1 Introduction

The public knowledge and support of major International Organizations (IOs) are vital for institutionalized cooperation on the global level (Dellmuth et al., 2022a, p. 3). By exploring the patterns in reporting on IOs in news media around the world, this paper sheds new light on the underlying mechanisms through which public knowledge and support of IOs are formed. It asks how the media visibility of IOs – the frequency with which IOs appear in news media – varies by country. It asks this question both for the affluent Western liberal democracies, often at the center of analysis of public attitudes toward IOs and of IO contestation and politicization (Lake et al., 2021; Walter, 2021), and—crucially— for the vast majority of the other states of the world, including emerging powers and smaller developing states. This global perspective provides key comparative insights unavailable when we only study Western states or a small group of selected non-Western powers.

My central question is how states differ in their media visibility of key global IOs and why. Media coverage of IOs matters enormously for how IOs work and how the public receives them. Recently, we have seen much interest in IR in how IOs are reported on in media in the context of mass protests against IOs, elite contestation, or scandals connected to their work (Ecker-Ehrhardt, 2018a, p. 530; Rauh & Zürn, 2020; Sommerer et al., 2022). My approach differs from these studies insofar as my analytical focus is not primarily on any individual IO or variation across IOs, and it is not on immediate instances of contestation in media. Instead, this study explores the most prominent macro patterns of everyday, routine media coverage of a large number of IOs around the globe and over several years. Through this everyday coverage, loyalty, trust, and “reservoirs of confidence” in IOs can be built over the long term (Tallberg & Zürn, 2019, p. 587; Dellmuth et al., 2022a, p. 28). Certainly, not all media coverage of IOs is positive, as the literature on IO politicization prominently discusses (Rauh & Zürn, 2020; Sommerer et al., 2022). Visibility alone does not bring legitimacy. Nevertheless, both analytically and normatively, visibility is a crucial condition for public scrutiny and IOs not being seen as instruments of governance by stealth (Steffek, 2007, p. 190; Bernauer et al., 2020, pp. 506–507). The reservoirs of public confidence in IOs cannot “fill up” if IOs do not systematically succeed in communicating their mission to the public and gain media visibility as an inherent part of the political order (Castells, 2008, p. 80; Hooghe et al., 2019, pp. 13–17). At the same time, visibility is a key—yet surprisingly understudied— component of IO politicization (Zürn et al., 2012, pp. 75–76). The opportunities for politicizing IOs by political parties or entrepreneurs differ in states where IOs figure continuously in public debates and in states where, under normal circumstances, they are close to invisible (Vries et al., 2021, p. 313).

Cross-national differences in the media visibility of IOs and the associated differences in knowledge of IOs (Dellmuth, 2016) are relevant beyond the questions of legitimacy and contestation. Quite plausibly, knowledge of IOs may influence public support for different IO designs and reform proposals regarding IO authority and state control (Anderson et al., 2019; Bernauer et al., 2020; Brutger & Clark, 2023). Equally significantly, the media visibility of IOs is closely associated with the ability of IOs to cue domestic audiences about cooperation problems (Brutger & Li, 2022; Chapman & Li, 2023; Thompson, 2006) and to successfully introduce new norms and ideas into public discourses around the world (Tallberg et al., 2020, p. 638). If IOs seek to inform the public and promote normative change, their media visibility is essential (Ecker-Ehrhardt, 2023a).

To contribute to the debate on these substantive issues, I develop a novel theoretical framework that synthesizes insights from Media Studies with those from prominent strands of theorizing about IOs in International Relations (IR). It highlights the tangibility of IOs’ activities as the key driver of their media visibility. At the core of the framework is the argument that, for most people, the work and functions of most IOs are abstract and distant. In short, intangible. These are features that make for a generally low news value of IO reporting (O’Neill & Harcup, 2009). In many countries, much of the work of IOs is just not interesting enough for everyday reporting.

At the same time, I argue that this does not apply uniformly across states. Building on the literature exploring the “boots-on-the-ground” implementation work of IOs (Eckhard, 2021), I argue that IOs may seem intangible for audiences in wealthy nations, especially in the West. But for low-income countries, the work of many of the major IOs is very tangible: concrete and proximate. Most major global IOs conduct sizable operational activity in human and socio-economic development in low-income countries. This makes their work there highly tangibly present. In these states, the media visibility of IOs can be expected to be much higher than in the West. I complement this theoretical argument with additional propositions reflecting IOs’ role in managing interdependence and globalization, the importance of IOs and multilateralism for weaker states, and states’ political regimes and cooperative values. IOs can be expected to be more visible in states that are more embedded in the global economy, less powerful, and more liberal-democratic.

I test the theoretical framework with a large novel dataset mapping the coverage of IOs in news media in 135 states—accounting for 96% of the world population—in 2018–2021. I analyze the content of more than 3.4 million articles and 2,000 carefully sampled media outlets in 63 languages. I then map the media visibility of 40 IOs and IO bodies jointly forming the UN System. This system consists not only of the UN central bodies, funds, and programs but also of all the UN specialized agencies and UN-related organizations. The former category includes, among others, the WHO, the IMF, and the World Bank. The latter consists of, among others, the WTO and the IAEA.Footnote 1 Jointly, this set of IOs and IO bodies forms most of the core of the current international order. Having said that, the UN System does not represent the universe of formal IOs, and many important multilateral and other IO formats are not covered in this study. Natural language processing tools – dictionary techniques combined with named-entity recognition – were applied to the more than 3.4 million articles to extract relevant references to IOs in media. This quantitative evidence is further complemented with human hand-coding of a representative sample of close to 1200 news articles from 2019 to 2020.

The results of the empirical analysis show that key global IOs appear in the media around 75% more frequently in countries with very low levels of human development compared to high-human development countries, controlling for various factors at the country, medium outlet, and article level. These substantively large differences support my core theoretical argument that the tangibility of IOs’ work drives their media visibility. “Less in the West:” Global IOs are far less visible in media in the West than in much of the rest of the world. In addition, the data provide limited support for the expectation that IOs should be more visible in less powerful states. It disproves the expectation that they should be more visible in states more embedded in the global economy or more democratic. Empirically, this is not the case.

The evidence further allows exploration of the mechanisms underlying the theoretical argument. First, hand-coded data show that the variation in IOs’ visibility is driven by variation in reporting on the local operational activity of IOs, which is more than twice as frequent in low-income countries compared to high-income countries. Second, an important observable implication is that the central hypothesis should not apply to IOs without sizable operational development-focused activities. The WTO and the IAEA are good examples. I show that the main results hold across most of the UN System and almost all its major bodies. However, in line with the theory, the differences in media visibility of IOs across states’ income levels are not detectable for the WTO and the IAEA. Third, the COVID-19 pandemic provides a unique opportunity to explore the validity of the argument by comparing media coverage of the WHO in pre-pandemic times and during the pandemic. The theoretical framework would have us expect the WHO to be more visible in low-income states in the pre-pandemic period, but this effect should disappear during the pandemic as the WHO emerged as a critical source of expertise on COVID-19 for states across the world. Empirical data confirm this. Variation in the media visibility of the WHO across states disappeared in the COVID-19 pandemic years of 2020–2021, in stark contrast to the pre-pandemic period 2018–2019.

