated with greater exposure, while increases in economic and geopolitical polarization are associated with less exposure. Between countries, only sociocultural and economic polarization remain significant predictors, while geopolitical polarization is not associated with disinformation exposure. These results largely reaffirm the aggregate country-level findings, which also showed that within-country increases in sociocultural polarization were associated with greater disinformation supply. Importantly, however, the individual-level data also highlight that when party competition revolves around economic issues, citizens are less likely to report disinformation exposure – suggesting that economically structured political conflict may be less conducive to the spread of disinformation than polarization along sociocultural lines.

3.3 Discussion

Aggregate country-level and individual-level survey data indicated that party-system polarization over sociocultural issues is positively associated with disinformation supply – both across countries and within countries over time. In contrast, ideological divisions over economic issues and geopolitics exhibited less consistent – and generally negative – associations with disinformation prevalence. Taken together, these findings suggest that the growing salience of sociocultural issues and the subsequent reorientation of party competition around the sociocultural cleavage have created the conditions necessary for disinformation to spread. Disinformation is not only more prevalent in socioculturally polarized countries, but its prevalence also increases as sociocultural polarization deepens within countries. Moreover, divisions over economic issues do not appear to create the same conducive environment for disinformation. Rather, disinformation proliferates when political competition revolves around contested values, norms, and group identities. Importantly, while the analyses operationalized disinformation supply in two different ways – expert assessments and self-reported exposure – neither operationalization explicitly measured pro-Kremlin or Kremlin-linked disinformation. Unfortunately, such measures with comparable cross-national and temporal coverage are not readily available.

These findings also motivate the selection of the Czech Republic and Slovakia as comparative cases. As Figure 1 shows, Slovakia exhibits markedly higher levels of party-system polarization over sociocultural issues compared to the Czech Republic. This variation can shape disinformation supply through multiple mechanisms. Polarization may shape the Kremlin’s strategic incentives, motivating it to disseminate disinformation where it may have the largest impact. Likewise, polarization may increase demand for ideological or partisan news, creating a niche for outlets that disseminate pro-Kremlin disinformation. Polarization may also shape elite behavior around disinformation, incentivizing the reproduction of false narratives for political gain and reducing willingness to identify disinformation as a societal threat. These elite-level variations may in turn shape the degree to which governments prioritize building anti-disinformation capacity. The next section evaluates these claims through a comparison of Czech and Slovak elite-level perspectives on disinformation.

Figure 1

Sociocultural polarization in the Czech and Slovak Republics

4.1 Data and Methods

I analyze in-depth semi-structured elite interviews conducted between September 2019 and April 2020, May and July 2023, and September and December 2024. All interviews were conducted in person, with the exception of four respondents who preferred to provide written answers and three respondents who were interviewed through online video conferencing. All interviews were conducted in Czech or Slovak. To recruit politicians, I contacted members of parliament who had committee assignments related to the topics – for example, foreign affairs or defense – and central party offices/secretariats. To recruit government officials and bureaucrats, I contacted government offices responsible for disinformation-related policies. Additionally, I reached out to non-governmental organizations – for example, civil-society initiatives and policy organizations – working on disinformation resilience. In total, I interviewed sixty-eight respondents: thirty-seven in the Czech Republic and thirty-one in Slovakia. Table 3 details the number of interview respondents by position and country.

To analyze the interviews, I rely on thematic analysis. Thematic analysis identifies larger patterns of meaning or core ideas within and across respondent answers. The method allows for a nuanced and detailed assessment of elite-level perspectives, relying on interviewees to describe their views in their own words with the contextual depth they deem sufficient. In thematic analysis, a theme is composed of smaller “building blocks” known as codes (Clarke and Braun, Reference Clarke and Braun2017, 297). I took an iterative strategy, combining both deductive and inductive approaches to identifying codes. Such an approach “involves the application of an explicit theoretical framework” while also focusing on “the generation of themes from the data” (Proudfoot, Reference Proudfoot2023, 310). First, I open-coded all interviews. Second, I consolidated the codes, sorting them into emerging themes. Third, I defined each theme for re-coding. Fourth, I re-coded the interviews, keeping only themes that appeared in multiple interviews and were afforded extended discussion by participants.

4.2 Results

Thirteen themes related to the four research questions emerged. Where relevant, I highlight how themes varied across: politicians from different ideological camps; the Czech Republic and Slovakia; and the three fieldwork time periods. I begin with politicians’ assessments before turning to the perspectives of government officials and non-governmental experts involved in anti-disinformation efforts.

4.3 Discussion

This section offered a comprehensive consideration of how politicians in the Czech and Slovak Republics think about both disinformation and the policies intended to build disinformation resilience. Moreover, the analysis explored how elite-level views shape the work of the government officials and non-governmental experts responsible for the technical implementation of anti-disinformation efforts. Disinformation is subject to political contestation between conflicting ideological camps. These conflicting perspectives align with the wider sociocultural and geopolitical divisions shaping political competition in both countries. While many liberal and centrist politicians identified disinformation as a clear problem and acknowledged their country’s vulnerabilities, conservative politicians suggested that the threat of disinformation was exaggerated. The most radical respondents argued disinformation was a politicized fabrication, created to disparage alternative, conservative views.

Politicians’ views on disinformation shaped both short- and long-term implementation of anti-disinformation policies. Clear country differences emerged. In the Czech Republic, due to greater ideological consensus, anti-disinformation capacities were built early and have not faced the same degree of politicization as in Slovakia. Thus, despite the lack of a centralized coordinating mechanism and rhetoric rarely matched by committed resources, government officials and non-governmental representatives acknowledged the incremental yet meaningful steps taken to build disinformation resilience. In contrast, Slovakia’s anti-disinformation efforts only began in earnest in 2020, and while respondents positively assessed the work done, worries about durability were pervasive. With Smer’s 2023 electoral victory, government priorities shifted away from disinformation resilience. Strategic policy documents were rewritten to de-emphasize foreign threats, resources and personnel were reallocated or cut, and government officials largely stopped publicly addressing disinformation and other hybrid threats. With the change of government, Slovakia’s disinformation resilience suffered a significant setback.

Ultimately, this analysis emphasizes that the battle against disinformation in both countries is far from over. Disinformation remains a deeply politicized and ideologically divisive issue. Ideological polarization not only undermines the credibility of anti-disinformation efforts but also creates an environment where legitimate concerns about information manipulation can be dismissed as partisan rhetoric. Building disinformation resilience thus remains a significant challenge, hampered by limited resources, fragmented responsibilities across multiple actors, and polarized elite discourse. These systemic challenges have direct implications for how individuals consume and respond to disinformation – the focus of the next section.

