When conspiracist beliefs matter: conditional associations between media-source reliance and pro-Russian disinformation beliefs during the war in Ukraine

Belief in pro-Russian disinformation about the war in Ukraine has important societal consequences. However, less is known about how media-source reliance and distrust-related dispositions jointly relate to such beliefs. Across two cross-sectional surveys conducted in Poland during the first year of Russia's full-scale invasion of Ukraine (Study 1: N = 397; Study 2: N = 1,066), we examined whether selecting social media, broadcast media, and online news portals as primary sources of war-related information was associated with pro-Russian disinformation beliefs, and whether these associations were moderated by political paranoia (Study 1) and conspiracy endorsement (Studies 1–2). Path analyses with multiple-testing correction showed that conspiracy-related belief indicators were consistently positively associated with disinformation beliefs across both studies, with political paranoia as a positive correlate in Study 1. Associations involving broad self-reported media-source indicators were weaker and less consistent, although direct comparisons across these predictor domains should be interpreted cautiously because they differed in measurement format, granularity, and conceptual proximity to the outcome. Reliance on social media was not positively associated with disinformation beliefs. Media-source main effects were generally weak; only online portal reliance in Study 2 remained robustly negatively associated with disinformation beliefs after correction for multiple testing. Moderation analyses suggested that source–disinformation associations were conditional on conspiracy endorsement, with different patterns emerging for classic CT endorsement vs. COVID-19 conspiracy beliefs. These findings caution against uniform channel-based accounts of disinformation susceptibility and suggest that distrust-related individual differences are relevant correlates of geopolitical disinformation endorsement within the present measurement design.

Russia's invasion of Ukraine has been widely framed as a watershed for European security, heightening nuclear tensions and disrupting food and energy supplies (Ash et al., 2023; Khorram-Manesh et al., 2023). The war has affected EU member states to an exceptional degree, fueling threat perceptions that can increase conspiratorial thinking and political paranoia—dynamics further amplified by coordinated Russian disinformation and propaganda (Beeson, 2005; Ji et al., 2019; Van Prooijen and Douglas, 2017, 2018; Soares et al., 2023). Endorsement of geopolitical disinformation is not merely an epistemic outcome. In the context of war, such beliefs may shape perceptions of victimized and aggressor groups, perceived legitimacy of political responses, and public support for humanitarian or security-related actions. This relationship is likely reciprocal: prior attitudes toward the war, Russia, Ukraine, refugees, or Western institutions may shape receptivity to pro-Russian narratives, while endorsement of such narratives may in turn reinforce or rationalize these attitudes. Understanding the psychological conditions under which people endorse pro-Russian disinformation, therefore, contributes to broader accounts of the societal consequences of conspiracist worldviews.

During the initial months of Russia's full-scale invasion, the war quickly became a focal point for research on public opinion, polarization, and susceptibility to pro-Russian disinformation and propaganda (Ash et al., 2023; De Vries, 2022; Erlich and Garner, 2023; Erlich et al., 2023; Flakus et al., 2025; Hameleers et al., 2024; Masch et al., 2023; Pokropek et al., 2026). This early phase is particularly informative because uncertainty and rapid narrative contestation created a volatile information environment, offering a strong test of how citizens form and update judgments under acute threat. Understanding reactions in such moments is therefore crucial, as they illuminate broader patterns of information processing, credibility evaluation, and alignment with ideological or identity-based narratives.

The Polish context is theoretically informative for studying pro-Russian disinformation narratives. Research on misinformation and disinformation has often focused on the United States and other broadly Western settings, whereas Central and Eastern Europe is strategically important because of Russia's geopolitical interests and influence operations surrounding the war in Ukraine. At the same time, the region is internally heterogeneous: comparative evidence from Poland, the Czech Republic, and Slovakia suggests that susceptibility to Russia-related disinformation varies substantially, with Poland appearing relatively resistant, Slovakia more receptive, and Czechia located between these poles (Wenzel et al., 2024). Poland is therefore a particularly informative and relatively conservative case: pro-Russian narratives circulate in a context marked by geographical proximity to the war, high public attention to Ukraine, strong public antipathy toward Russia, and high support for Ukrainian refugees. In 2022, nearly all Poles reported an unfavorable view of Russia, and large majorities perceived Russia as a major threat and expressed no confidence in Vladimir Putin (Poushter et al., 2022); support for accepting Ukrainian refugees reached 94% shortly after the invasion (CBOS, 2024). Thus, endorsement of pro-Russian disinformation narratives in Poland is unlikely to reflect a broadly pro-Russian public climate and may instead reveal more specific patterns of susceptibility linked to media-source reliance and conspiracy-related beliefs.

Building on this crisis-focused rationale, we ground our expectations in cognitive accounts of how people form and revise opinions under uncertainty. Cognitive research distinguishes three broad mechanisms of opinion formation: affective responses (Marcus, 2000), reasoning processes (Lupia et al., 2000), and cultural mechanisms, including outsourcing opinions to trusted others (Sloman and Fernbach, 2017; Sloman and Rabb, 2019). Reasoning processes are especially relevant in contested information environments because credibility judgments may be shaped by both analytical scrutiny and motivational alignment with prior attitudes or identities. Classic work on motivated skepticism shows that people tend to evaluate politically congenial arguments as stronger than uncongenial ones (Taber and Lodge, 2006), whereas research on cognitive reflection and fake news suggests that analytical thinking can improve discernment between true and false political information (Pennycook and Rand, 2019).

Whereas affective and reasoning accounts foreground individual differences (e.g., personality, intelligence, critical thinking), the cultural perspective emphasizes shared meanings, trust, and communication in social environments. In contemporary conflicts and disinformation, this mechanism is especially salient because people rely on social cues and mediated information to manage uncertainty (Traberg et al., 2024), often via media and interpersonal networks that signal what to believe and whom to trust (Shin et al., 2022). Accordingly, examining how citizens acquire and interpret information through broadcast and social media is a promising approach (Sloman and Rabb, 2019), particularly as digital platforms increasingly shape exposure, attention, and interpretation.

Social and broadcast media reliance and opinion formation

Although social media were not built for news, they have substantially accelerated information diffusion and become major news sources for many users (Fedeli and Matsa, 2018; Gottfried and Barthel, 2018; Gottfried and Shearer, 2016). Their role in public communication is therefore consequential but ambiguous. On the one hand, social media can amplify civic mobilization, as illustrated by movements such as #MeToo and Black Lives Matter (Alaggia and Wang, 2020; Mundt et al., 2018). On the other hand, their speed, scale, and algorithmic prioritization of engaging content can facilitate misinformation, reinforce biases, and support the circulation of conspiracy beliefs and political dis-/misinformation, especially in polarized or crisis-related contexts (Ahmed et al., 2021a,b, 2024; Enders et al., 2021; Guess et al., 2020; Hunter, 2023; Valenzuela et al., 2024). However, channel-based expectations may be too coarse, especially in high-salience crisis contexts, where social media can simultaneously circulate misinformation, journalistic updates, corrective information, peer evaluation, and norm-enforcing reactions.