Overall, the paper makes three contributions to the literature on IOs and their public image. First, it brings into the debate a novel perspective that puts systematic, long-term reporting on IOs in news media on the agenda, with a view toward the macro patterns in the information on IOs provided to the general public. Second, the article presents data on media coverage of IOs on a global scale. The results show that this is essential for understanding global patterns and interpreting the patterns observed in Western countries. Third, the study provides novel theoretical insights into how IOs are reported in media, highlighting the role played by the tangibility of their work. This matter is bound to raise attention in the years to come as public and elite contestation of “distant” and “out of touch” IOs continues to intensify.

2 Theorizing media visibility of international organizations across states: The tangible presence of IOs and alternative accounts

In the era of mediated politics (Bennett & Entman, 2000), “the media have become the most important source of information and vehicle of communication” in politics (Strömbäck, 2008, p. 230). For any political order, but especially for political institutions outside the realm of nation-states, it is crucial to successfully communicate to the general public their mission, relevance, the appropriateness of their procedures, and their accomplishments (Ecker-Ehrhardt, 2018a, b; Hooghe et al., 2019, p. 77; Schmidtke, 2019). IOs need loyalty developed over time—particularly in crises and when their decisions negatively affect some constituents (Easton, 1975, p. 444; Gronau & Schmidtke, 2016, p. 538). We know that individual-level attitudes to IOs are to some extent driven by personal characteristics as well as IO-level characteristics and only partly shaped by incoming information (Dellmuth et al., 2022b, p. 286; Dellmuth & Tallberg, 2020, p. 318). Yet, without widespread political knowledge of IOs, the public is unlikely to support their policies or existence in the first place. The alleged legitimacy gap IOs face can hardly be bridged unless the public knowledge gap is addressed at the same time.

Granted, media visibility is not necessarily what IOs always enjoy, as discussed in the large literature on IO politicization and contestation (De Wilde, 2011; Stephen & Zürn, 2019). Empirical research on IO politicization often uses IO media visibility, in connection to protest events, as a key component of quantitative measures of politicization (Tallberg et al., 2014, p. 759; Rauh & Zürn, 2020; Hooghe et al., 2019, p. 97; Sommerer et al., 2022). Media coverage of IOs is also likely to suffer from negativity bias (Brutger & Strezhnev, 2022), which is prevalent in news media reporting in general (Galtung & Ruge, 1965, p. 69; Soroka & McAdams, 2015). A key limitation of my focus on IO media visibility, rather than the positivity or negativity of reporting on them, is that this critical content will not be directly addressed. Despite that, the motivation for this study is that media visibility links directly with public knowledge of IOs and because visibility is a condition for any public debate on the legitimacy of IOs, whether legitimating or de-legitimating.

2.1 IO media visibility: Insights from Media Studies and International Relations

Two large strands of research and theorizing can inform an analysis of the media visibility of IOs across states. One is the rich literature on news value in Media Studies. The other is the literature on IOs in IR. I use the Media Studies literature as my starting point, as journalists and news editors decide most immediately on news media content. My theorizing draws on the literature that explores the key factors that typically make events, institutions, or persons newsworthy – that provide them with news value (Harcup & O’Neill, 2001; O’Neill & Harcup, 2009). These news value criteria have been discussed for decades in Media Studies and appear applicable to the newsworthiness of foreign affairs in general (Galtung & Ruge, 1965; Koopmans & Vliegenthart, 2011). Further, they are usefully completed with literature on foreign news coverage across media systems (Aalberg et al., 2013; Segev, 2015), on journalistic practices in selecting news (Schultz, 2007), and on the economic pressures on publishing news that cater most immediately to the tastes of consumers (Welbers et al., 2016). Drawing on these, I presume that in long-term reporting on IOs, media largely adhere to what editors and journalists expect their audiences to prefer, i.e., they are guided by the perceived news value of the reports.

While the Media Studies literature provides useful general frameworks, IR literature fills them with substance specific to IOs. In a recent contribution, de Wilde seeks to bring Media Studies and IR literature into communication in the context of the European Union. He uses the news value framework to identify theoretically those factors that can be expected to drive media visibility of the European Union (de Wilde, 2019; O’Neill & Harcup, 2009, p. 165; Schulz, 1982). I build on this approach as the factors discussed by de Wilde apply to reporting on IOs generally. He identifies four such factors, the last two directly relevant to my interest in cross-national variation in the visibility of many IOs.

The first news value factor is IOs’ status (1). An important reason why (some) IOs are likely to be reported on is that they constitute elite global governance bodies, enjoying the political authority to make binding decisions and the epistemic authority to interpret world affairs (Rauh & Bödeker, 2016; Zürn et al., 2012). This criterion appears particularly important for differences in media visibility across IOs. The descriptive data on the aggregate visibility of the individual IOs presented later in this text speak to this status criterion. At the same time, cross-IO differences are not the focus of this study, so the theoretical importance of this criterion for this specific analysis is limited.

The second news value factor is the valence of news (2), or its emotional charge. This may become relevant to IO reporting when, for example, we expect IOs to receive substantial negative media attention in crises (cf. Schlipphak et al., 2022; Debre & Dijkstra, 2021) or when they are connected with scandals (Ecker-Ehrhardt, 2018a). Sommerer et al. study situations of protests against the key global governance bodies by tracking references to IOs in connection to protest events, for example (2022). I fully acknowledge the importance of status and valence as news value criteria, especially for studies on IO contestation and exploration of cross-IO variation. However, my theoretical focus on media visibility at large, rather than on the valence of reporting as such, limits the use of valence for this specific analysis. And since my focus is on the everyday coverage of a larger aggregate of IOs over several years, this abstraction from valence is justified, and I do not build my theory around it.

In contrast, de Wilde’s last two news value criteria appear highly relevant for explaining variation in IO media visibility across states. The third news value criterion is relevance (3) and refers to whether an IO and its actions are consequential for the news audiences (O’Neill & Harcup, 2009, p. 279). The final factor on de Wilde’s list is audience identification (4), which for IOs refers to the feeling of proximity audiences have with the IO.