5.1 Recognition, Readership, and Trust of Disinformation News Sources

To evaluate Hypothesis 2, Figure 2 plots the percentage of respondents who recognized (light gray) and read (dark gray) at least one, two, or four disinformation outlets in the Czech Republic and Slovakia. The number of respondents who recognized and read at least one mainstream outlet is plotted on the left. Disinformation outlets were widely recognized in both countries. However, recognition was slightly higher in Slovakia. 76 percent of Czechs and 82 percent of Slovaks recognized at least one disinformation source but, as the threshold increased (to two or more outlets and to four or more outlets), differences in recognition became larger. In the Czech Republic, only 42 percent of respondents recognized two or more disinformation sources, and just 13 percent were aware of four or more. In Slovakia, the corresponding rates were 62 percent and 36 percent.

Figure 2

Media recognition and readership

Both Czechs and Slovaks read disinformation outlets less frequently than mainstream media. Twenty-two percent of Czechs reported reading one or more disinformation news source. The rate in Slovakia was 43 percent – nearly twice as high. Again, the gap between countries increased substantially at higher thresholds: It was nearly three times as large for two or more sources and nearly four times as large for four or more. This descriptive analysis provides indicative evidence for Hypothesis 2: Slovaks consume disinformation media at substantially higher rates than Czechs. Perhaps the more important takeaway, however, is that a substantial portion of individuals in both countries consume disinformation, indicating that disinformation has achieved significant penetration in both media environments.

Next, I take a descriptive look at trust in mainstream and disinformation outlets. Figure 3 plots average trust (and associated 95 percent confidence intervals) in mainstream and disinformation outlets conditional on recognition (gray) and conditional on both recognition and readership (black).

Figure 3

Media trust

Conditional on recognition, Czechs and Slovaks demonstrated significantly greater trust in mainstream than in disinformation media. In the Czech Republic, the difference was around 0.5 on the five-point scale. In Slovakia, the difference was about half the size. The difference-in-differences estimate – that is, the extent to which the trust gap between mainstream and disinformation outlets differed across the countries – was 0.21 and significant at $p<0.001$. Importantly, both lower levels of trust in mainstream media and higher levels of trust in disinformation outlets among Slovaks account for this difference.

Respondents expressed more trust in outlets they read than in sources they only recognized. This readership “premium” was substantially smaller for mainstream news sources than for disinformation outlets. In both countries, trust in mainstream sources was approximately 0.45 points higher among readers than non-readers. For disinformation sources, the difference was 0.86 in the Czech Republic and 0.76 in Slovakia. This difference in trust between readers and non-readers statistically differed across media type ($\mathrm{\Delta }=0.35$, $p<0.001$); however, it did not differ between countries – that is, the triple difference-in-differences was not statistically significant ($\mathrm{\Delta }=-0.070$, $p=0.261$). In other words, the size of the trust premium afforded to disinformation outlets by readers, while greater than that afforded to mainstream outlets, did not differ between countries.

These results provide mixed evidence for Hypothesis 2. Slovaks were less trusting of mainstream media than Czechs, and showed greater trust in disinformation outlets they recognized but did not necessarily read. However, conditioning on readership eliminated these differences: Czechs and Slovaks were equally trusting of disinformation outlets they read. Substantively, while disinformation audiences differ in size across the two countries, once self-selection is accounted for, the two audiences do not exhibit meaningfully different levels of trust in disinformation media. An additional analysis in Online Appendix B disaggregates evaluations by outlet and underscores that there are substantial variations in trust across disinformation outlets in both countries.

5.2 Predictors of Disinformation Readership and Trust

Next, I identify the correlates of disinformation readership and trust, and assess whether they differ across the two countries. The first dependent variable, Readership, takes on a value of 1 if a respondent reported reading at least one disinformation news source and 0 otherwise. Readership is not conditional on recognition – that is, the analysis focuses on the differences between those who read disinformation and those who do not, no matter the latter group’s reasons for not being consumers. The second dependent variable, Trust, measures the average level of trust a respondent expressed toward recognized disinformation outlets (1–5 scale).

For each outcome, I estimate three models. I start with a baseline set of demographic and cognitive covariates: age, education (primary, secondary, university), sex, locality size, political knowledge, and cognitive reflection test (CRT) scores. Political knowledge is a binary measure coded as 1 if respondents answered at least two of three EU-related factual questions correctly, and 0 otherwise. The CRT measures an individual’s propensity to engage in System 2 analytical reasoning (Pennycook et al., Reference Pennycook, Cheyne, Koehler and Fugelsang2016). The questions are designed to present an intuitive, seemingly correct answer that is actually incorrect. I use the expanded seven-item CRT (Thomson and Oppenheimer, Reference Thomson and Oppenheimer2016), which adds non-numeracy-related questions to the original three-item inventory (Frederick, Reference Frederick2005). The CRT shows marginally acceptable reliability in both the Czech Republic (0.65) and Slovakia (0.67), with no improvement possible by removing items. CRT scores range from 0 to 7 based on the number of items answered correctly. The second model adds geopolitical alignment (should the Czech Republic/Slovakia belong to the West, East, or between), radical and populist party support (0, 1; parties identified as radical-left, radical-right or populist with the PopuList dataset; Rooduijn et al., Reference Rooduijn, Pirro, Halikiopoulou, Froio, Van Kessel, De Lange, Mudde and Taggart2024), trust in government (1–4), and trust in mainstream media (1–4). A final model replaces these measures with respondents’ self-placements on the left-right and liberal-conservative ideological dimensions (0–10).

Three estimands are of interest: the association between each covariate and the outcome in (1) the Czech Republic and in (2) Slovakia; and (3) the difference between these two associations. To recover these estimands, I interact all covariates with an indicator variable for country (Czech Republic serves as the reference). I estimate readership using logistic regression and trust using OLS, using HC2 robust standard errors (see Online Appendix B for full regression results).

5.2.1 Readership

Figure 4 presents results from the first readership model as group-average marginal effects (with 95% confidence intervals). Brackets show the magnitude of cross-country differences in estimates and associated p-values. Older individuals are more likely to be disinformation consumers in both countries, and though the cross-country difference in this estimate is statistically significant, it is substantively negligible. In Slovakia, women are about five percentage points more likely than men to consume disinformation, while in the Czech Republic the pattern flips. The roughly ten-point gap between countries is statistically significant. Slovaks with a university degree are about 20 percentage points less likely to consume disinformation than those with only a primary education, and roughly six points less likely than those with a secondary education. In contrast, education has a statistically negligible association with disinformation readership in the Czech Republic. Locality size had little association with disinformation consumption in either country. CRT scores had no relationship with disinformation readership in the Czech Republic, but were negatively associated with readership in Slovakia. Each additional correct answer on the CRT predicted approximately a three-percentage-point decrease in the probability of reading disinformation news sources. Finally, political knowledge was linked to disinformation readership in both countries, with associations of similar magnitude. The politically knowledgeable were six to nine percentage points more likely to be disinformation readers. This finding is worth emphasizing as it suggests disinformation readership is not simply explained by a lack of political knowledge. Instead, other factors – such as motivated reasoning, media distrust, or active information-seeking behavior – may lead the politically knowledgeable to consume disinformation.