Social media and broadcast media differ in their communicative logic. Social media are participatory, interactive, and engagement-oriented, allowing recipients to become content creators and participate in public discussion (Dolan et al., 2015). Broadcast media, by contrast, are more strongly shaped by editorial gatekeeping, verification norms, and crisis-information functions (Carruthers, 2000; Kasińska-Metryka and Pałka-Suchojad, 2023; Müller, 2019). Although broadcast media are not bias-free and can also reproduce threatening or politicized frames (Dinas and Van Spanje, 2011; Jacobs et al., 2017), their institutionalized source-checking procedures distinguish them from lower-accountability online environments.

In early-crisis conditions, source selection may matter because it is likely to be associated with the dominant credibility cues people encounter, i.e., interpersonal endorsement and rapid recirculation on social platforms (Mirbabaie et al., 2020) vs. editorial gatekeeping and elite cueing in broadcast news (Cikanek, 2023). Thus, if pro-Russian claims are more prevalent in low-accountability environments, reliance on social media should be associated with higher endorsement. In contrast, reliance on broadcast news (i.e., where verification norms are stronger) should be associated with lower endorsement.

From this perspective, media-source reliance should not be treated as operating independently of the recipient's prior distrust-related orientations. Individuals high in conspiracist beliefs or political paranoia may interpret the same information environments through different assumptions about agency, intentionality, and hidden motives.

Media-source reliance, political paranoia, and conspiracy-narrative endorsement

Media use – particularly via social media and partisan outlets – is a key predictor of belief in misinformation, reinforcing fear-based and distrustful narratives (Enders et al., 2021; Weeks, 2018) and promoting conspiratorial thinking and political paranoia in contexts of institutional distrust and perceived outgroup threats (Douglas et al., 2019; Imhoff and Bruder, 2014). Longitudinal studies show media consumption often precedes such beliefs (Romer and Jamieson, 2021). During the Ukraine war, Czech citizens who avoided public broadcasting and relied on online media were more likely to endorse pro-Kremlin disinformation (Grygarová et al., 2025).

Political paranoia—unfounded suspicion of personal or in-group persecution (Imhoff and Lamberty, 2018)—differs from conspiracy beliefs, which explain major events via secret, malevolent plots by powerful groups (Imhoff and Bruder, 2014; Van Prooijen and Douglas, 2017). Paranoia emphasizes personalized threat and interpersonal distrust, while conspiracism focuses on group-level threats and institutional distrust (Imhoff and Bruder, 2014; Imhoff and Lamberty, 2018). Conspiracy beliefs are frequently conceptualized as relatively stable political attitudes or worldviews, whereas political paranoia is more often discussed in terms of perceived threat and suspiciousness. However, both constructs overlap empirically and can co-occur, particularly under conditions of uncertainty (Imhoff and Bruder, 2014). The two are moderately correlated (Imhoff and Bruder, 2014) but differ in their relationship to media: conspiracy beliefs are strongly tied to media exposure and misinformation (Enders et al., 2021), while political paranoia reflects broader mistrust patterns (Imhoff and Lamberty, 2018).

Both political paranoia and conspiracist ideation can be conceptualized as relatively stable cognitive–affective orientations that bias threat appraisal, trust calibration, and epistemic vigilance, thereby shaping how individuals select, interpret, and weight mediated information (Imhoff and Lamberty, 2018; Douglas et al., 2019). In media contexts, these dispositions may operate via (a) motivated or confirmatory processing (privileging congenial cues and discounting counter-attitudinal information), (b) generalized distrust toward institutional sources, and (c) heightened sensitivity to intentionality and agency, which can jointly amplify or attenuate acceptance of particular narratives depending on perceived source alignment (Enders et al., 2021; Imhoff and Lamberty, 2018). Importantly, the same orientations may sometimes function as a “skepticism filter,” leading individuals to reject dominant misinformation when it is construed as emanating from hostile or untrustworthy actors (Jolley and Paterson, 2020; Pummerer et al., 2022). Accordingly, early-crisis periods may be a particularly informative context for testing moderation, because heightened uncertainty and rapid recalibration of credibility cues can increase the extent to which stable pre-dispositions condition how media exposure translates into belief updating and disinformation acceptance.

Thus, conspiracist beliefs and political paranoia may not only predict disinformation beliefs directly but also condition how reliance on different information sources is associated with such beliefs. Moreover, as political paranoia and conspiracy beliefs share a distrust- and threat-related core, their unique roles are best evaluated in models that consider both their separate and overlapping contributions.

The present research

Given the societal and political implications of the war in Ukraine (particularly for public opinion, information processing, and intergroup attitudes), we conducted two studies examining how reliance on different media as primary sources of information about the war relates to belief in disinformation, considering the moderating roles of political paranoia and endorsement of classic CT. The present research contributes to the psychology of conspiracy beliefs by examining whether conspiracist and paranoia-related orientations condition the association between self-reported reliance on information sources and endorsement of geopolitical disinformation. We test whether source–disinformation associations depend on distrust-related dispositions, including classic and content-specific conspiracy beliefs. As a proxy for classic conspiracy beliefs, we measured endorsement of two classic CT (9/11 conspiracy and New World Order). As an example of content-specific conspiracy beliefs, we used COVID-19 conspiracy beliefs. These conceptually related but empirically distinct constructs may differentially interact with individuals' selection of social vs. broadcast (and, additionally, online news portals) outlets among their main information sources, addressing a gap in the literature on cognitive-affective mechanisms in geopolitical crises.

Study 1. First, drawing on prior research on social media (Ahmed et al., 2021a,b, 2024; Chao et al., 2020; Enders et al., 2021; Evrony and Caplan, 2017; First et al., 2020; Gao et al., 2020; Houston et al., 2018; Hunter, 2023) and broadcast media (Dinas and Van Spanje, 2011; First et al., 2020; Jacobs et al., 2017), we hypothesized that selecting social media as one of one's primary sources of information about the war in Ukraine would be positively associated with belief in disinformation (H1), whereas selecting broadcast media as a primary source would be negatively associated with belief in disinformation (H2). In addition to the confirmatory hypotheses for social and broadcast media (H1–H2), online news portals were included as exploratory analyses, as existing theory does not yield a clear directional prediction for this category.

Second, based on research on sub-clinical paranoia (Cichocka et al., 2016; Kay, 2021) and generic CT beliefs (Cislak et al., 2021; Bertin et al., 2021, 2022; Kay, 2021), we hypothesized that higher scores in paranoia and classic CT endorsement would predict stronger belief in disinformation (H3).

Finally, considering links between media use and generic CT (Enders et al., 2021; Valenzuela et al., 2024) or paranoia (Lamba et al., 2023; Larsen et al., 2020), their conceptual overlap (Grzesiak-Feldman and Ejsmont, 2008; Imhoff and Lamberty, 2018), and their influence on media interpretation (Imhoff and Bruder, 2014), we hypothesized that both political paranoia (H4) and classic CT endorsement (H5) would moderate the associations between social vs. broadcast media reliance and belief in disinformation. We outlined two competing, context-dependent scenarios: (a) a susceptibility amplification pattern, in which higher levels of classic CT endorsement or political paranoia strengthen the association between media reliance and belief in disinformation, consistent with confirmatory or threat-based processing (Enders et al., 2021); and (b) a skepticism/attenuation pattern, in which higher levels of these dispositions weaken (or potentially reverse) the association, reflecting heightened distrust and selective rejection of messages perceived as misaligned with one's worldview or local political context (Eberl et al., 2021). Given these competing possibilities, the moderation hypotheses were directional regarding the presence of conditionality but open regarding the direction of the interaction pattern.