Building on these two general factors of relevance and identification, I posit that the critical challenge for IOs’ everyday media visibility worldwide is connected with the low tangibility of IOs and their work. By tangibility, I refer to the degree to which IOs and their work are abstract or concrete from the perspective of the news media audiences and the degree to which they are distant or proximate to their everyday lives geographically and socially. I argue that media visibility of IOs’ across states is likely to be crucially influenced by the extent to which they are perceived as abstract and distant – intangible – or concrete and proximate – tangible. I assume that for most people, most of the time, most IOs are quite intangible. IOs are abstract, depersonalized, and often complex in nature, and they are difficult to reach physically and socially, scarcely associated with audiences’ daily lives. To the extent this is the case, IOs are unlikely to score high on the perceived relevance and identification of news.

Again, synthesizing insights from Media Studies and IR provides concrete support for these intuitions. First, journalists know that media reporting, in general, tends to strongly prefer concreteness against abstractness, and that it tends to prefer personalization against facelessness of institutions – for example through focus on political leaders (Galtung & Ruge, 1965, p. 68; Harcup & O’Neill, 2001, p. 273). Yet, by their nature, IOs are primarily bureaucratic bodies. They draw their authority from depersonalized, impartial procedures, often abstract and legal principles, and from impersonal expertise (Barnett & Finnemore, 2004). Indeed, major IOs recognize the challenge and increasingly engage in personalizing the media content they seek to publish (Ecker-Ehrhardt, 2023b). Yet, the matters the IOs communicate about are often abstract (Duan et al., 2019), and their language is highly technocratic (Rauh, 2022). The possible gains from cooperation are often diffuse and long-term, not immediate and tangible (Baturo & Gray, 2023, p. 3). If that is the case, the ability of audiences to relate to the message from IOs or about IOs will be limited. This abstractness and complexity make it difficult for audiences to recognize the relevance of IOs in most everyday situations, and to identify with these abstract institutions.

Second, the feeling of proximity is essential for newsworthiness (Aalberg et al., 2013, p. 390; Hester, 1973, pp. 246–247). The fact that IOs are geographically distant to most people decreases their chances for media coverage (Koopmans & Vliegenthart, 2011, pp. 641–642). Moreover, the social distance to IOs is exceptionally high for most individuals. This applies on the individual level, with few people having a direct personal connection with IOs. Most importantly, global IOs cannot enjoy the advantage of generating a sense of community, belongingness, and we-feeling among people that nation-states often generate (Hooghe et al., 2019; Russett & Oneal, 2001, p. 75). That is, IOs are only weakly related to most audiences, not only at the individual level but also at the community and societal level. These are arguments often raised in connection with the EU, with its “capital” Brussels allegedly suffering from being too distant from the everyday lives of EU citizens (Berezin & Díez-Medrano, 2008, p. 9).

Taken together, the abstractness of IOs and their distance from news media audiences – in short, their low tangibility – directly maps onto lower fulfillment of the news value criteria of 3) relevance and 4) identification when decisions whether or not to report on IOs are made (de Wilde, 2019, p. 1196). Many IOs appear aware of this and are concerned with effective public communication (Ecker-Ehrhardt, 2018a, 2023a, b). Still, the low tangibility of IOs does not support their robust everyday media coverage.

One prominent implication is what Dellmuth calls IOs’ low attitude importance for the public, driving poor IO political knowledge (2016). In the most extensive available empirical survey mapping this topic, Dellmuth et al. found a sizeable gap in knowledge of IOs between elites and the general public (2022a). While 62% of elites in their sample correctly answered basic questions about three well-known international bodies (UN Security Council, IMF, and Transparency International), only 12% of the general public did (2022a, pp. 40, 48). The public appears not particularly knowledgeable of IOs.

Nevertheless, this evidence also highlights that individuals vary in their consumption of political information on foreign affairs and IOs and that there are likely sizable systematic cross-national differences. If my reasoning is correct, we should expect the intensity of media reporting on IOs to reflect the degree to which IOs are tangibly present in the states. If, for some states, IOs and their work can be expected to be systematically more tangible than for others, media reporting is likely to reflect that.

This general tangibility-focused reasoning has one prominent manifestation that should apply to most global IOs that form the UN System. The most obvious way IOs become tangible across many countries and over time is through their everyday “boots-on-the-ground” implementation work, especially in human and socio-economic development. Development is the policy field to which the vast majority of UN System IOs and IO bodies is closely related (Reinsberg & Westerwinter, 2021, pp. 70–73) and, indeed, “social progress and better standards of life” are enshrined in the UN Charter preamble as one of the four motivations for the establishment of the UN itself. Major UN System IOs’ work is connected with crisis financing (Stubbs et al., 2020), humanitarian assistance, and peace-related operations (Eckhard, 2021, p. 297) but especially with long-term development financing and project implementation (see e.g. Honig, 2020; Eckhard et al., 2023; Heinzel & Liese, 2021; Marchesi & Masi, 2021 for discussion of implementation by the World Bank). It is the socio-economic development agenda where money is spent globally, where much IO staff works, where IOs deliver concrete goods. The raising of development to a prime position on the global agenda is, after all, what the overarching architecture of the Millennium Development Goals and the Sustainable Development Goals has been seeking to accomplish.

My argument about the importance of tangibility can go beyond socio-economic development and on-the-ground implementation work. Tangibility may take on different forms in specific circumstances. One example is the WHO. Especially in the early phases of the COVID-19 pandemic, the expertise of the WHO became critically important – and thus highly tangible – for virtually all states of the world. Other examples may include highly contentious IMF programs or key UN Security Council resolutions concerning concrete conflicts. In specific instances, tangibility will be heavily driven by factors other than development. Indeed, in Sect. 5, I leverage some of these cases in exploring the mechanisms underpinning the theory.

However, my target here is to provide a macro-perspective of the bulk of the entire UN System and explore cross-national variation in the mass of everyday reporting on IOs. Given this goal, the systematic sizable differences in the presence or absence of tangible development-related work of IOs present the most promising theoretical take and the most prominent embodiment of the tangibility-focused argument. Consequently, in empirical terms, the media visibility of the UN System and most of its key bodies is likely to be systematically higher in low-income or low-human development states, where the IOs’ development-oriented activities are carried out. This is reflected in my core “tangibility of IOs’ work” hypothesis:

H1: Media visibility of IOs is higher in states where they are tangibly present. For most of the UN System IOs and IO bodies over a longer period, these are states with lower levels of human and socio-economic development.

In sum, the core empirical expectation derived from my theoretical framework is that most major formal global IOs will be significantly less visible in highly economically developed countries, typically associated with the West, than in most of the Global South.