Figure 4

Demographics, cognitive reflection, political knowledge, and disinformation readership

Figure 5 presents group-average marginal effects estimated from the second model. While trust in government had little association with disinformation readership in the Czech Republic, it had a positive association in Slovakia. The positive relationship likely reflects the ideological convictions of supporters of the economically left-wing but socially conservative (Smer-led) governing coalition in power at the time of the survey. Trust in mainstream media was negatively associated with disinformation readership among Czechs, but showed no significant association among Slovaks. Populist voters in both countries were also more likely to consume disinformation than the supporters of non-populist parties. The size of the association – about 10 percentage points – was statistically indistinguishable between the two countries. Finally, geopolitical attitudes were also substantively associated with disinformation readership. Individuals who believed their country belongs to the West were less likely to consume disinformation than those favoring alignment with the East or somewhere in between. In Slovakia, this difference ranged from 12 (between vs. West) to 15 percentage points (East vs. West), while in the Czech Republic it ranged from three to nine percentage points.

Figure 5

Institutional trust, populist support, geopolitics, and disinformation readership

Figure 6 presents associations between self-placements on the economic and sociocultural dimensions and disinformation readership. In the Czech Republic, both dimensions had statistically significant associations with disinformation readership. Individuals with right-leaning economic views were less likely to consume disinformation. A one-unit shift toward right-leaning economic positions decreased the probability of being a disinformation reader by just over one percentage point. Conversely, each one-unit shift toward social conservatism increased the probability of consuming disinformation by more than three percentage points. In Slovakia, left-right economic positioning was not associated with disinformation readership. However, each one-unit shift toward social conservatism was associated with a nearly four-percentage-point increase in the probability of being a disinformation reader. Overall, sociocultural conservatism was the stronger predictor of disinformation readership in both countries. The negative association between right-leaning economic positions and disinformation consumption in the Czech Republic may reflect the fact that socioculturally liberal Czech parties tend to maintain right-leaning economic programs.

Figure 6

Ideology and disinformation readership

5.2.2 Trust

I follow the same analytical strategy to assess predictors of disinformation trust. Figure 7 presents the first set of results. Age had a statistically significant and positive association with trust in disinformation in Slovakia, but not in the Czech Republic. However, the association was substantively small: A ten-year increase in age corresponded to only a 0.15-point increase in trust. In both countries, women were more likely than men to trust disinformation, and the association was of similar magnitude. Having a university education was associated with lower trust in disinformation news sources. However, in the Czech Republic, this was only the case relative to primary education, whereas in Slovakia, university-educated respondents trusted disinformation less than both those with primary and secondary education. Moreover, the negative association between university education (relative to primary education) and disinformation trust was substantively larger in Slovakia – approximately 0.45 compared to 0.15 in the Czech Republic. In both countries, respondents from larger cities were less trusting of disinformation than those from smaller cities and towns. While each additional correct CRT answer was associated with a 0.08-point decrease in disinformation trust in Slovakia, CRT performance had no association with trust in the Czech Republic. Finally, political knowledge had little association with trust in disinformation in either country.

Figure 7

Demographics, cognitive reflection, political knowledge, and disinformation trust

Figure 8 presents group-average marginal effects for the second set of covariates. Trust in government is predictive of trusting disinformation in Slovakia, but not in the Czech Republic. However, each one-unit increase in trust in mainstream media was associated with a rise in trust in disinformation: 0.08 in the Czech Republic and 0.15 in Slovakia. In the Czech Republic, the difference in disinformation trust between populist and non-populist voters was substantively large at 0.31. In Slovakia, the association was both substantively and statistically smaller – approximately half the magnitude. In Slovakia, geopolitical orientation had a pronounced association with trust in disinformation, with a 0.45 difference in disinformation trust between those who viewed Slovakia as belonging to the West relative to those who thought Slovakia should belong to the East or somewhere in-between. In the Czech Republic, the relationship between geopolitical views and trust in disinformation was less clear-cut. Individuals who thought the Czech Republic should belong to the West were neither more nor less trusting of disinformation than those who thought it should belong to the East. However, this null finding may reflect limited statistical power, as only 105 respondents selected the East option. Supporting this interpretation, respondents with geopolitically Western-leaning views exhibited significantly lower trust in disinformation than those who placed the country somewhere in between. Finally, in both countries, there was little evidence of statistically significant differences in disinformation trust between individuals who favored the East versus those who preferred a position in between East and West.

Figure 8

Institutional trust, populist support, geopolitics, and disinformation trust

Finally, Figure 9 shows that left-right and liberal-conservative ideology both have substantive, statistically significant relationships with trust in disinformation outlets, and that these relationships are largely similar across the two countries. First, every one-unit shift toward the right decreases trust in disinformation by 0.06 in the Czech Republic and 0.04 in Slovakia. Second, conservatives were more trusting of disinformation than liberals. The association was substantively larger than that of left-right economic ideology, with a one-unit shift toward conservatism corresponding to a 0.09 increase in trust in the Czech Republic and 0.075 in Slovakia.

Figure 9

Ideology and disinformation trust

5.3 Discussion

Disinformation media are less recognized and read than mainstream media; yet, over 20 percent of Czechs and over 40 percent of Slovaks still report reading at least one disinformation news source. While disinformation sources are generally less trusted than mainstream sources, this difference disappears among actual consumers. In other words, individuals who read disinformation and those who read mainstream media have similar levels of trust in the sources they consume. Importantly, trust among disinformation readers did not differ significantly between the two countries, suggesting that selection into disinformation produces similar effects on trust across diverse national contexts.

Analyzing the predictors of disinformation readership and trust provides correlational evidence for ideologically motivated selection into news consumption. Conservative, populist voters who believe that their country should belong to the East were more likely than their liberal counterparts to be disinformation readers. The substantial predictive purchase of sociocultural and geopolitical attitudes underscores that individuals trust disinformation because it accords with their preexisting ideological commitments. More substantively, trust in media is not necessarily a function of its objective credibility, but rather of the extent to which it produces content that affirms ideological and partisan convictions.

Most predictors show similar directional relationships with disinformation readership and trust in both the Czech Republic and Slovakia. However, the magnitude of these associations often varies. This pattern suggests that while the politically motivated underpinnings of disinformation consumption operate consistently across contexts, variation in factors that make a given country more or less susceptible to disinformation – for example, ideological polarization, anti-disinformation capacity, elite reinforcement – may amplify or attenuate these individual-level processes. In other words, individuals across countries select into disinformation for the same reasons: doing so affirms their preexisting ideological and attitudinal convictions. However, differences in countries’ political environments condition the strength of this relationship, as well as the extent to which “protective” factors like education and analytical reasoning influence disinformation readership and trust.