By skepticism/attenuation, we refer to a conditional pattern in which distrust-related dispositions weaken or reverse the association between reliance on a given information source and disinformation belief. Theoretically, this pattern can be grounded in motivated skepticism and epistemic vigilance: individuals may evaluate communicated claims through the lens of prior attitudes, perceived source trustworthiness, and communicative intent, scrutinizing claims that appear incongruent, manipulative, or source-misaligned more strongly (Sperber et al., 2010; Taber and Lodge, 2006). In the case of conspiracy-related dispositions, such vigilance may become generalized because conspiracy mentality involves suspicion toward powerful actors, hidden motives, and institutional sources of authority (Imhoff and Bruder, 2014). Thus, attenuation does not imply that conspiracist thinking is protective or accuracy-enhancing; rather, it captures a context-dependent redirection of distrust, whereby individuals may also reject disinformation narratives when these are perceived as source-misaligned, institutionally driven, or inconsistent with their broader worldview (Wood et al., 2012).

Importantly, analyses involving social media and broadcast media reliance were hypothesis-driven. In contrast, online news portals were included exploratorily because this category combines heterogeneous sources, and existing theory does not yield a clear directional prediction.

Study 2. The study re-tested the media-source hypotheses regarding reliance on social and broadcast media (H1–H2) and the direct association between conspiracy-related beliefs and belief in the pro-Russian disinformation (which were a part of H3). Given these measurement and sampling differences, we treated Study 2 as a conceptual extension of the conspiracy-belief analyses and a partial replication of selected media-source and conspiracy-belief associations, not as a direct replication of the full Study 1 model.

Specifically, political paranoia was not measured in Study 2. Therefore, the paranoia-related components of Study 1, including moderation by political paranoia, could not be retested. Study 2 also differed from Study 1 in recruitment procedure, sample composition, timing, and the operationalization of conspiracy-related beliefs. Accordingly, cross-study comparisons should be interpreted as evidence of partial convergence or divergence across related but non-identical designs.

The extension (compared to Study 1) concerned the conspiracy-belief component. Whereas Study 1 included only classic CT endorsement, Study 2 distinguished it from COVID-19 conspiracy beliefs. This allowed us to examine whether media-source associations with belief in disinformation were conditioned similarly or differently by two brief, closely related indicators of conspiracy-narrative endorsement. Classic CT endorsement was treated as a brief indicator of relatively decontextualized conspiracy-narrative endorsement, based on familiar conspiracy themes that are not directly tied to the war in Ukraine or to the COVID-19 pandemic. COVID-19 conspiracy beliefs, in contrast, were treated as a content-linked conspiracy-belief indicator. Although such beliefs are conspiratorial in form, they may also reflect more specific distrust toward public institutions, science, health authorities, political elites, and pandemic-related policies. Thus, the comparison between these two indicators allowed us to examine whether moderation patterns differed between a more decontextualized conspiracy-belief measure and a more politically and institutionally situated conspiracy-belief measure.

Participants and procedure

Both studies were conducted in accordance with APA ethical guidelines and institutional and national research ethics standards (e.g., approval of the Research Ethics Committee of the Institute of Philosophy and Sociology at the Polish Academy of Sciences, Warsaw, Poland; application no. 9/2022). Both studies were part of a larger data collection. All data and Supplementary Information that support the findings are openly available in the OSF repository: https://osf.io/xtczy?view_only=0bb0e241848f41fabb07645973d60387. The hypotheses and analytic model specifications reported in this manuscript were not pre-registered.

Study 1. The sample consisted of 397 participants (women: 65%) aged 18 to 79 years (M = 37.3, SD = 15.2). The majority reported secondary (n = 204, 51%) or tertiary (n = 186, 47%) education; the rest reported elementary/vocational education. Recruitment used two Facebook-based streams: (a) paid advertisements targeted to age and sex strata approximating population quotas (n = 203, 51%; number of ad impressions: approx. 17,100, link clicks: 10,677, started surveys: 1,056, completed surveys: 203—with no IP duplicates); (b) snowball sharing via participants' networks (n = 194, 49%). The final analytic sample included respondents with complete data on focal variables. Study 1 was conducted in May 2022.

Study 2. The sample consisted of 1,066 participants (women: 52%) aged 18 to 86 years (M = 44.84, SD = 16.25). The majority reported secondary (n = 365, 34%) or tertiary (n = 473, 44%) education; the rest reported elementary/vocational/post-secondary education. We recruited participants via an online research panel, considering quotas of age, education, and place of residence representative of the Polish population. Study 2 was conducted in August 2022.

Measures

The following measures were used to assess the focal variables across the two studies. The full wording of all items used in the present analyses was provided in Supplementary materials A1–A4.

Media reliance regarding the war in Ukraine

In both studies, media reliance was operationalized as the selection of up to three primary sources of information about the war in Ukraine (question: “From what sources do you get the most information about the war in Ukraine?”). Participants chose from a pre-defined list including broadcast and online TV/radio, printed press, online news portals, and social media (e.g., Twitter/X, Facebook). For analysis, we created binary indicators capturing whether each source was selected among the respondent's top three (0 = not selected, 1 = selected). We also created three category-level binary indicators, namely “Reliance on any social media as a main information source (top-three)” (Facebook or Twitter/X selected), “Reliance on any broadcast media as a main information source (top-three)” (broadcast and online TV or radio selected) and “Reliance on online news portals as a main information source (top-three)” (online news portals selected).

The “Broadcast/online television” and “Broadcast/online radio” options were aggregated into a single broadcast-media indicator to capture reliance on professionally produced, editorially mediated, institutionally identifiable news sources. The “Facebook” and “X/Twitter” options were aggregated into a single social-media indicator to capture reliance on platform-based and interaction-oriented information environments. Online news portals were retained as a separate indicator because the survey measured portal-based news use as a source category rather than specific outlets. All three categories remain internally heterogeneous and should not be interpreted as measures of exposure to identical content, editorial standards, platform affordances, or credibility cues.

All indicators reflect the selection of primary information sources (top three). They should be interpreted as indicators of relative reliance rather than frequency or intensity of media use, direct measures of exposure, time spent with a given medium, trust in a given medium, platform-use intensity, encountered content, exposure to specific credibility cues, or content-level consumption. Because participants could select up to three sources, the indicators were non-mutually exclusive and should be interpreted as overlapping self-reported source-prioritization measures rather than distinct media diets. The full list of source-response options, selection frequencies, and source-combination distributions for each study is reported in Supplementary material A4.

Political paranoia

We measured political paranoia using a ten-item “Political Paranoia Scale” (Korzeniowski, 2010), e.g., “In politics, nothing ever really happens openly, and all key political decisions are made in secret” (α = 0.92). Political paranoia was measured only in Study 1.