2.2 Complementary hypotheses

Three theoretically plausible hypotheses for IO media visibility complement H1. First, the literature on IOs and their functions in world politics focuses on their role in managing material interdependence in an increasingly globalized world (Keohane, 2001). Following this reasoning, the ever-more intense economic ties among states and peoples should translate to a demand for information about the bodies that govern these areas of interdependence (Castells, 2008; Ecker-Ehrhardt, 2012; Rauh & Zürn, 2020). In a cross-sectional setting, information about IOs should be in higher demand in states more engrained in the globalized economy. News media in those states should meet this demand. The functionalist interdependence management hypothesis follows this reasoning:

H2: Media visibility of IOs is higher in states more embedded in the global relations of economic interdependence.

Second, the functionalist reasoning in H2 can be usefully complemented with insights from realist scholarship. Some works in this tradition show that not all states are equal in IOs (Stone, 2011). IOs may be perceived as instruments of powerful states’ influence when effectively controlled by them (Heinzel et al., 2023), and information about own state’s ability to control an IO may boost public support for that IO (Brutger & Clark, 2023). Yet it seems unlikely that the ability of government officials to exercise informal control in some IOs will make the public systematically more interested in reading about the IOs at large. The traditional realist perspective that dismisses the relevance of IOs for international politics seems more fruitful here. Audiences in less powerful states should be interested in information about IOs because, for these countries, IOs are relatively more important. For powerful states for which unilateral, bilateral or minilateral options are attractive, global IOs may be less relevant (O’Neill & Harcup, 2009, p. 165; Stone, 2011, p. 21). For weak states, IOs and multilateralism are often their only venues for exercising influence in world politics. Their public may be expected to reflect this, and the media should cater to this interest in their reporting on IOs. Hence, I formulate the importance of multilateralism hypothesis:

H3: Media visibility of IOs is higher in less powerful states.

Finally, non-material factors can be expected to play a role in public interest in information about IOs. Constructivist and liberal theorizing suggest that peoples’ attitudes toward IOs may be associated with their identities and values, and their identification with IOs and multilateralism. To the extent to which the current order is considered liberal (Lake et al., 2021, pp. 229–232), we should expect public interest in IOs to be most pronounced in liberal democracies (Tallberg et al., 2020, p. 631). News media should reflect this interest in institutions and report on IOs more. This is also in line with the basic concepts of democratic peace theory, which sees the combination of international organizations, democracy and interdependence as sources of cooperative behavior (Russett & Oneal, 2001). I thus formulate the liberal values hypothesis:

H4: Media visibility of IOs is higher in liberal democracies.

An alternative take on the domestic politics side of reporting on IOs would highlight that, in principle, autocracies might be critical of the “liberal” UN System, and negativity bias in news media reporting might drive media visibility of IOs in these countries up (cf. Brutger & Strezhnev, 2022). In practice, however, we observe support for the UN System and its key bodies from autocratic regimes, mainly due to the symbolic equality of all states, including non-democratic and weaker states, granted in the UN General Assembly and across the UN System (for related findings, see Tallberg et al., 2020, p. 635). The key condition for this mechanism to play out thus appears not to be met.

There are likely partial conceptual overlaps between H1 and H2–H4. For example, the management of economic interdependence logic may translate into high tangibility levels when jobs or entire business sectors are affected by IOs’ decisions. Similarly, the power-oriented considerations embodied in H3 may translate into the very high tangibility of an IO when a country relies for its security on the collective security mechanism of the UN Security Council. It is thus likely that tangibility is, in specific instances, linked to the forces embodied in H2–H4 but also that a comprehensive picture of IO media visibility across states will reflect forces underpinning more than one of the hypotheses.

3 Data and descriptive evidence

My analysis maps the media visibility of 40 IOs and IO bodies jointly forming the entire United Nations System. Table A1.1 in the appendix contains the complete list of bodies studied. In the data collection, separate entries were used for the central bodies of the UN: Security Council, General Assembly, Secretariat, and ECOSOC. In the analysis, these are all subsumed under the UN label. The same applies to the World Bank’s subunits. The measurement period covers 2018–2021, the most recent period for which data are fully available, and, at the same time, the period ending (just) before the Russian invasion of Ukraine.

3.1 Empirical data sources, sampling, pre-processing, and IO references detection

The data I use in this article come from an extensive new database developed within the project Global Flows of Political Information (GLOWIN). Footnote 2 The database builds a quasi-representative picture of online news media content worldwide. In describing the database, I use the plural “we,” as multiple scholars participated in its creation, deploying it for different analytical tasks (Parizek & Stauber, 2024; Parizek 2023; see also Rauh & Parizek, 2024). Once I return to the specific application of this database for this article, I also return to using “I.” The data collection infrastructure we developed was designed to meet two significant challenges. One is to secure coverage of news media that is truly global. To the best of our knowledge, a systematic map of media content around the world does not exist in social sciences. Second, we implemented several often technically demanding and highly laborious steps to make sure that our map of media content is as accurate and representative as possible. As I discuss more below, this includes bringing in data on audience geography, extensive manual checks of every news media outlet covered, and working systematically across languages.

Our key source for mapping news media content is the Global Database of Events, Language, and Tone (GDELT, 2019; Leetaru & Schrodt, 2013). GDELT is unique in that it covers media in virtually all countries. This is the only existing source that offers coverage even remotely close to this. GDELT processes the content of around 60,000 media outlets worldwide and publishes the results. This amounts to about 400,000 to 500,000 news article records per day. At the same time, to retain complete control over the data collection process, including sampling, we only take GDELT as a starting point. For all tasks except the first, we rely entirely on our procedures and careful sampling from this mass of data. The technical details of the data collection and pre-processing are described in the appendix, with the workflow summarized in Table A2.1. I briefly present the core features here in six steps.

First, for each day of data, we started with a list of URLs of news articles tracked by GDELT. Since our hardware and translation capacity is ultimately a limiting factor for the volume of data processing we can perform, we collected a random sample of 10% of these GDELT article records, accessing the URL of each news article. We succeeded in downloading the content of the article in approximately 70% of the cases. Second, we used the readability heuristic in Python to clean the files and extract the plain full texts of the articles (van Cranenburgh, 2022).

Third, we bring into the sampling process systematic data on the audience geography of each website (media outlet) provided by Amazon Alexa Web Services (Alexa Web Information Service, 2021). This commercial database estimates, for each domain, the size of readership of the domain in up to fifty countries. This is the key data we use to connect our raw news articles from GDELT with information on where their content is consumed. In addition, we used this extensive data to drastically filter the sources covered by GDELT and only kept in the analysis media outlets that ranked 500 or higher in at least one country. Only 30–35% of downloaded articles qualified based on this audience size criterion. Note that this also results in that broadly comparable numbers of media outlets are included for (almost) all states. The numbers range from low dozens of media outlets covered for smaller states to around 90 outlets for the most represented. This massively reduces imbalances in the sizes of media markets, where states such as the US have several orders of magnitude more media outlets than small developing countries. Later in the analysis robustness checks, I also narrow the data for each country down to only news articles in locally spoken languages.