In the broader context of CEE, the Czech-Slovak comparison suggests that conservative sociocultural positions and anti-Euro-Atlantic orientations are likely to correlate with disinformation consumption across the region. However, the findings also suggest that the relationship between individual attitudes and disinformation consumption may be stronger in countries resembling Slovakia – those characterized by high ideological polarization, electorally successful radical populist parties, and weak institutional capacity for countering disinformation.

Several limitations are worth highlighting. First, the survey did not measure readership frequency. Although the question was phrased to suggest regular engagement – using “do you read/watch” rather than “have you ever read/watched” – capturing habitual consumption is essential for understanding belief in disinformation (Arendt et al., Reference Arendt, Northup and Camaj2019; Pennycook et al., Reference Pennycook, Cannon and Rand2018). Second, the survey measured disinformation consumption by focusing on readership of disinformation media outlets. While disinformation media are certainly one prominent pathway through which disinformation is disseminated, they are by no means the only pathway. Disinformation also spreads through social media, on podcasts, and in interpersonal conversations. Though it is unlikely that the predictors of disinformation consumption differ dramatically across these channels, future research may wish to account for both the frequency of consumption and the different channels of consumption. Third, the survey relied on an online, opt-in panel of respondents. Given the potential threats to representativeness – especially as it pertains to media diets – future studies may wish to replicate these analyses with samples recruited using face-to-face methods.

6.1 Belief in True and False Headlines

Across all false treatment arms, Czech respondents evaluated sixteen false headlines, while Slovak respondents evaluated fourteen. Table 8 presents the most credible false headlines in the Czech Republic, while Table 9 does the same for Slovakia. For Slovakia, the analysis includes headlines from both the 2021 and 2024 surveys. While changes in the information environment between the surveys may limit the comparability of belief evaluations, the ranking still provides a useful overview of which disinformation narratives Slovaks found most and least credible.

Table 8Most and least believed disinformation (Czech Republic)

Note: Means presented with 95% confidence intervals; % True indicates percent of respondents who indicated that a headline was “likely” (3) or “completely true” (4).

Table 9Most and least believed disinformation (Slovak Republic)

Note: Means with 95% confidence intervals; % True indicates percent indicating a headline was “likely” (3) or “completely true” (4); Belief adjusted for certainty was not asked in 2021 Slovakia survey; if ranking by adjusted belief the least-believed category includes “Ukrainians who really need help flee to Russia, not to the EU.” (-39.34) and “There is no war in Kyiv. People there live normally and peacefully.” (−19.11).

In both countries, the most widely believed disinformation headlines related to Ukrainian refugees. In the Czech Republic, the three most-believed headlines were assessed as likely true or true by over 50 percent of respondents. Evaluations were similar in Slovakia, where only the third-most-believed headline dipped below 50 percent. Mean belief scores averaged around 2.5, the scale’s midpoint. When adjusted for certainty, belief scores in the Czech Republic hovered around zero, indicating substantial uncertainty about the truthfulness of even the most-believed headlines. In Slovakia, the two most believed headlines had adjusted belief scores of 15.68 and 11.80, suggesting more certainty. The third-most believed headline in both countries had a negative adjusted belief score, indicating that respondents, while uncertain, leaned toward believing the headline to be false. More importantly, despite some uncertainty, respondents assessed the remaining thirteen headlines in the Czech Republic and the remaining eleven headlines in Slovakia as more false than true.

In the Czech Republic, the least-believed headline related to the Russo–Ukrainian War. Headlines about Ukrainians seeking refuge in Russia and the EU adding insect protein to groceries rounded out the bottom three. Belief scores for these headlines averaged around 1.70, adjusted belief scores hovered around −50, and only 13 to 15 percent of respondents considered the headlines likely or completely true. In Slovakia, the two least-believed headlines (both from the 2021 survey) focused on the EU’s interference with traditional values and its promotion of LGBTQ+ rights. The third least-believed headline was the same as the Czech Republic’s least-believed claim: Russia’s purported right to invade Ukraine. Comparing the least-believed headlines in the 2024 surveys indicates that Czechs were more certain about the falsity of headlines than Slovaks. Ranking by adjusted belief (and thus excluding the 2021 EU-related headlines), Slovakia’s three least-believed headlines all had higher belief ratings than the Czech Republic’s three least-believed headlines.

Next, I test whether respondents evaluated true and false headlines differently, and whether showing true and false headlines together influenced these evaluations – an important test of ecological validity, as individuals’ real-world news consumption entails exposure to both true and false narratives. Three of the six experiments included a combined true+false condition, enabling: (1) a within-subjects analysis of how respondents assessed false versus true headlines; and (2) a between-subjects analysis comparing evaluations of false headlines presented in isolation with evaluations of false headlines presented together with true headlines. I preregistered the following hypotheses:

To test Hypothesis 3a, I construct three analytical samples from respondents assigned to the three true+false treatment arms present in the experiments. This arm was included in the 2021 Slovakia EU experiment ($N=323$), the 2024 Czech Republic Russo–Ukrainian War experiment ($N=598$), and the 2024 Slovakia Ukrainian refugee experiment ($N=595$). I restructure the data so the unit of analysis is the respondent-headline evaluation: eight evaluations per respondent in the 2021 experiment and ten per respondent in the 2024 experiments. I then create an indicator for headline type (true vs. false). I model belief (as well as the other evaluations) using a linear mixed model with random intercepts for respondents and headlines, and random slopes for headline type across respondents. Respondent intercepts allow each individual their own baseline evaluation and account for within-person correlation; headline intercepts account for inherent differences in believability across headlines. Random slopes allow the effect of headline type to vary across respondents.

Figure 11 presents the standardized effect of a headline being false (relative to true) on evaluations. The left panel analyzes headlines related to the Russo–Ukrainian War in the Czech Republic, and shows that respondents expressed less belief in false headlines than in true ones. Substantively, the difference amounted to 0.75 standard deviations in both the unadjusted and adjusted belief measures. The middle panel focuses on EU-related headlines in Slovakia. There was little evidence that respondents evaluated true and false headlines differently (though conclusive evidence is limited by the absence of the adjusted belief measure). Finally, the third panel, focused on refugee-related headlines in Slovakia, provides mixed evidence. Unadjusted belief did not differ between true and false headlines at conventional levels of statistical significance; however, there was a statistically significant difference in adjusted belief evaluation (with the effect size amounting to just over one-third of a standard deviation). Importantly, there was little evidence that headline type influenced willingness to share, recognition, and registration intent. In sum, the analysis provides mixed evidence in support of Hypothesis 3a. Importantly, since the combined true+false treatment arm was present in only three of the six experiments, findings are limited to the three corresponding disinformation topics. Given the mixed findings, future research may wish to evaluate generalizability by replicating this design across additional topics.