Conspiracy theories endorsement

In Study 1, we assessed endorsement of two widely circulated, classic conspiracy narratives (i.e., 9/11 inside job, New World Order) derived from Swami et al., 2011 as a brief indicator of classic CT endorsement (rather than a full-scale conspiracy mentality measure; α = 0.79). In Study 2, we reused those items (α = 0.75), adding two other statements referring to COVID-19 plots (e.g., “The government wants to restrict the rights and freedoms of citizens under the pre-text of fighting the COVID-19 pandemic”)—measuring COVID-19 conspiracy beliefs (α = 0.70). Importantly, these variables cannot capture the full breadth of conspiracy mentality and should be interpreted only as brief indicators of endorsement of selected conspiracy narratives.

Belief in disinformation

In both studies, we assessed endorsement of pro-Russian disinformation about the war in Ukraine. Study 1 used a seven-item questionnaire (α = 0.90), whereas Study 2 used a six-item version of the same measure (α = 0.89; both rated on a 5-point scale ranging from 1 = “definitely disagree” to 5 = “definitely agree” and with sums as composite scores)1. The six items common to both studies referred to narratives justifying Russia's attack on Ukraine, denying or minimizing Russian responsibility for civilian destruction, and portraying Ukrainians or Ukrainian refugees as undeserving of support or unfairly privileged (e.g., “The photos of ruined civilian buildings in Ukraine are photomontages distributed by the Ukrainians to blame the Russians.”).

The seventh item, included only in Study 1, stated that Ukrainian refugees posed a threat to the health of Poles. This item was not included in Study 2 because it referred to a temporally specific health-threat narrative that was more salient in the earlier phase of the war and closer to the COVID-19 pandemic context, but was no longer prominent in the monitored public discussion when Study 2 was fielded several weeks later2. All items were used in related research (e.g., Flakus et al., 2025; Pokropek et al., 2026).

Statistical analyses

Descriptives

First, for both studies, we computed descriptive statistics and zero-order Pearson's r correlations to summarize all variables.

Path models

In Study 1, to test whether reliance on different media sources, political paranoia, and classic CT endorsement were associated with belief in disinformation, we estimated a conditional process model implemented and estimated as structural equation models (SEMs) with manifest variables (path analysis).

Given the strong conceptual overlap between political paranoia and classic CT endorsement, and the high latent correlation between these constructs observed in the supplementary CFA analyses (see Supplementary materials A5, A6), the primary Study 1 analysis included both moderators simultaneously. This joint specification allowed us to evaluate whether political paranoia and classic CT endorsement showed independent associations with belief in disinformation and whether either construct independently moderated media-source associations. Thus, the primary model included the aggregated media indicators (reliance on any social media, any broadcast media, and online news portals) together with political paranoia and classic CT endorsement (with corresponding media-source reliance x moderator interactions).

In Study 2, the distinction between classic CT endorsement and COVID-19 conspiracy beliefs was used to examine whether moderation patterns differed between the two. Thus, using similar statistical techniques, we estimated a single path model including the three media-reliance indicators, classic CT endorsement, COVID-19 conspiracy beliefs, and the corresponding media × conspiracy-belief interaction terms. This specification allowed us to examine whether classic CT endorsement and COVID-19 conspiracy beliefs showed similar or divergent moderating patterns, while accounting for their shared variance.

To account for potential confounding, the both models were adjusted for demographic covariates (age, gender, education, and place of residence) and political conservatism (single item: “How would you rate your overall political beliefs?”, assessed from 1 = “liberal” to 7 = “conservative”). In all cases, standard errors and p -values were computed using the delta method. Percentile bootstrap confidence intervals were estimated using 5,000 bootstrap replications and are reported as auxiliary indicators of estimate precision.

Robustness

To assess the robustness of the observed associations and moderation effects, we re-estimated the primary models using alternative specifications: single-moderator models, unadjusted models, and models additionally controlling for engagement in discussions about the war in Ukraine and interest in war-related news. Discussion engagement was operationalized as: (a) public online communication (participation in discussions on online forums or social media, e.g., Facebook, Twitter/X), (b) online personal communication (discussions via online communication tools, e.g., WhatsApp), and (c) offline personal communication (discussions with relatives and/or peers offline). Interest in war-related news was measured with the item: “Are you interested in the war in Ukraine and follow the news related to it?” (rated from 1 = “definitely no” to 5 = “definitely yes”). These analyses were used to evaluate whether the pattern of findings changed under alternative model specifications and were not used to identify additional significant effects. These robustness models are reported in Supplementary materials B1–B6; full primary-model estimates and conditional effects are reported in Supplementary materials B7–B10.

Familywise error control

To control the familywise error rate across multiple regression paths, we report both unadjusted p -values and Holm-adjusted p -values within pre-defined families of tests, separately for each study and each primary joint model. In each study, the primary model included the relevant moderators simultaneously in a single joint specification. Holm correction was applied separately to (a) all non-interaction paths and (b) focal interaction terms. The non-interaction family included the media-source indicators, the relevant moderator variables, and covariates. We used these conceptually defined families because they map onto two distinct components of the models: direct associations with belief in disinformation, including adjustment paths, and moderation of media-source associations, respectively. A single Holm correction across all regression paths would be more stringent because it would combine conceptually distinct non-interaction and interaction tests into one larger correction family, thereby increasing the number of comparisons used to determine the adjusted p -values.

Primary inferences were based on Holm-adjusted p -values. Online portal reliance was included in the same correction families as the other media-source indicators, but was interpreted as exploratory.

Conditional effects were used descriptively to probe supported interactions and were not treated as independent confirmatory tests. Sensitivity analyses were used to evaluate whether the pattern of findings changed under alternative model specifications and were not used to identify additional significant effects.

Moderation

To examine moderation, we specified interaction effects between each media indicator and the focal moderator in both studies. Variables involved in the moderation effects were mean-centered before computing product terms. When an interaction term remained supported after Holm correction, conditional effects were estimated at the 16th, 50th, and 84th percentiles of the moderator to describe the direction of the interaction pattern. These conditional effects were treated as descriptive probes and were not used as independent confirmatory tests.

Software

All path models were calculated in JASP (v0.18.3). Holm's corrections were applied using R.

Descriptive statistics (studies 1 and 2)

Correlation analyses (Table 1) showed that belief in pro-Russian disinformation correlated weakly with broadcast media (Study 1: r = −0.17, p < 0.001; Study 2: r = −0.09, p = 0.005), and online portal reliance (in both studies; Study 1: r = −0.16, p = 0.001; Study 2: r = −0.27, p < 0.001), moderately with COVID-19 conspiracy beliefs (r = 0.42, p < 0.001), and strongly with both political paranoia (r = 0.63, p < 0.001) and classic CT endorsement (in both studies; Study 1: r = 0.66, p < 0.001; Study 2: r = 0.65, p < 0.001).

Classic CT endorsement was also weakly related to social media (only in Study 2: r = 0.07, p = 0.034; with non-significant association in Study 1), broadcast media (in both studies; Study 1: r = −0.17, p < 0.001; Study 2: r = −0.10, p = 0.002), and news portals reliance (in both studies; Study 1: r = −0.13, p = 0.011; Study 2: r = −0.26, p < 0.001), while belief in COVID-19 CT—to online portal reliance (r = −0.08, p = 0.013). Political paranoia also correlated exclusively with reliance on online portals (r = −0.12, p = 0.015).