Fourth, to ensure that our database only covers news media outlets, we performed a full manual check of every single web domain (media outlet) retained after the previous steps. A group of trained research assistants and the author of this article checked the content of the individual domains to distinguish news media sites from, for example, organizational or personal domains. This led to a further approximately 30% reduction in the number of domains analyzed.

Our data from GDELT covers news content in 63 languages. Our fifth step was translating a random sample of 20% of our downloaded content using Google Translate.Footnote 3 We maintained the proportion of originally English and non-English articles, to secure the representativeness of our final sample. Approximately 20% of our analyzed articles were originally in English. The remaining 80% underwent automated translation. All the steps above were designed to ensure the representativeness of the sample and of the empirical picture in the GLOWIN project database. After additional clean-up and the exclusion of very small states and states with few source outlets covered, or with only a minimal number of news articles, we were left with around 2,300 news articles per day of data, in total 3,409,360 articles for the period 2018–2021 ready for analysis, across 2036 domains.

Sixth, with the database in place, I could search for references in the articles to the 40 IOs and IO bodies. For this task, I used a dictionary method to detect various forms of IO names and their abbreviations in the text, enhanced with regular expressions. The full dictionary and the details of its construction are reported in the appendix section A8. This method was complemented with the flairNLP named-entity recognition (NER) tagger (Akbik et al., 2019). Both procedures yield very similar aggregate results, but the NER tagger underperforms with three-word IO names, often failing to recognize the entire three-word sequence as one entity. On the other hand, the critical contribution of the tagger is its ability to distinguish the simple pronoun ‘who’ and the acronym for the World Health Organization. After extensively validating the procedure, I opted for the regex-based dictionary search as the key NLP method but complemented it with the NER-based procedure specifically for the WHO case. At the article level, the measurement of the dependent variable is a binary variable indicating whether a particular IO or IO body is present in the article. These individual IO scores are aggregated across the UN System for most analyses. In the country-level aggregates, the dependent variable is defined as the share of articles in the given audience country that refer to one or more of the IOs. In selected analyses later in Table 2 and the appendix section A11, the measurement focuses on individual IOs.

3.2 Operationalization and analysis setup

The key theoretical proposition embodied in hypothesis H1 is centered on the tangibility of IOs’ work across states and their work in socio-economic development. To capture this tangibility of IOs’ work most comprehensively, I use the country’s overall level of human development, measured with the Human Development Index (HDI). The HDI is the most direct (inverse) measure of a state’s need for IO work (United Nations Development Programme, 2022). The reason for opting for HDI as the preferred measure is practical. More specific markers of IOs’ development work worldwide, such as UNDP development assistance funding volumes or World Bank loans could serve as alternatives. As I discuss below and show in the appendix section A5, their use produces very similar results to the primary analysis. However, these measures only pertain to individual IOs, and specific scores may be driven by respective countries’ relations with the particular IO. For example, conditionalities attached to World Bank loans may make them unattractive to specific states (Dreher et al., 2022). Since my primary target is to evaluate the overall aggregate visibility of the entire UN System and to avoid these possible biases, I opt for the use of HDI in the main analysis.

For the interdependence management hypothesis H2, the degree of integration of states into the global economy is measured with each country’s KOF globalization index score (Dreher, 2006; Gygli et al., 2019). The power of states (H3) is measured with the country’s economic size – its aggregate economic output (GDP (Log 10) (World Bank, 2022).Footnote 4 While more nuanced measures may be more suitable for security-related matters, economic size reflects the underlying power in the international system well (Beckley, 2018).Footnote 5 Finally, I approximate the liberal democracy of states (H4) using the V-DEM Liberal democracy scores (Coppedge et al., 2019).

A central problem with this list of theoretically derived predictors is that some are mutually highly correlated. The pairwise correlation between the country HDI (H1) and its KOF Globalization Index (H2) is r = 0.88***. With very few exceptions, the same countries that enjoy very high levels of human and economic development are also highly embedded in the global economy and vice versa. The correlation of HDI and power (GDP (Log 10)) is, in a cross-national analysis, r = 0.59***. Finally, the HDI and the V-DEM Liberal democracy scores (H4) are also correlated at r = 0.57***. This raises a multicollinearity issue in regression models. I thus concentrate the analysis in the body of the article on model specifications that include these predictors in different constellations, but not all at once. Pairwise correlations of the covariates are reported in section A4 in the appendix.Footnote 6

In the analysis, I include four control variables at the level of states, media outlets, and individual articles. The first control is the degree of foreign news coverage in respective media outlets (variable Foreign news media coverage). It reflects the average number of references to foreign states in news articles in a given media outlet or, in cross-national analysis, in media articles analyzed for a given country. The measure is constructed with the same news content dataset used to measure the visibility of IOs. Instead of references to IOs, though, it counts references to foreign states, building on Parizek & Stauber (2024). Including this variable controls for the possible heterogeneity of media outlets across countries and that some countries are more open to foreign news in general. Second, I add information on the audience size of the individual media outlets, both at the media outlet level and as a national average. Third, the length of the individual articles is controlled for as well. IOs may be expected to appear more in analytically deeper (and presumably longer) articles. Fourth, I include a dummy for articles originally in English. This controls for the possibly imperfect automated translation. At the aggregate cross-national level, country shares of articles in English are used. Combined, these four variables ensure that the IO visibility scores draw on comparable empirical data across states.

4 Evaluation of the IO visibility model: Tangibility as the key driver of IO media visibility

I now turn to my evaluation of the key theoretical proposition embodied in hypothesis H1 and the complementary hypotheses H2–H4. Relevant descriptive findings are presented first. The analysis follows, consisting of a series of OLS and logistic regression models, with alternative modeling techniques covered in the appendices.

4.1 Descriptive results

The dataset is limited to states with at least 1 million inhabitants, GDP higher than 1 billion USD, at least 5,000 news articles analyzed, and at least ten news outlets covered to ensure the numerical robustness of the results. This leaves us with 135 states, accounting jointly for 96% of the world population and 99% of the world GDP. From the 3,409,360 articles analyzed, a reference to at least one IO or IO body from the list of 40 was found in 150,172 articles (4.4%). This is a relatively substantial number, as revealed in the following comparison. Besides tracking references to IOs, we also adapted our database to map references to all the world’s states (Parizek & Stauber, 2024). The overall value of 4.4% for all the 40 IOs and IO bodies combined is approximately comparable to that of Italy. Based on our data, Italy ranks eight in global visibility among states. Of course, this compares the entire UN System against a single country, but still, the volume of media coverage appears relatively substantial.