Figure 11

False headline effects (v. true) on evaluations

To evaluate Hypothesis 3b, I compare respondents’ belief in false headlines presented in isolation to respondents’ belief in false headlines that are presented together with true headlines. I pool respondent evaluations of false headlines in the false group and the true+false group, respectively. The random assignment of respondents to these between-subjects conditions allows me to estimate how evaluations change when true and false headlines are presented together (note that fatigue effects may bias this estimate given the greater number of evaluations in the true+false condition). I assess statistical differences using unpaired t-tests with Benjamini–Hochberg corrected p-values (adjustments based on four tests in SK 2021 and five tests in CZ 2024 and SK 2024). Across all three topics, respondents evaluated false headlines presented in isolation similarly to false headlines presented together with true headlines (Table 10). In fact, only one statistically significant difference emerged: Slovak respondents evaluated false headlines about Ukrainian refugees as less familiar in the combined condition than when the same headlines were presented in isolation.

Table 10False headline evaluations: isolation vs. with true headlines

Note: ${}^{*}$ indicates difference from False at $p<0.05$; p-values adjusted with Benjamini–Hochberg correction (5 tests for 2024 experiments; 4 tests for 2021 experiment); $N=592$, 598, 344, 323, 605, 595.

6.1.1 Motivated Reasoning, Analytical Reasoning, and Headline Evaluations

Individuals are more likely to believe disinformation when it confirms their preexisting attitudes and ideological commitments (Batista Pereira et al., Reference Batista Pereira, Bueno, Nunes and Pavão2022; Mazepus et al., Reference Mazepus, Osmundsen, Bang-Petersen, Toshkov and Dimitrova2023; Vegetti and Mancosu, Reference Vegetti and Mancosu2020). However, whether a stronger propensity to engage in analytical reasoning (as measured by performance on the cognitive reflection test) exacerbates or attenuates directionally motivated reasoning is unclear (Bago et al., Reference Bago, Rand and Pennycook2022; Pennycook and Rand, Reference Pennycook and Rand2019; Tappin et al., Reference Tappin, Pennycook and Rand2020 versus Roozenbeek et al., Reference Roozenbeek, Maertens, Herzog, Geers, Kurvers, Sultan and van der Linden2022a). Evidence in this debate largely comes from the American context. However, recent cross-national studies (Gehle et al., Reference Gehle, Hameleers, Tulin, de Vreese, Aalberg, Van Aelst, Cardenal, Corbu, van Erkel, Esser and Halagiera2024; Tulin et al., Reference Tulin, Hameleers, de Vreese, Aalberg, Corbu, Van Erkel, Esser, Gehle, Halagiera, Hopmann and Koc-Michalska2024), as well as evidence from Hungary (Faragó et al., Reference Faragó, Krekó and Orosz2023) and Ukraine (Erlich et al., Reference Erlich, Garner, Pennycook and Rand2023), also suggest that analytical reasoning is associated with a better ability to identify disinformation. Given this debate, I preregistered the following hypotheses:

Hypothesis 4: Respondents will express stronger belief in false headlines when the disinformation content aligns with their ideological convictions.

Hypothesis 5: Respondents with lower cognitive reflection scores will express stronger belief in false headlines than respondents with higher cognitive reflection scores.

As before, I test these hypotheses with the sample of respondents randomly assigned to the true+false experimental conditions. As a first test of Hypothesis 4, I consider whether variations in attitudes toward Russia influence evaluations of disinformation related to the Russo–Ukrainian War in the Czech Republic ($N=598$). I measure these attitudes with a pretreatment question – that is, one asked before exposure to the headlines – about stances toward the Russian Federation. Answers were measured on a 1 to 4 scale, ranging from “strongly negative” to “strongly positive.” I collapse the “negative” answers (1 and 2) and the “positive” answers (3 and 4) into two categories. I then estimate the same linear mixed model as above, but add an interaction between headline type and the binary Russia attitudes measure. I calculate p-values using the Benjamini–Hochberg adjustment for multiple comparisons. The estimand of interest is the difference in evaluations of the same headline set between attitudinal subgroups – for example, the difference in evaluations of false headlines between individuals with anti-Russia and pro-Russia attitudes.

Figure 12 displays results as predicted values, with the gray brackets and text indicating the size of the difference between subgroups and the (adjusted) p-values. Results indicate clear evidence of ideological moderation. Pro-Russia respondents exhibited greater belief in false headlines than anti-Russia respondents. The difference was substantively large – amounting to over one point on the belief scale, and over 75 points on the adjusted belief scale. While there was a statistical difference in adjusted belief evaluations of the true headlines, it was substantively small and did not replicate with the unadjusted belief measure. Respondents evaluated true headlines similarly, regardless of their positions on Russia. The other evaluations also provide evidence of ideologically motivated information processing. Respondents with more positive attitudes toward Russia were more likely than anti-Russia respondents to share both disinformation and true headlines. Likewise, these respondents indicated that they were more familiar with both headline sets.

Figure 12

War headline evaluations across Russia attitudes (Czech Republic)

Next, I consider whether ideologically motivated information processing also conditions belief in headlines related to Ukrainian refugees in Slovakia ($N=595$). As the relevant ideological moderator, I use a pretreatment measure of immigration attitudes. The measure asked respondents to evaluate the influence of immigrants on Slovakia (1 = “very negative” to 4 = “very positive”) and was dichotomized to create anti-immigration and pro-immigration subgroups.

Figure 13 mirrors the previous analysis. Individuals with attitudinal predispositions aligned with disinformation content – in this case, individuals with anti-immigrant views – were more likely to believe false headlines. Differences were substantial in magnitude, amounting to 0.71 and 53 points on the respective belief scales. Pro-immigration respondents were slightly more likely to believe true headlines than anti-immigration respondents, though the difference was substantively small. Unlike with the Russo–Ukrainian War headlines, preexisting attitudinal commitments did not condition the extent to which individuals were willing to share or recognized the different headline sets.

Figure 13

Refugee headline evaluations across immigration attitudes (Slovakia)

Next, to test Hypothesis 5, I evaluate the relationship between analytical reasoning and belief in false headlines. I also explore how analytical reasoning interacts with ideologically motivated information processing. While the existing literature agrees that analytical reasoning matters for how individuals consume and respond to disinformation, it has yet to come to a conclusion about how analytical reasoning conditions the effects of directionally motivated reasoning. On the one hand, System 2 analytical reasoning may amplify motivated reasoning because individuals deploy their analytical capabilities to arrive at attitude-consistent conclusions. On the other hand, if individuals rely on System 2 reasoning rather than System 1 intuitive information processing, they may be more discerning consumers of information, and thus more able to distinguish between true and false content (Pennycook and Rand, Reference Pennycook and Rand2019).