Finally, classic CT endorsement was strongly and positively associated with political paranoia (r = 0.79, p < 0.001) and moderately with COVID-19 conspiracy beliefs (r = 0.58, p < 0.001). Overall, zero-order correlations suggested that political paranoia and conspiracy-related belief indicators showed larger bivariate associations with pro-Russian disinformation-narrative endorsement than did the broad media-source indicators. This comparison is descriptive and should be interpreted cautiously because the belief-based measures and binary media-source indicators differed in scale format, measurement precision, and conceptual proximity to the outcome.

Across the focal variables, skewness ranged from −0.24 to 1.72 for Study 1 and from −1.11 to 1.18 for Study 2, and kurtosis ranged from −2.01 to 2.74 and from −1.99 to.01 for Studies 1 and 2, respectively.

Associations between media reliance, political paranoia, classic conspiracy theories endorsement, and belief in disinformation—Path model (study 1)

Main effects

To examine whether media reliance, political paranoia, and classic conspiracy theories endorsement were associated with belief in disinformation, we estimated a path model (Table 2). It included main information channels, political paranoia, classic CT endorsement, and their interactions (adjusted R 2 = 0.52), and was adjusted for age, sex, education, residence, and political conservatism (for the coefficients of the controlled variables, see Supplementary material B7).

In this model, we controlled for basic sociodemographic variables, i.e., age (in years), sex (woman vs. man), education (high school or lower vs. post-secondary or university), place of residence (rural vs. urban), and political views (conservatism). For controlled variables' coefficients, see Supplementary material.

Results suggested that both political paranoia and classic CT endorsement were positively associated with belief in disinformation. Media-reliance main effects were weak and none remained statistically significant after Holm correction.

Conditional effects

The moderation pattern differed across the two distrust-related moderators. Interactions involving political paranoia were not supported. By contrast, the interaction between broadcast media reliance and classic CT endorsement remained supported, indicating that the association between broadcast media reliance and belief in disinformation varied as a function of classic CT endorsement.

Conditional effects (or the full list, see Supplementary material B8) showed that associations between media reliance and belief in disinformation were generally weak or non-significant at lower and median levels of classic CT endorsement. Negative conditional associations were observed primarily at higher levels of classic CT endorsement, particularly for reliance on broadcast media.

Robustness and sensitivity

Robustness and sensitivity analyses indicated that including additional variables (engagement in war-related discussions and interest in war-related news) did not materially change the key results. Specifically, the direction and magnitude of the media coefficients and interaction terms remained comparable, and the overall pattern of conditional effects persisted.

Notably, engagement in public online communication showed an independent positive association with disinformation beliefs. However, it did not attenuate the focal media associations or the moderation pattern. This suggests that the main findings are not attributable to general discussion engagement or online expressiveness.

As for the single-moderator specifications, both political paranoia and classic CT endorsement appeared to moderate the association between media reliance and belief in disinformation. However, as the primary joint model indicated (Table 2), the pattern changed when both moderators were entered simultaneously. Political-paranoia interactions no longer remained supported, whereas the broadcast media × classic CT endorsement interaction persisted. Thus, although the single-moderator models suggested broader distrust-related conditionality, the joint model indicated that this conditionality was not paranoia-specific and was more consistently attributable to classic CT endorsement. Main effects were more stable across specifications: both political paranoia and classic CT endorsement remained positive predictors of belief in disinformation, while media-reliance associations were weaker and did not remain robust after Holm correction.

Discussion

Study 1 showed that political paranoia and classic CT endorsement were strong positive correlates of belief in pro-Russian disinformation, whereas associations with media-source reliance were not supported. Although single-moderator models performed as sensitivity analyses suggested broader distrust-related conditionality, the main analyses specified as joint two-moderator path model indicated that the moderation by political paranoia did not persist when classic CT endorsement were modeled simultaneously. Thus, the evidence for paranoia-specific moderation was limited, whereas the conditional role of conspiracy beliefs appeared comparatively more robust. This pattern motivated Study 2, which focused on conspiracy beliefs and distinguished between classic and more content-specific (i.e., COVID-19) conspiratorial endorsement.

Associations between media reliance, classic vs. covid-19 conspiracy endorsement, and belief in disinformation—Path model (study 2)

Main effects

To examine whether classic CT endorsement and COVID-19 conspiracy beliefs were associated with belief in disinformation, we estimated a path model (Table 3) including main information channels (top-three), classic CT endorsement vs. COVID-19 conspiracy beliefs, and their interactions (adjusted R 2 = 0.46). The model was adjusted for age, sex, education, residence, and political conservatism.

Regarding information sources, reliance on online portals was robustly and negatively associated with belief in disinformation, indicating lower disinformation belief among respondents who reported online portals as their main information source. Reliance on social media and broadcast media was not significantly associated with the outcome.

Consistent with H3, conspiracy beliefs were positively associated with belief in disinformation. Specifically, classic CT endorsement emerged as a moderate predictor, and COVID-19 conspiracy beliefs as a weak positive predictor. Both associations remained statistically significant after Holm correction.

Among sociodemographic covariates (see Supplementary material B9), education was negatively associated with belief in disinformation, but this association did not survive Holm correction. Political conservatism was positively associated with disinformation beliefs and remained significant after Holm correction.

In general, classic CT endorsement were by far the strongest positive correlate of disinformation beliefs, while online portal reliance was the only media-source indicator showing a robust main association.

In this model, we controlled for basic sociodemographic variables, i.e., age (in years), sex (woman vs. man), education (high school or lower vs. post-secondary or university), place of residence (rural vs. urban), and political views (conservatism). For controlled variables' coefficients, see Supplementary material.

Conditional effects

Several interaction terms indicated that associations between reliance on information sources and belief in disinformation depended on conspiracy beliefs.

The interactions between broadcast media reliance, online portal reliance, and classic CT endorsement were negative and statistically significant. Such patterns suggest that as classic CT endorsement increase, the association between broadcast/online portal reliance and belief in disinformation becomes more negative (i.e., the simple slope for broadcast/online portal reliance decreases with higher classic CT endorsement). For social media reliance × classic CT endorsement, the interaction was not statistically significant.

As for COVID-19 conspiracy beliefs, only the positive interaction with broadcast media reliance was statistically significant after Holm correction, suggesting that as COVID-19 conspiracy beliefs increase, the association between broadcast reliance and belief in disinformation becomes more positive (i.e., the simple slope for broadcast reliance increases with higher levels of COVID-19 conspiracy beliefs). In contrast, the interactions of social media reliance × COVID-19 conspiracy beliefs and online portal reliance × COVID-19 conspiracy beliefs were not statistically significant.

Thus, classic CT endorsement and COVID-19 conspiracy beliefs did not operate in the same way. Classic CT endorsement strengthened the negative association between reliance on broadcast and online portals and disinformation beliefs, whereas COVID-19 conspiracy beliefs shifted the association in the opposite direction.