However, there is much variation across countries: the share of articles referencing at least one of the 40 IOs and IO bodies varies between approximately 1.5% and 10%. At the high end are especially developing countries, for example, Togo, Yemen, or the Democratic Republic of Congo, joined by, among others, Palestine. At the low end is a large and heterogeneous group of states; as many as 65 out of the 135 states have IO visibility scores lower than 5%. Figure 1 provides a visual representation of the variation, with darker colors indicating a higher percent share of articles referring to one or more of the 40 IOs and IO bodies. To highlight the differences across states, results only for media in locally spoken languages are displayed here. In several instances, the visualization shows specific countries that do not fit well with their overall regional patterns. Zambia is one case with a low IO visibility compared to most neighboring countries. A closer inspection reveals that the record for the country is heavily impacted by international (Western) media in the English language, which our source of audience geography data also assigns to a readership in Zambia. As there is no systematic way to avoid this, caution is required when interpreting the data for individual countries, especially those with few observations, as opposed to larger country aggregates. Table A3.1 in the appendix shows the full numerical results.

Fig. 1
figure 1

UN System IOs’ visibility across the world

Full size image

The overall 2018–2021 global ranking of IO visibility is led by the UN by a large margin, with a reference in 78,952 articles (2.3%). This makes the UN approximately comparable in global media visibility to Korea or Mexico (Parizek & Stauber, 2024). The UN occurs in media more than five times as frequently as the IMF, third on the list across the period. This makes the IMF approximately comparable to countries such as Finland or Philippines. The WHO ranks second, with 0.89%, mainly due to a sudden spike in visibility during the COVID-19 pandemic, a matter explored in Sect. 5.2. In the pre-pandemic years, it scored third in visibility at 0.3% of articles, after the IMF and similarly to the World Bank, comparable to such countries as Bolivia or Kuwait. The WTO (sixth), to take one more example, appeared in around 0.15% of articles, making it comparable to Madagascar or Cambodia. The overall media visibility of the 40 IOs and IO bodies combined is fairly substantial, and the same applies to the UN itself. The media visibility of all the other IOs is much lower, however. Even some of the most powerful global IOs only receive media attention comparable to small states.

4.2 Hypotheses evaluation: Explanatory evidence

Figure 2 presents a simple bivariate view of the four hypotheses with the overall visibility of IOs on the vertical axes. The figure reveals strong patterns, though only partly in the expected direction. Most importantly, IOs are systematically more visible in media in states with lower levels of human development. At the same time, they are more visible in media in states less embedded in the global economy, in less powerful states, and in less liberal-democratic states. That is, the bivariate plots directly support my central hypothesis H1 and also hypothesis H3. In contrast, the evidence directly contradicts the interdependence management hypothesis H2 and the liberal values hypothesis H4. The numerical expressions of these relations are presented in Table A4.2 in the appendix.

Fig. 2
figure 2

IO visibility bivariate plots

Full size image

Next, I turn to the results of multivariate analysis presented in two parts, both summarized in Table 1. First, Models 1–5 show results from a set of OLS regressions at the cross-national level. As in Fig. 2 above, the dependent variable is the share of news articles in the given country referring to at least one of the 40 IOs and IO bodies. Second, Models 6–10 bring the analysis down to the level of individual articles. These are logistic regression models in which the dependent variable scores ‘one’ when any of the 40 IOs is mentioned in the article and zero otherwise. All the models include the controls described in Sect. 3.2. The reported standard errors are heteroskedasticity-robust. In the case of the logistic regression models, standard errors are clustered by audience country and media outlet. This is needed because on average, one news article has an audience in more than three audience countries.

Table 1 Linear regression (OLS) and logistic regression results
Full size table

In Model 1 in Table 1, I only include the HDI as my key predictor representing hypothesis H1, together with all the controls. Model 2 adds the country’s GDP (Log 10), and Model 3 adds Liberal democracy. Next, because Fig. 1 above also indicates support for hypothesis H3, in Model 4, I include both GDP and Liberal democracy while excluding HDI to avoid high multicollinearity. Finally, Model 5 includes the KOF Globalization index, again excluding HDI for the same reason. All Models 1–3 confirm the significant negative association between IO visibility and states’ level of socio-economic development (H1), as shown in Fig. 1. The substantive effect of the variable is sizeable. In Model 2, an increase from the empirical minimum of HDI, around 0.4, to the empirical maximum of 0.96 is associated with a decrease from around 7% to 4% of articles making a reference to an IO, holding all other variables at their means. The visibility of IOs in the lowest-HDI states is approximately 75% higher than in the highest-HDI states, ceteris paribus. Seen in reverse, in highest-HDI states the reporting on IOs is less than 60% as frequent as in lowest-HDI states.

Results for the alternative theoretical accounts explored in hypotheses H2–H4 are also interesting. Contrary to the descriptive evidence in Fig. 2, a country’s economic size (GDP Log 10) does not show the expected negative effect on IO media visibility in this multivariate setting. For Liberal democracy, in Model 4, where HDI is not controlled for, a statistically significant but weaker effect is identified, dropping from around 5.9% to about 4.4% as we move from the least liberal-democratic to the most liberal-democratic states. In Model 3, though, no significant association is detected. Finally, in line with observations in Fig. 2, H2 is manifestly refuted with a significant but negative association between the KOF index and IO visibility.

Models 6–10 in Table 1 show the results of five logistic regression models analogous to the linear regression Models 1–5 but at the level of individual media articles. The results are substantively aligned with those reported earlier. Across all Models 6–8, HDI shows a consistent negative association with IO visibility expected in hypothesis H1. In Model 7, for example, the predicted probability of a reference to an IO decreases from around 5% to around 3% as we move from lowest-HDI to highest-HDI states, holding other variables at their means. In Model 6, which does not include the (collinear) GDP (Log 10), the effect is stronger, with a decrease from around 6% to around 3%, effectively halving the predicted probability of an IO reference. Models 7, 9, and 10 show support for H3 (GDP (Log 10), in contrast to the linear regression Models 2, 4, and 5. Liberal democracy shows a negative association with IO visibility in Model 8, as in Model 4 before, but loses statistical significance in Model 7 where HDI is included. Overall, the models show mixed support for H3 and a clear refutation of H2 and H4. In addition, the analysis shows the importance of inclusion of Foreign news coverage as a control. Media with an average of 0.1 references to a foreign state in an article show a predicted probability of a reference to an IO at around 0.5%. In contrast, media with an average of 3 foreign state references per article predict a reference to an IO in more than 10% of articles.

In the appendix, I provide a battery of analyses to test the robustness of the main results, consisting of three deep changes to the empirical estimation strategy. First, I present a series of robustness tests showing that the results hold with the use of relevant alternative indicators for the core hypothesis H1 and the complementary hypotheses H2–H4. They are reported in section A5. Most important are the tests in which I replace HDI as the variable capturing H1 with countries’ GDP per capita (Log 10) as a measure of economic development and with two direct – while admittedly also highly selective – measures of IOs’ tangible “boots-on-the-ground” work in developing countries. For that purpose, I use the financial volume of development projects implemented worldwide by the UNDP and the financial volume of World Bank lending projects. All measures deliver substantively identical results to those reported in Table 1 using HDI.