I again rely on the subsample of respondents assigned to the true+false condition in the 2024 Czech and Slovak surveys. Using the same linear mixed model setup, I regress belief evaluations on binary indicators for false versus true headlines and for attitude consistency versus inconsistency, and CRT scores. To identify conditional relationships, I include a three-way interaction between these variables, along with all constituent two-way interactions. For robustness, I model both belief measures. Two estimates are of interest. First, the interaction between headline type and CRT scores tests Hypothesis 5. Second, the three-way interaction between headline type, attitudinal commitments, and CRT scores tests how analytical reasoning and preexisting attitudes jointly condition headline evaluations. Table 11 presents results.

Table 11Ideology, analytical reasoning, and belief

DV: Belief (1 to 4), Adjusted Belief (−100 to 100); * $p<0.05$; ** $p<0.01$; $p$-values adjusted with Benjamini–Hochberg correction (32 tests)

Analytical reasoning clearly shapes belief in disinformation headlines. The statistically significant and negative interaction term between false headlines and CRT indicates that as cognitive reflection scores increase, the gap in belief between false and true headlines widens. Or, in simpler terms, higher cognitive reflection guards against believing disinformation. Substantively, in the Czech sample, a one-unit increase in cognitive reflection scores enhances discernment between false and true headlines by roughly 36 percent for unadjusted belief and 40 percent for adjusted belief. In the Slovak sample, the effects are even larger at 41 and 47 percent, respectively.

The triple interactions – which account for how preexisting attitudes moderate the relationship between cognitive reflection and belief in false headlines – suggest that the “protective” effect of cognitive reflection is stronger among individuals whose preexisting attitudes conflict with disinformation content. In other words, analytical reasoning is a stronger protective buffer against believing false headlines when preexisting attitudes do not align with the headline’s content. Substantively, among individuals with attitudes inconsistent with disinformation content, each additional point on the CRT reduces belief in false headlines by 0.09 points on the unadjusted belief scale. For those whose attitudes align with disinformation, cognitive reflection has virtually no effect (−0.005 points).

To clarify substantive quantities, I plot conditional predicted values of belief estimated from the Czech (Figure 14) and Slovak (Figure 15) unadjusted belief models. Left panels plot predicted belief in false headlines across CRT scores, while right panels do the same for belief in true headlines. Dashed lines indicate respondents whose preexisting attitudes were inconsistent with the headlines, while solid lines indicate respondents with attitudes consistent with the headlines. The left panels show that belief in false headlines decreases as a function of CRT scores among respondents for whom disinformation is attitude inconsistent. In both countries, a one-unit increase in CRT scores corresponds to roughly a 0.09 decrease in predicted belief. Finally, the right panels show that CRT scores and preexisting attitudes have little association with predicted belief in true headlines.

Figure 14

Predicted belief: Attitude consistency and CRT (Czech Republic)

Figure 15

Predicted belief: Attitude consistency and CRT (Slovakia)

6.2 The Effects of Fact-Checking

The preceding analyses focused on the extent to which individuals believe disinformation news headlines and on the individual-level characteristics associated with this belief. This section investigates whether fact-checking and prebunking can correct misinformed beliefs. Fact-checking and prebunking can reduce misperceptions (e.g., Martel and Rand, Reference Martel and Rand2023; Roozenbeek et al., Reference Roozenbeek, van der Linden, Goldberg, Rathje and Lewandowsky2022b), though questions about the durability of effects (Carey et al., Reference Carey, Guess, Loewen, Merkley, Nyhan, Phillips and Reifler2022) and ideologically asymmetrical responses remain (DeVerna et al., Reference DeVerna, Guess, Berinsky, Tucker and Jost2024). Importantly, recent work has also broadened the study of these interventions to multiple contexts (e.g., Arechar et al., Reference Arechar2023; Porter and Wood, Reference Porter and Wood2021), occasionally finding limited correction effects (Altay et al., Reference Altay2025; Batista Pereira et al., Reference Batista Pereira, Bueno, Nunes and Pavão2022; Porter et al., Reference Porter, Scott, Wood and Zhandayeva2024).

To assess, two experiments – one in the 2024 Czech survey and one in the 2024 Slovak survey – included a fact-checking condition. This condition included the same headlines as the false condition, but also presented detailed fact-checks. These fact-checks combined labeling – warning individuals that independent fact-checkers have identified a given piece of information as false (Brashier et al., Reference Brashier, Pennycook, Berinsky and Rand2021) – with corrections – information that explains why the information is misleading (Porter et al., Reference Porter, Scott, Wood and Zhandayeva2024). The fact-checks were sourced from the third-party fact-checking organizations listed in Online Appendix A. Using real-world fact-checks ensures that the experiment measures responses to actual anti-disinformation efforts being deployed in both countries. I preregistered the following hypothesis:

In the experiments, Czechs evaluated anti-Ukrainian refugee disinformation, while Slovaks evaluated disinformation related to the Russo–Ukrainian War. These topics were selected to investigate the efficacy of fact-checking on the most salient disinformation issue in each country. While anti-Ukrainian refugee sentiment is on the rise in both countries, the issue is more contentious in the Czech Republic (Jelínková et al., Reference Jelínková, Plaček and Ochrana2024, 2), where far-right politicians openly question state support for refugees. In Slovakia, attitudes toward the War are more contested, and sympathies for Russia are among the strongest in Europe (Učeň, Reference Učeň2023). Figure 16 presents examples of the fact-check stimulus alongside the false stimulus from both countries.

Figure 16

False and fact-checked headlines

Table 12 compares headline evaluations for the false and fact-check conditions in both experiments. Fact-checking had little impact. The sole exception was in the Czech Republic, where fact-checks reduced participants’ expressed familiarity with the false headlines. An additional analysis, presented in Online Appendix C, analyzes fact-checking at the headline level, confirming the pooled analysis. Fact-checks did not reduce belief in any of the ten headlines presented across the two experiments.

Table 12Effects of fact-checking

Note: ${}^{*}$ indicates difference from False at $p<0.05$; p-values adjusted with Benjamini–Hochberg correction (5 tests); $N=589$, 596, 605, 595.