Across two cross-sectional studies, we tested whether selecting social media, broadcast media, or online news portals as primary sources of information about the war in Ukraine was associated with belief in pro-Russian disinformation, and whether these links were moderated by political paranoia (Study 1) and by classic conspiracy theories endorsement (Studies 1–2) and COVID-19 conspiracy beliefs (Study 2). Overall, the evidence for our hypotheses was mixed. H1 was not supported: social media reliance was not positively associated with disinformation beliefs in either study. H2 did not receive support: broadcast reliance was not associated with disinformation beliefs in Studies 1 and 2. H3 was consistently supported, as political paranoia and conspiracy-related belief indicators were positively associated with pro-Russian disinformation-narrative endorsement. These associations were descriptively larger than those observed for broad media-source indicators, but this comparison should be interpreted cautiously because the predictor domains differed in measurement precision and conceptual proximity to the outcome. H4 was not supported: in Study 1, the joint model did not support unique moderating effects of political paranoia. Finally, H5 was partially supported: conspiracy beliefs moderated selected media–disinformation associations (i.e., regarding broadcast media reliance), but the direction of moderation differed for classic vs. COVID-19 conspiracy beliefs.

On the exploratory level, online news portals showed a weak yet non-robust negative trend in Study 1, but emerged as a robust negative predictor in Study 2 and showed robust conditionality with classic CT endorsement. This pattern is consistent with, but does not test, the possibility that source reliance reflects not only content access but also exposure to distinct credibility cues, including corrective information.

The moderation results further clarified the role of distrust-related dispositions. In Study 1, political-paranoia conditionality was not supported when classic CT endorsement was also included in the analyses (although single-moderator sensitivity analysis suggested otherwise), suggesting certain variance shared with conspiracist orientation rather than a paranoia-specific mechanism. By contrast, moderation by conspiracy beliefs, especially for broadcast reliance, remained comparatively more robust.

Study 2 further showed that classic and COVID-19 conspiracy beliefs did not operate in the same way. Classic CT endorsement amplified the skepticism/attenuation pattern, making broadcast- and portal-reliance slopes more negative at higher levels of generic CT. COVID-19 conspiracy beliefs showed the opposite tendency for broadcast reliance, shifting this association toward a more positive slope. No moderation effects appeared for social media reliance. Thus, the direction of conditional media–disinformation links depended on whether conspiracy beliefs were measured as broad generic suspicion or as endorsement of more content-specific conspiracy narratives, consistent with the idea that classic CT may reflect a broad interpretive style, whereas content-specific CT (e.g., COVID-19 conspiracies) may function as content-committed priors.

Paranoia/conspiracy beliefs and associations with disinformation

Across two cross-sectional studies, the most consistent descriptive pattern was that conspiracy-related belief (including political paranoia in Study 1) indicators were positively associated with belief in pro-Russian disinformation. These associations were descriptively larger and more stable than those observed for broad indicators of reliance on primary information sources. However, this pattern should not be interpreted as definitive evidence that distrust-related dispositions are substantively more important than information environments. The two predictor domains differed in measurement precision, scale granularity, and conceptual proximity to the outcome: conspiracy-related beliefs and belief in disinformation were assessed as self-reported belief measures, whereas media reliance was assessed using coarse binary indicators of source selection.

At the same time, the observed pattern remains informative descriptively: within the present measurement design, belief-based indicators were more consistently associated with belief in pro-Russian disinformation than were broad indicators of respondents' selection of a given media category as one of their main sources of war-related information. Moreover, this result matches evidence that conspiracist ideation reliably relates to misperceptions and misinformation susceptibility in politicized contexts (Douglas et al., 2019; Douglas and Sutton, 2023). It also aligns with findings that paranoia and conspiracy thinking overlap (shared distrust/threat sensitivity), which can attenuate “paranoia-specific” patterns once conspiracist orientation is modeled simultaneously (Alsuhibani et al., 2022; Freeman and Bentall, 2017; Greenburgh and Raihani, 2022).

Mechanistically, these dispositions may operate as interpretive lenses. Conspiracy beliefs are tied to epistemic, existential, and social motives (often uncertainty and threat), fostering intentionality-based explanations and suspicion of covert agency, which can increase the plausibility of disinformation framed as deliberate plots by powerful actors (Douglas et al., 2019; Van Prooijen and Douglas, 2018). Conspiracy endorsement also covaries with generalized distrust and negative evaluations of high-power groups, making malign-intent attributions more accessible when judging political claims (Imhoff and Bruder, 2014).

Political paranoia can be understood as a threat belief: cognitive models emphasize sustained appraisals of interpersonal harm maintained by negative interpretation bias, worry/anxious arousal, and safety-seeking strategies, which bias ambiguous or contested information toward hostile-intent interpretations (Freeman and Bentall, 2017; Trotta et al., 2021). The strong correlation yet distinctness of paranoia and conspiracy thinking supports a shared distrust/threat core and helps explain why paranoia-related variance may shrink in joint models (Alsuhibani et al., 2022; Greenburgh and Raihani, 2022).

Finally, classic and COVID-19 conspiracy endorsement should be treated as distinct predictors of disinformation belief. Classic CT endorsement reflects a broad, content-independent tendency to suspect covert, malevolent agency behind major events (Imhoff and Bruder, 2014), whereas specific (e.g., COVID-19) beliefs are commitments to concrete narratives that shape credibility judgments in a more selective, “fit-based” manner (Douglas and Sutton, 2023; Imhoff and Bruder, 2014). Accordingly, classic CT endorsement may function as a general lens of suspicion. At the same time, COVID-19 beliefs act as topic-linked priors that amplify acceptance of aligned disinformation and rejection of incongruent accounts (Greenburgh and Raihani, 2022).

One size does not necessarily fit all: media reliance and disinformation beliefs

The present results challenge the assumption that reliance on social media necessarily heightens vulnerability to disinformation and other forms of misinformation (Enders et al., 2021; Guess et al., 2020; Hunter, 2023; Valenzuela et al., 2024). In the highly salient, fast-moving, and normatively charged information environment surrounding the war in Ukraine, source reliance may shape exposure to different credibility cues, such as interpersonal endorsement and rapid recirculation on social media vs. editorial gatekeeping and elite cueing in broadcast media (Mirbabaie et al., 2020). However, the present measures did not assess exposure frequency, content encountered, trust in sources, platform affordances, editorial gatekeeping, or corrective cues. Therefore, the observed media-source associations should be interpreted as conditional associations between self-reported source selection and endorsement of pro-Russian disinformation narratives, rather than as evidence of specific media-exposure mechanisms.

One possible interpretation, requiring testing in future research, is that high-stakes crises may strengthen corrective dynamics and norm enforcement in public communication. Morally charged issues may elicit frequent corrective cues—debunking, counterspeech, norm signaling—and increase the reputational costs of endorsing fringe narratives, potentially making alignment with corrective information more likely, especially among individuals prone to suspicious interpretations (Badrinathan and Chauchard, 2024; Cialdini, 2003; Geber, 2024; Nyhan and Reifler, 2010). For broadcast media, negative conditional associations may reflect selection into more institutionally mediated information environments or exposure to content structured by information authority, official statements, professional gatekeeping, and elite consensus (Zaller, 1992). However, because the present study did not measure exposure to corrective content, perceived norms, reputational concerns, specific broadcast outlets, trust in broadcast media, or perceived elite cues, these mechanisms should be treated as possible explanatory processes rather than as processes demonstrated by the current data. If broadcast coverage more strongly reflected pro-Ukraine framing and recurrent correction of pro-Russian claims, broadcast reliance could co-occur with lower endorsement of such disinformation.