Second, I alter the news articles sampling strategy by only including in the analysis articles written in the locally spoken languages of the audience countries. So, for example, a widely read English language source will not be considered in the analysis of a Spanish-speaking country. This is the same modification as applied in Fig. 1 above. This cuts the number of country-articles by around 30% and thus reduces somewhat the number of countries in the analysis, but it provides a stronger focus of national (as opposed to global) media. The analysis is presented in the appendix in section A6, again with results consistent with those reported above, but even stronger substantive effect sizes for H1.

Third, I replicate the analysis using a completely different source of data to map news. Instead of GDELT, I use Google News accessed via its RSS feed, the second largest available source. I use a representative sample of headlines from the 82 national-language variants of Google News in 2019–2021. The 82 national-language variants represent 69 states, dominantly those with larger markets and only including few smaller developing countries. This is a significant disadvantage for my purposes here. However, the advantage of this source is that the content is curated, and we can use Google News’ editorial decisions to determine a particular headline in a specific country. The analysis results are reported in appendix section A7 with results fully consistent with those reported in Table 1, and again with even stronger substantive effect sizes for H1.

5 Exploring the tangibility mechanism

In this section, I offer two avenues of further exploration of the mechanisms underpinning the theoretical framework. First, I study qualitatively the content of a sample of news and the reporting on IOs’ local tangible activities. Second, I explore empirically two key observable implications of the theoretical argument.

5.1 Tangibility and the content of reporting on IOs: Human coding

The central focus of this article is on the intensity of IO media coverage. But to obtain further insights into the mechanisms producing the main observed effect, I complement this quantitative view with qualitative insights on the content of reporting on IOs. With a group of research assistants, I estimated the prominence of four basic types of references to IOs in a sample of articles from the dataset. The following four classification categories—or types of news reporting on IOs—are distinguished in the scheme:

  1. 1.

    Reporting focused on central IO politics, typically headquarters-based, for example, central IO decision-making;

  2. 2.

    Reporting focused on the local implementation work, typically associated with a specific country and specific IO operational activity;

  3. 3.

    Reporting focused on the provision of data, information, and expertise by the IO, e.g., in the form of reports or expert announcements;

  4. 4.

    Reporting focused on explicit promotion of the norms underpinning the IOs’ mandates, e.g., the promotion of human rights.

The codes are not exclusive, so a concrete article referring to an IO can, for example, refer to local IO politics (2) and the provision of data and expertise (3). The four categories are not designed to correspond to the four hypotheses of this paper; they are theoretically derived from literature on the sources of IOs’ authority (Barnett & Finnemore, 2004; Zürn et al., 2012). At the same time, category (2), “local IO politics,” reflects closely the reasoning underpinning hypothesis H1.

In total 1,192 articles randomly sampled from the large dataset from years 2019 and 2020 were coded, always by two independent trained coders. In the 1,192 articles, 1,498 references to a concrete IO were made (more than one IO can appear in an article). More details on the coding procedure are provided in section A9 in the appendix to this article.

Evidence from this coding reveals two insights that speak directly to the core interest of this article. First, local IO politics is the most frequent reference type (42%) in pre-COVID 2019. In 2020, the value dropped to 20% and was replaced by information and expertise provision as the most frequent type of news. Yet, even if we combine the two years, reporting on local IO politics is as frequent as reporting on central IO politics, both appearing in total in 31% of cases. This prominence of localized news on IOs, with explicit references to specific states and IO activities in them, corresponds closely with the reasoning behind hypothesis H1. Explicit promotion of norms is the least frequent of the four types of references (17%).

Second, the data reveal a systematic difference in how IOs are reported across states. In each of the four charts in Fig. 3, coded article characteristics are connected with the states in which the articles have an audience. I plot the frequency of the given reference type against states’ HDI. Figures (A), (C), and (D) show no systematic relationship between how IOs are referred to in media and the states’ level of human development. For local IO politics in chart (B), however, there is a sizeable effect in which IOs in high-HDI states are only referred to in connection with local IO implementation work in around 25% of cases. In contrast, in low-HDI states, this is the case in around 50% of news on IOs or more. This evidence strongly supports the intuitions behind hypothesis H1.

Fig. 3
figure 3

Types of references to IOs across states

Full size image

5.2 Tangibility and variation across IOs and over time

Hypothesis H1 is focused explicitly on IOs’ work in development as the key area in which most UN System IOs and IO bodies gain tangibility over time and across many states. However, the theoretical argument is broader, and, as acknowledged earlier, the tangibility of IOs’ work and presence need not be only material. The applicability of the reasoning used here may be broader than that captured in H1, while, at the same time, hypothesis H1 should not apply to global IOs without development-focused operational work. Both these implications of the theoretical framework can be leveraged.

First, one possibility is provided by that the data cover both the pre-pandemic period 2018–2019 as well as the pandemic years 2020–2021. This allows us to explore a change in the position of the WHO, the central platform for global health governance and pandemic response. In the pre-pandemic years 2018–2019, the WHO is the third most visible of all IOs, after only the UN and the IMF. As with other operational IOs, much of the WHO’s everyday agenda is centered on developing countries, and hence, the pattern expected in H1 should apply. During the COVID-19 pandemic, however, the WHO became unprecedentedly relevant around the globe, regardless of development level. This is reflected in that in 2020, the WHO was virtually equally visible as the UN, appearing in 1.9% of articles, six times its usual visibility. In March–April 2020 specifically, it appeared in as much as 3.3% of articles and became twice as visible globally as the UN itself. The COVID-19-related work of the WHO became equally tangible – concrete and understandable, proximate and with direct ties to people’s everyday lives – across states’ income levels.

Models 1 and 2 in Table 2 support this. In 2018–2019 (Model 1), the WHO showed a strong negative association between its visibility and country HDI. Substantively, the WHO’s visibility decreases from around 0.54% of articles to about 0.26% of articles as we move from the lowest-HDI to the highest-HDI states. This halving of media visibility along the range of HDI is perfectly in line with H1. In 2020–2021 (Model 2), the negative association between the WHO visibility and HDI disappeared, and the WHO became equally visible in media around the world irrespective of the development status, at around 1.7% of articles on average. A profound change in the visibility patterns of the WHO appeared, and it is closely connected with how tangible the WHO’s work became across the world.