6.3 The Effects of Prebunking

Prebunking is also an effective intervention for increasing disinformation resilience (e.g., Roozenbeek et al., Reference Roozenbeek, van der Linden, Goldberg, Rathje and Lewandowsky2022b). Prebunking is motivated by inoculation theory – the idea that preemptive warnings about disinformation strategies can reduce individuals’ vulnerability to misleading narratives (Cook et al., Reference Cook, Lewandowsky and Ecker2017). Importantly, while prebunking can effectively lower belief in false headlines, it may also increase scepticism of true headlines (Hoes et al., Reference Hoes, Aitken, Zhang, Gackowski and Wojcieszak2024). Thus, I predicted:

To test, the Green Deal disinformation experiment in the 2024 Czech survey included five between-subject conditions: the standard control, false, and true conditions, plus two additional prebunking conditions where a prebunking message appeared prior to either the false or true headlines. The prebunking warning was adopted directly from content used by the Czech Centre Against Hybrid Threats (Figure 17).Footnote ¹⁰⁵ The warning defines disinformation and offers several tips for identifying potentially misleading content.

Figure 17

Prebunking stimulus: Czech Centre Against Hybrid Threats

Table 13 presents evaluations pooled across headlines within each condition. As in the other experiments, respondents had little difficulty distinguishing between false and true headlines. However, there was little evidence that the prebunking intervention influenced headline evaluations. Respondents who evaluated false headlines provided statistically equivalent assessments regardless of whether they received the prebunking warning. More optimistically, the prebunking warning did not make respondents more sceptical of true headlines, which were also evaluated similarly with and without the warning. I confirm these results using respondent-headline-evaluations as the unit of analysis in Online Appendix C.

Table 13Environment headline evaluations (Czech Republic)

Note: ${}^{*}$ indicates a difference from False at $p<0.05$; p-values adjusted with Benjamini–Hochberg correction (20 tests); $N=486$, 470, 491, 484.

6.4 Discussion

Across two countries, two time periods, and multiple disinformation topics, this analysis explored how individuals evaluate disinformation headlines. Individuals were generally adept at distinguishing between false and true headlines. However, some narratives were much more believed than others. This variation is likely a product of issue salience and politicization. For example, the most believed disinformation headlines in both the Czech Republic and Slovakia concerned Ukrainian refugees – a salient, politically polarized issue. Importantly, these assessments did not differ based on whether false headlines were presented in isolation or blended together with thematically similar, but completely true, headlines.

Preexisting attitudinal and ideological commitments played an important role in how individuals evaluated headlines. Individuals for whom disinformation was attitude-consistent demonstrated significantly higher levels of: (1) belief in false headlines, (2) willingness to share such content, and (3) familiarity with the disinformation than individuals for whom disinformation was attitude-inconsistent. These patterns were consistent across countries and occurred regardless of differences in baseline beliefs across disinformation topics. In other words, no matter the disinformation topic or how believable or unbelievable the claims were, ideological and attitudinal alignment consistently predicted higher levels of belief. In sum, these findings provide compelling evidence in support of directionally motivated information processing.

The analyses also evaluated how analytical reasoning influences headline evaluations, both independently and in conjunction with directionally motivated information processing. On average, higher levels of System 2 analytical reasoning were associated with lower belief in false headlines. More importantly, these effects were almost entirely localized to individuals whose preexisting attitudes conflicted with disinformation content. Individuals ideologically predisposed to agree with false headlines believed the headlines regardless of their analytical reasoning capacity. In sum, analytical reasoning does not uniformly guard against disinformation, nor does it amplify belief in congenial falsehoods. Rather, the relationship is asymmetric: analytical reasoning reduces belief in attitude-incongruent disinformation but has little effect when prior attitudes align with false content. Preexisting ideological commitments attenuate the protective benefits of analytical reasoning.

Finally, the analyses explored the effectiveness of fact-checking and prebunking as potential strategies for reducing belief in false headlines. In both the Czech Republic and Slovakia, and across two different disinformation topics, fact-checking had little effect on respondent assessments. Likewise, prebunking, tested only in the Czech Republic, also did not influence respondent evaluations. As such, these results echo a small set of studies documenting mixed or limited effects of fact-checking beyond the commonly studied Western contexts (Batista Pereira et al., Reference Batista Pereira, Bueno, Nunes and Pavão2022; Porter et al., Reference Porter, Scott, Wood and Zhandayeva2024). There are several potential explanations for the null findings. First, the disinformation used in the fact-checking experiments covered highly salient topics. Thus, the fact-checks may not have presented sufficiently novel information to alter beliefs (Ecker and Ang, Reference Ecker and Ang2019; Tulin et al., Reference Tulin, Hameleers, de Vreese, Aalberg, Corbu, Van Erkel, Esser, Gehle, Halagiera, Hopmann and Koc-Michalska2024). Second, the surveys did not measure individuals’ levels of trust in fact-checkers, which is an important determinant of fact-checking’s effectiveness (Bruns et al., Reference Bruns, Dessart, Krawczyk, Lewandowsky, Pantazi, Pennycook, Schmid and Smillie2024; Martel and Rand, Reference Martel and Rand2024). Finally, fact-checking and prebunking are most effective when used in combination (Pennycook et al., Reference Pennycook, Berinsky, Bhargava, Lin, Cole, Goldberg, Lewandowsky and Rand2024; Roozenbeek et al., Reference Roozenbeek, van der Linden, Goldberg, Rathje and Lewandowsky2022b). Thus, in isolation, the prebunking and fact-checking interventions may not have been strong enough to alter respondents’ beliefs.

7.1 Does Disinformation Influence Attitudes?

The experiment fielded in the 2021 Slovakia survey focused on disinformation related to the EU’s propagation of socioculturally liberal values and policies ($N=1,560$, attention check failures included). The experiment considers effects on three attitudinal outcomes: (1) Euroscepticism; (2) generalized outgroup prejudice; and (3) nationalist policy preferences. Each outcome represents a core public opinion target for Kremlin-linked disinformation: support for the EU, identity-based grievances with non-dominant groups, and support for those who seek to reaffirm the rights of the titular majority (and also tend to be sympathetic toward Russia).

I operationalize each outcome as a preregistered mean index of several questions. For EU-related attitudes, respondents indicated what feelings the EU evokes for them, whether Slovakia should leave the EU, and whether Slovakia benefits from EU membership (1–4; Cronbach’s $\alpha =0.88$). Three questions asked respondents about the extent to which they agree that some groups or cultures are better than others and operationalize generalized outgroup prejudice (1–4; $\alpha =0.82$). Finally, for nationalist policy preferences, four questions asked whether respondents think that the rights of Slovaks should be privileged over the rights of other groups (1–4; $\alpha =0.66$). Higher values indicate stronger preferences – that is, more positive EU attitudes, stronger outgroup prejudice, and stronger nationalist policy preferences. I preregistered the following hypothesis:

Figure 18 presents intent-to-treat effects associated with the false, true, and true+false headline sets. The control group serves as the reference category. Across all three outcomes, false headlines evoked attitudinal backlash, though the effect on EU attitudes only approached conventional levels of statistical significance (p = 0.066). In other words, relative to the control group, individuals exposed to disinformation messages were more pro-EU, indicated lower support for nationalist policy preferences, and were less prejudiced. Standardized effect sizes (Cohen’s d) range from small (0.15) to moderate (0.32). True headlines had similar backfire effects for outgroup prejudice, but not for EU attitudes or nationalist policy preferences. Finally, there was little statistical evidence that the combined true and false headlines influenced EU attitudes and nationalist policy preferences. However, the combined headline set did weaken outgroup prejudice. In sum, false headlines, whether presented in isolation or together with true headlines, tended to cause attitudinal backlash rather than achieve their intended persuasive effect.