For social media, the pattern should be treated as context-dependent and platform-heterogeneous rather than as evidence of a general negative association. Social media environments may host misinformation alongside corrective content, journalistic updates, peer evaluation, collective sensemaking, rumor evaluation, fact-checking, community-based notes, explainers, and peer correction (Bode et al., 2024; Drolsbach et al., 2024; Fischer-Preßler et al., 2019; Huang et al., 2018; Porter and Wood, 2021; Bode et al., 2024; Bode, 2018). Aggregate estimates may therefore mask substantial heterogeneity, especially because exposure to untrustworthy sources is often concentrated in relatively small user segments (Guess et al., 2020). In principle, selecting social media as a primary source may, in some contexts, indicate embedding in networks where corrective content circulates quickly and disinformation is contested rather than normalized. However, in the present study, social media reliance was measured only as selection of Facebook or X/Twitter among respondents' main information sources. The data therefore cannot distinguish whether respondents encountered misinformation, corrective content, peer correction, journalistic updates, or other forms of platform-mediated information. Because Study 2 did not replicate a main effect or moderation for social media reliance, any protective interpretation should remain context- and specification-dependent rather than a stable cross-study conclusion.

Generic vs. content-specific conspiracy beliefs as distinct moderators

Moderation results further support a context-dependent, disposition-contingent account. In Study 1, although political-paranoia conditionality was not supported, moderation by conspiracy beliefs (especially for broadcast reliance) was more robust. Thus, media-source reliance should not be treated as uniformly risky or protective: distrust-related pre-dispositions may shape how people interpret, accept, or resist information encountered via preferred sources. However, the data cannot distinguish verification from selective trust, selection effects, or shifts in reliance toward or away from particular sources. Accordingly, the present evidence supports political paranoia as a correlate of disinformation belief but not as a robust independent moderator of source–disinformation associations, especially given that it was assessed only in Study 1.

The skepticism/attenuation pattern can be understood through three related mechanisms. First, motivated skepticism implies that individuals scrutinize worldview-incongruent claims more strongly than congruent claims (Taber and Lodge, 2006). Second, epistemic vigilance and trust calibration suggest that recipients evaluate communicated information not only by its content but also by perceived source reliability and communicative intent (Sperber et al., 2010). Third, conspiracy-related dispositions are associated with generalized suspicion of powerful actors and hidden motives (Douglas et al., 2019; Imhoff and Bruder, 2014). Taken together, these processes may increase skepticism not only toward mainstream or institutional narratives but also toward disinformation narratives when these are perceived as emanating from hostile, manipulative, or worldview-incongruent actors. In this sense, attenuation should be interpreted as a context-dependent redirection of distrust rather than as accurate discernment or reduced susceptibility to misinformation.

Moderation patterns also varied by channel and type of conspiracy belief. In Study 2, reliance on the online portal showed robust conditionality with classic, but not COVID-19, conspiracy beliefs, suggesting that even coarse “portal use” may reflect different informational environments depending on dispositional priors. More importantly, classic and COVID-19 conspiracy beliefs showed divergent moderation patterns: classic CT endorsement strengthened the negative association between broadcast/online portal reliance and disinformation beliefs, whereas COVID-19 conspiracy beliefs shifted the association between broadcast reliance and disinformation beliefs in the opposite direction. This suggests that broad conspiracist suspicion and commitment to concrete conspiratorial narratives may not play identical roles in geopolitical disinformation contexts. Classic conspiracy beliefs may reflect generalized suspicion that is not necessarily aligned with pro-Russian narratives, whereas more content-specific conspiracy beliefs, for instance COVID-19 CT, may function more like content-linked priors organizing credibility judgments. Importantly, this should not be interpreted as evidence that classic conspiracy beliefs are protective; rather, it shows that source–disinformation associations depend partly on how conspiracist orientation is operationalized.

One possible mechanism underlying the positive COVID-19 conspiracy beliefs × broadcast reliance interaction is that content-specific conspiracy beliefs may function as anti-institutional priors rather than as generalized conspiracist suspicion. COVID-19 conspiracy beliefs often concern the motives and trustworthiness of governments, public-health authorities, scientific institutions, and other powerful actors, and may therefore organize credibility judgments around institutional distrust (Pummerer et al., 2022). Among respondents high in such beliefs, broadcast media coverage may have been interpreted less as corrective information and more as part of a distrusted mainstream communication environment. This interpretation is consistent with motivated skepticism and the hostile media effect, which suggest that attitude-incongruent or identity-threatening media content may elicit counter-arguing or perceived bias rather than belief revision (Taber and Lodge, 2006; Vallone et al., 1985). It also aligns with evidence suggesting that links between exposure to information sources and COVID-19 conspiracy or misinformation beliefs depend partly on trust in those sources (De Coninck et al., 2021). However, because we did not measure trust in broadcast media, perceived bias, counter-arguing, or exposure to specific broadcast content, this mechanism remains a theoretically plausible interpretation rather than a directly tested process.

Social and interpersonal implications of pro-Russian disinformation endorsement

Beyond identifying correlates of pro-Russian disinformation endorsement, the present findings speak to the broader social and interpersonal consequences of conspiracy beliefs in conflict contexts. Endorsing disinformation about the war in Ukraine is unlikely to be merely a private belief or an isolated misperception. In highly politicized and morally charged contexts, such beliefs may shape how individuals interpret the intentions of governments, media, scientific or expert institutions, and other social groups. To the extent that pro-Russian disinformation is embedded in conspiracist worldviews, it may therefore contribute to broader patterns of interpersonal and institutional distrust, suspicion toward official accounts of the conflict, and more antagonistic perceptions of groups seen as supporting Ukraine or opposing Russia. This interpretation is consistent with research showing that conspiracy beliefs are associated with reduced trust, lower willingness to engage in normative collective action, and more hostile or polarized social attitudes (e.g., Jolley and Douglas, 2014; Pummerer et al., 2022; Van Prooijen and Douglas, 2018).

Importantly, our results do not directly test these downstream consequences. However, they indicate one pathway through which conspiracy beliefs may become socially consequential: by increasing the credibility of conflict-relevant disinformation narratives. In the context of the war in Ukraine, such narratives may affect not only beliefs about the conflict itself, but also interpersonal conversations, perceptions of refugees or political outgroups, support for collective responses, and the degree to which citizens view mainstream institutions and media as legitimate sources of information. Future research should therefore examine whether the conjunction of conspiracist beliefs and disinformation endorsement predicts conflict-relevant social outcomes, including interpersonal trust, intergroup attitudes, political polarization, willingness to support humanitarian or military aid, and tolerance for hostile rhetoric toward groups implicated in the conflict.