Table 2 Differences in media visibility patterns across selected IOs (OLS)
Full size table

Second, we can explore the applicability of H1 by comparing patterns of media visibility for several highly visible selected IOs. Not all IOs and bodies forming the UN System have sizable operational development-related portfolios. I take the United Nations and the World Bank as reference points as the two most visible IOs with a strong focus on operational work in development. The reasoning underpinning H1 should apply to these two IOs. This compares with the two most visible IOs without sizable on-the-ground operational activity focused on the developing world: the WTO and the IAEA. I expect that the WTO and the IAEA should be equally or more tangible for developed countries as they are for developing countries. The WTO is responsible primarily for global trade rule-making and is thus relevant for states at different levels of development. Similarly, the IAEA’s agenda is highly pertinent for economically developed and large middle-income states because of its focus on advanced nuclear technology. The reasoning underpinning H1 should not apply in these cases.

Models 3–6 in Table 2 confirm these expectations. All core results hold robustly for the UN and the WB (Models 3 and 4). Moving from the lowest country HDI to the highest is associated with a decrease from approximately 3.6% to 2.4% of articles for the UN and from 0.63 to 0.17% of articles for the World Bank, other things equal. In contrast, in the WTO and IAEA (Models 5 and 6), no systematic association between media visibility and HDI is detected. The WTO is predicted to appear in approximately 0.18% of articles and the IAEA in around 0.08% of articles, irrespective of the state’s level of development. As shown in the appendix in section A10, these results also hold in the logistic regression setup.

While these insights only pertain to a selection of two likely (UN and WB) and two unlikely cases (WTO and IAEA) for the theorized association with HDI to apply, the results offer further credibility for the reasoning behind H1 and invite more systematic exploration of cross-IO variation in future research. There are other bodies in the UN System for which the reasoning underpinning H1 should not be expected to apply, given their limited involvement in operational development-related work. Beyond IAEA and the WTO, especially small technical UN agencies, such as the ITU or the UPU and some others, should be cases at hand. A limitation in analyzing this cross-IO variation is the sparse reporting on most IOs in the UN System, especially the small ones. This makes a cross-national pattern identification with 135 states unreliable for them. However, to at least probe the logic underpinning H1 further, I selected those IOs for which my dataset reports at least 1000 observations. I replicated Models 1–3 from Table 1 for each separately. The analysis shows that the core pattern representing H1 is robustly present across the models in 13 out of 17 IOs. Indeed, these 13 IOs jointly account for 93% of all the news media references in the dataset, demonstrating the broad applicability of H1 in the mass of the UN System. All four IOs for which H1 does not apply – the WTO, the IAEA, the ICJ, and the UNRWA – represent clear cases where H1 should not apply, based on the theoretical framework. In the case of UNRWA, this is due to its unique geographical focus on the Middle East. Only two more IOs from the list can be considered borderline cases for the theoretical applicability of H1 – the ILO and the ICC. The tangibility of the ICC for low-HDI countries is likely driven primarily by the ICC’s broadly discussed ‘Africa problem,’ including several high-profile ICC cases (García Iommi, 2020). The explanation is less evident for the ILO, although development figures prominently on its core agenda, primarily via SDG Goal 8: Decent Work and Economic Growth. Appendix section A11 shows the full results of this analysis. These cross-IO results need to be treated with caution due to the sometimes very low numbers of observations. Yet, the overall pattern supports the broad applicability of H1 in the UN System. At the same time, from the at least somewhat more visible IOs for which the analysis can be reliable, the few cases where H1 is not applicable appear clearly in line with the theoretical framework.

6 Conclusion

The analysis offered in this article provides important insights into the macro patterns of IO media visibility around the world. The media visibility of global IOs in Western states is less than 60% of their visibility in low-income, low-human development states. These patterns are driven—I argue—by the tangibility of IOs’ work in developing countries. In contrast, in the West, where IOs are more abstract and distant from people’s everyday experiences, their media visibility is much lower. Perhaps surprisingly, IO visibility is also lower in states more integrated into the global economy and in liberal democracies. All these findings speak directly to the rapidly growing attention to the “non-Western” world in analyzing global politics (for example, Stephen, 2021; Zarakol, 2022). Our efforts to understand global governance and major IOs will be systematically biased if we lack good understanding of their image among emerging powers and developing countries more broadly (e.g. Verhaegen et al., 2021).

Differences in the visibility of IOs across states are likely to translate directly into differences in the knowledge of IOs and cooperation at large (cf. Aalberg et al., 2013). This may have consequences for two-level games of IO-related intergovernmental bargaining (Vries et al., 2021, p. 307). It may also impact the ability of IOs to shape the normative environment of world politics, put new topics on the international agenda, or generally cue domestic audiences about cooperation problems (Brutger & Li, 2022). In the long term, this may have implications for IO legitimacy. If IOs want legitimacy, they need to be seen in media as delivering concrete results and being proximate rather than distant to the audiences and their everyday experiences. This may be simpler in states in which they conduct large-scale operational ‘on-the-ground’ activity. In high-income countries, it is difficult to identify the tangible, non-abstract results of IOs’ work that could be presented to the general public in media-attractive terms (de Wilde, 2019, p. 1203). If in high-income countries the positive public image of IOs must be built primarily on their abstract contribution to the provision of global public goods, then explaining the benefits of IOs to the general public will be necessarily tricky. This is bad news for the long-term legitimacy of IOs (Dahl, 1999, p. 30).

These findings invite much further exploration of this research agenda. First, the prime contribution of this article is in exploration of worldwide macro patterns. The analysis of the content of reporting on IOs is limited to a highly structured human classification of a sample of just below 1,200 articles. A significant step forward would be a more systematic exploration of the qualitative content of news reporting on IOs. An important limitation of my approach is that the analysis does not cover the valence of media references to IOs. Our understanding of the global order would be much more robust if we had evidence on the prevalence of criticism and praise of major global IOs (Kentikelenis & Voeten, 2021). Second, another avenue for further research is in applying the framework developed here for studying regional IOs and global IOs beyond the UN System. It is not apparent how far the results obtained here for the UN System, or more precisely the bulk of the major UN System IOs and IO bodies, generalize. It would be interesting to see how this framework applies to major informal IOs, such as the Group of 20, or prominent transregional IOs, such as NATO or the AIIB. Relatedly, regarding regional IOs, media reporting on regional IOs and global IOs could prove to be complementary, or reporting on one of the levels could function as a substitute for reporting on the second. Given the importance of regional integration for everyday politics, much insight might be gained from such analysis (Schmidtke & Lenz, 2023). Third, the theoretical model presented here can be further developed to explore variation across IOs. One possibility would be to study systematically the relationship between IO media visibility and their authority and design. Another more specific avenue could link visibility with IOs’ efforts at active communication (Ecker-Ehrhardt, 2018a). The theoretical framework presented here can be modified to illuminate diverse research tasks concerning the public knowledge of IOs, their legitimacy, and related topics.