Figure 18

Effects of EU-related disinformation (Slovakia)

Next, I analyze the effects of disinformation related to the Russo–Ukrainian War. Specifically, I consider whether headlines affected the extent to which respondents thought their country should support the Ukrainian war effort, whether respondents saw Russia’s aggression as justified, and whom they blamed for starting the war. To measure the first two outcomes, I use two preregistered indices (Cronbach’s $\alpha =0.86$ and 0.89 in the Czech Republic, and 0.85 and 0.90 in Slovakia), with higher values indicating stronger pro-Ukraine positions. The final outcome is a preregistered binary indicator of whom respondents blame for the war (“the West/NATO” or “Ukraine” coded as 0; “Russia” coded as 1). Higher values for all measures indicate stronger pro-Ukrainian positions. I preregistered the following hypotheses:

Figures 19 and 20 present results for the Czech ($N=\mathrm{2,406}$) and Slovak ($N=\mathrm{2,409}$) samples. There is little statistical evidence of false headlines impacting attitudes in the Czech Republic, and none of the headline sets influenced attitudes in Slovakia. However, in the Czech Republic, both true headlines and the combined set of true and false headlines pushed attitudes in a pro-Ukraine direction. While these effects were substantively small – ranging from 0.10 to 0.15 of the outcomes’ standard deviations – they indicate that exposing individuals to factual information about the war can bolster support for Ukraine.

Figure 19

Effects of Russo–Ukrainian War disinformation (Czech Republic)

Figure 20

Effects of Russo–Ukrainian War disinformation (Slovakia)

The next set of results focuses on disinformation related to Ukrainian refugees. For these experiments, I measured effects on three preregistered outcomes: an index that measures the extent to which Ukrainian refugees are a burden or benefit for the Czech Republic/Slovakia (1–4; $\alpha =0.94$ in the Czech Republic and Slovakia), a single-item measure of cultural distance between Ukrainians and Czechs/Slovaks (1–4), and a single-item feeling thermometer measuring personal evaluations of Ukrainian refugees (0–100). I preregistered that:

Figures 21 and 22 plot intent-to-treat effects. As with the war disinformation headlines, results indicate evidence of attitudinal backlash, not direct persuasion. Compared to respondents in the control group, respondents in the false headline group considered Ukrainians more of a benefit to their host country (Czech Republic p = 0.029, Slovakia p = 0.047) and indicated warmer feelings toward them (Czech Republic p = 0.011, Slovakia $p<0.001$). However, there was little evidence that the false headlines shaped perceptions of cultural distance. Backlash was also present in the Czech fact-checking condition (burden versus benefit p = 0.022; thermometer p = 0.051), and in the Slovak true+false condition, but only for the thermometer outcome (p = 0.008). Standardized effect sizes were substantively small, hovering around 0.15 for the burden versus benefit outcome, and between 0.15 and 0.20 across the different conditions for the feeling thermometer. Finally, there was little statistical evidence that true headlines affected attitudes.

Figure 21

Effects of Ukraine refugee disinformation (Czech Republic)

Figure 22

Effects of Ukraine refugee disinformation (Slovakia)

The last experiment focuses on the effects of disinformation related to the European Green Deal in the Czech Republic. I evaluate effects on three preregistered attitudinal indices: (1) EU attitudes ($\alpha =0.90$); (2) perceptions of whether the Czech Republic has a say in EU policymaking ($\alpha =0.88$); and (3) a climate change assessment index that captures individuals’ beliefs about the occurrence of climate change, their perceptions of its negative impacts, and their evaluations of whether current responses are adequate to prevent adverse effects ($\alpha =0.84$; adapted from de Graaf et al., Reference de Graaf, Stok, de Wit and Bal2023). I preregistered the following hypotheses:

Figure 23 presents intent-to-treat effects. Across all attitudinal outcomes and experimental conditions, effects are precisely estimated near zero. In other words, headlines about the European Green Deal – whether true, false, or accompanied by a prebunking warning – had a statistically negligible impact on attitudes toward the EU, assessments of whether the Czech Republic has a say in EU policymaking, and climate change scepticism.

Figure 23

Effects of EU Green Deal disinformation (Czech Republic)

A final set of analyses examines whether directionally motivated reasoning and analytical thinking explain these results. For each experiment, I test whether treatment effects are moderated by respondents’ prior attitudes and their performance on the CRT. I present results in Online Appendix D. Overall, there is little evidence that preexisting attitudes or analytical reasoning shape how individuals respond to disinformation headlines. While some interactions reach conventional levels of statistical significance, there are few clear patterns across either disinformation topics or countries. Moreover, most p-values would exceed 0.05 if adjusted for multiple comparisons, and the interaction analyses – especially those considering the triple interaction between treatment, ideology, and analytical reasoning – are likely underpowered. Thus, while the evidence suggests that preexisting attitudes and analytical reasoning have little impact on how disinformation shapes public opinion, future research is necessary to confirm this conclusion.

7.2 Discussion

These results underscore that attitudinal backlash to disinformation is much more common than direct persuasion. Disinformation headlines moved public opinion in the opposite of the intended direction in three out of six experiments. Disinformation linking the EU to liberal policies on migration and LGBTQ+ rights made individuals less Eurosceptic, less nationalist, and more tolerant of outgroups; likewise, false narratives about Ukrainian refugees strengthened tolerance. In the other three experiments, there was little evidence that disinformation headlines had any effect.

Additional analyses indicate that these effects do not differ as a function of individual ideology or analytical reasoning. These findings do not necessarily mean that disinformation is ineffective. The experiments focus on a narrow component of disinformation exposure: immediate reactions to potentially clear instances of disinformation. While measuring these immediate reactions is important for understanding disinformation’s effects, disinformation is likely to shape attitudes through less drastic exposures that occur over longer periods of time and across multiple channels. Perhaps more importantly, the disinformation narratives also focused on highly salient topics widely covered by both mainstream and disinformation media. As such, it is likely that public opinion on these issues is crystallized, preventing attitudinal updating. More generally, disinformation’s attitudinal effects may be most pronounced on topics or issues where opinions are not yet crystallized. In other words, effective counter-disinformation efforts may require proactively addressing emerging controversies and exploitable events before they become politicized.