This possibility is supported by prior work showing that conspiracy beliefs can have consequences beyond belief accuracy. Experimental studies indicate that exposure to conspiracy theories can reduce intentions to engage in political and pro-environmental action (Jolley and Douglas, 2014), while intergroup conspiracy narratives can increase pre-judice and discrimination toward targeted groups, with effects sometimes generalizing to other outgroups (Jolley et al., 2020). In crisis contexts, conspiracy beliefs have also been linked to lower institutional trust, weaker support for collective regulations, and reduced social engagement (Pummerer et al., 2022).

Taken together, these findings suggest that conspiracy beliefs may matter not only because they are associated with false or misleading beliefs about the war, but also because such beliefs may provide a basis for broader social distrust, intergroup suspicion, and polarization around conflict-related issues.

Limitations and directions for future research

Several limitations of the presented studies should be addressed. First, both studies relied on cross-sectional self-report data, which precludes causal inference and leaves open the possibility of reverse causality (e.g., individuals less inclined to endorse disinformation may preferentially rely on particular sources). Relatedly, measuring both predictors and outcomes via the same survey instrument increases vulnerability to shared method variance, and self-reported media use may differ from behavioral indicators of exposure. Similarly, although the models adjusted for basic sociodemographic characteristics and political conservatism, this adjustment should not be interpreted as resolving confounding. Several relevant factors were not measured or not included in the primary models, including institutional trust, trust in mainstream media, attitudes toward Russia and Ukraine, party preference, ideological extremity, anti-establishment orientation, news avoidance, digital literacy, and prior endorsement of COVID-related misinformation. These omitted variables may be associated with both media-source selection and endorsement of pro-Russian disinformation narratives.

Accordingly, the observed moderation effects should be interpreted as conditional associations rather than as evidence that media reliance causally alters disinformation beliefs at different levels of conspiracy belief. Thus, future research should employ longitudinal panel designs and stronger identification strategies to test directionality and mechanisms more rigorously. For instance, integrating survey data with passive or trace-based measures (e.g., device/app logs) would improve exposure validity and enable content-sensitive, platform-specific tests.

Second, comparisons between conspiracy-related belief indicators and media-source reliance should be interpreted cautiously. Conspiracy-related beliefs and pro-Russian disinformation-narrative endorsement were both self-reported belief measures and may share suspiciousness, hidden-agency attribution, substantive content, and agreement-response tendencies, whereas media-source reliance was assessed using broad binary indicators of self-reported source prioritization. Thus, larger standardized coefficients for conspiracy-related belief indicators should be treated as descriptive evidence of stronger associations within the present measurement design, not as evidence that distrust-related dispositions are more important than information environments.

The Study 2 distinction between classic CT endorsement and COVID-19 conspiracy beliefs should also be interpreted cautiously: COVID-19 conspiracy beliefs may reflect conspiracist ideation, but they may also index pandemic-related attitudes, institutional distrust, political grievances, or ideological orientations. Therefore, differences between these indicators should not be interpreted as evidence that topic-specific conspiracy belief, independent of pre-existing attitudes, uniquely drives endorsement of pro-Russian disinformation narratives.

Third, some limitations concern sampling and generalizability. Study 1 was recruited via Facebook advertisements and snowball sampling, and was overrepresented in women and respondents with tertiary education; it should therefore not be treated as representative of the Polish population. The Facebook-based recruitment strategy is especially relevant because social-media reliance was one of the focal predictors: although being recruited via Facebook is not equivalent to relying on social media as a primary source of war-related information, this recruitment mode may have restricted or selectively shaped variation in social-media engagement, platform familiarity, and online information-seeking habits. Study 2 improved demographic balance by using quotas for age, education, and place of residence, and was closer to the population structure in terms of sex and age. Nevertheless, it was drawn from an online panel rather than a probability-based population sample and still included a relatively high share of respondents with tertiary education. Cross-study differences in media-source reliance may therefore reflect recruitment mode, sample composition, timing, or measurement context rather than substantive temporal change alone.

As for the sample, statistical power for interaction effects should also be considered, especially in Study 1. Because interaction terms generally require larger samples to estimate precisely, non-significant interactions in the smaller Study 1 sample should not be interpreted as strong evidence of absence. Supported interactions should also be interpreted cautiously, with attention to uncertainty estimation and correction for multiple testing.

Fourth, media reliance was captured via broad “primary source” categories (social media, broadcast media, and online news portals), which serve only as coarse proxies for media diets. This operationalization does not capture the intensity, duration, or composition of exposure—features that likely matter for belief formation in disinformation. It also does not distinguish how these sources are used (e.g., incidental exposure via feeds vs. intentional news seeking), nor does it measure key contextual features such as algorithmic curation, interpersonal sharing, or the extent to which respondents follow partisan, alternative, or fact-checking accounts. This concern is especially salient for the social media category, which is highly heterogeneous across platforms and content types, including news, opinion, entertainment, and corrective information; similarly, “broadcast” and “online portals” vary substantially in editorial standards, political slant, and crisis coverage. Also, respondents could select up to three primary sources, meaning that the media indicators were non-mutually exclusive and reflected overlapping, constrained source-selection patterns rather than distinct media diets or isolated media environments. The coefficients, therefore, represent conditional associations within a constrained set of self-reported source selections or source polarization, not as clean comparisons between mutually exclusive media-use profiles, nor as measures of trust in specific sources, actual media consumption, or exposure to particular types of content within these sources.

Because ‘top-three reliance' is a coarse, binary indicator, interaction effects may partly reflect selection into media environments rather than differential processing of the same content. Future work should incorporate more fine-grained and, where feasible, behavioral measures of media use (e.g., time spent, platform-specific engagement, content-level exposure, and passive trace-based indicators), and should model differential exposure to misinformation, credible journalism, and corrective content to better identify the mechanisms linking media environments to disinformation belief.

A similar limitation concerns the measurement of conspiracy beliefs. In both studies, classic and COVID-19 conspiracy endorsement was assessed with only two items per indicator, which should be treated as brief indicators rather than comprehensive measures of conspiracy mentality.

Fifth, the Ukraine-war context may impose distinctive normative and informational conditions, which could limit generalizability to other misinformation domains. In highly salient crises, elite cues and social norms may foster stronger corrective dynamics and discourage public endorsement of fringe narratives, potentially yielding attenuation patterns that might not emerge in more ambiguous or polarized contexts. Future work should therefore test whether the direction and magnitude of media-source associations replicate across topics that differ in consensus, polarization, and perceived stakes (e.g., health misinformation or domestic electoral disinformation).

Finally, pre-registered replications and robustness-focused modeling would help delineate boundary conditions and distinguish stable effects from sample-contingent findings, particularly given the multiple-testing burden and the presence of interaction terms in models of this kind.

Taken together, the present findings indicate that endorsement of pro-Russian disinformation is strongly associated with distrust-related dispositions, particularly conspiracy beliefs, than with broad indicators of media-source reliance. At the same time, source-reliance–disinformation associations were not uniform: they varied across media categories, model specifications, and types of conspiracy beliefs.

These results caution against simple channel-based accounts of disinformation susceptibility and suggest that conspiracist orientations may hypothetically shape how individuals interpret war-related information environments. Understanding geopolitical disinformation, therefore, requires attention not only to where people obtain information but also to the distrust-related lenses through which they evaluate it.