Avenues to News and Diverse News Exposure Online: Comparing Direct Navigation,Social Media,News Aggregators,Search Queries,and Article Hyperlinks

Introduction

The online environment is the primary source of news and political information for a growing number of citizens (Fletcher and Nielsen, 2018; Shearer, 2018). This profound shift to digital news means that the number of possible ways people are exposed to political content multiplies dramatically. Individuals now exist in unique media environments characterized by flows of information that are filtered by users themselves, their social contacts, and algorithms (Thorson and Wells, 2016). These changes have theoretical implications for how people encounter news and the types of news they see. On the one hand, digital technologies offer easy access to political information and diverse perspectives, including socially shared news. On the other hand, these technologies also make it possible for users to avoid news altogether and—for those politically inclined—to turn to partisan sources and congenial content. Identifying the extent to which people are exposed to (diverse) news through various online channels—be it webmail or social media—is needed to better understand the nature of the media environment today

And yet, relatively little is known about how various channels to news affect news exposure. This gap remains because research often focuses on how individual motivations or political biases drive exposure to entertainment versus news or to congenial political content (Feldman et al., 2013; Wojcieszak, 2019). Although this work illustrates why people select news, it says little about the avenues through which people are actually exposed to news online. Recent important work investigating the role of algorithms and social networks is also limited in that it studies either news exposure (Fletcher and Nielsen, 2018; Möller et al., 2020; Scharkow et al., 2020) or like-minded exposure (Cardenal et al., 2019). Focusing on either volume or congeniality masks important patterns of news use. For example, a single source (e.g., social media) may promote more news consumption (which is democratically beneficial) yet concurrently direct users to predominantly congenial sources (which many observers find problematic). Studies that examine both the volume and the diversity of news exposure across multiple online avenues are rare and only provide aggregate-level evidence (Flaxman et al., 2016) or focus on a single platform (Bakshy et al., 2015). As importantly, most work tests direct visits, social media, and search engines (Flaxman et al., 2016; Möller et al., 2020) and multiple other avenues remain understudied. As Thorson and Wells (2016) note, the complexity of the contemporary media environment requires scholars to test multiple, competing avenues in the same study to fully understand the nature of news exposure.

Heeding this call, we examine six avenues to news—including some that have not been analyzed before—(1) direct access to news websites, (2) social networks, (3) news aggregators, (4) search engines, (5) webmail, and (6) hyperlinks in other news. This is the first project to compare all of these avenues side by side. We also attend to both stages of news exposure simultaneously, testing the extent to which each avenue promotes exposure to news and also to congenial, dissimilar, and centrist news sources. Methodologically, we rely on web browsing history records from 636 social media users in the United States. The data contain millions of unique visits to websites, with hundreds of thousands of those to news domains, over a ninety-day time period. These behavioral data are paired with users’ survey responses, allowing us to assess whether the visited sites are politically congenial with the user. This approach also makes it possible to provide evidence on both aggregate and individual-level patterns, an advancement over past work.

We use these data to examine two parsimonious questions: (1) what is the proportion of news sites accessed via the six avenues and also (2) how do these avenues shape the political leaning of the news users encounter? In so doing, we shed new light on news consumption online and also speak to the literature on news diversity (see Hendrickx et al., 2020; Joris et al., 2020; Loecherbach et al., 2020) by showing whether a single avenue promotes exposure to both congenial and dissimilar sources. Our additional analyses also capture the extent of congenial vs. dissimilar exposure more granularly, as well as exposure to left- and right-leaning sites classified as politically extreme.

Democratic Importance of News Exposure and Diverse News Exposure

Understanding the avenues through which users encounter news online is theoretically and practically important. Theorists agree that democracies work best when citizens are up-to-date on political events. Therefore, news exposure—by enhancing knowledge and stimulating political engagement—is seen as key to a well-functioning society (Delli Carpini and Keeter, 1996). At the same time, citizens should attend to diverse outlets (as few regard heavy reliance on one-sided partisan sources as beneficial). Because the multiplicity of choice online makes it easy for some citizens to avoid news altogether, and also—for those politically interested—to access sources that are consistent with their viewpoints, the questions of (1) via which avenues do users encounter news and (2) which avenues promote exposure to sources from across the political spectrum are normatively important. To address this latter question, we draw on the literature on news diversity. Our core interest is in “exposure diversity” (Helberger et al., 2018; Napoli, 1999), or the extent to which audiences are actually exposed to an array of content and sources. Such exposure is hoped to promote “representative thinking” (Arendt, 1968, 241), “sound political judgment” (Page, 1996, 2), and to transform citizens into a cohesive collective (Barber, 1984). Establishing clear empirical benchmarks for diversity is difficult (Joris et al., 2020) and what aggregate balance of exposure to centrist, congenial, and dissimilar news is most beneficial remains an open normative question. At a minimum, however, exposure diversity means that people encounter opposing viewpoints and a range of ideological perspectives in public debates (Helberger et al., 2018).

We conceptualize diversity as disparity, which implies the distance between two elements (Loecherbach et al., 2020). Unlike most studies that consider the distance between news outlets or articles, we examine the distance between the ideology of the news outlets a user visits and the user’s self-reported political leaning, which allows for news exposure to be categorized as congenial, dissimilar, or centrist. According to this definition, both centrist and dissimilar exposure can be seen as diverse, as each represents different degrees of distance from an individual’s ideological stance.

This approach allows us to investigate which avenue promotes exposure to (1) centrist outlets, which tend to appeal to broad audiences, present balanced news from both sides, and give equal voice to various political factions (being diverse in terms of content and in relation to the user’s ideology), (2) dissimilar outlets that provide mostly counter-attitudinal views (being diverse in terms of political leaning), and (3) congenial partisan outlets from one’s own side. Furthermore, we can test whether any particular avenue facilitates (or limits) exposure to congenial outlets but simultaneously encourages consumption of centrist or dissimilar sources, thereby promoting (or inhibiting) diversity in one’s news diet. Practically, because several of the tested avenues (e.g., social media, search engines) are often blamed for putting users in echo chambers (Pariser, 2011), addressing the extent to which this in fact happens and how these avenues fare in comparison with other online sources is of importance.

News Exposure Online

The current media environment offers unprecedented ways to access news, including avenues that bring to the forefront of individual motivations or social connections (Thorson and Wells, 2016). The most straightforward way to access news is to seek it out directly, for example by setting a news site as one’s homepage or typing in the URL. Because people use media that satisfy their needs and desires, and because politics is sometimes perceived as complex or boring, few people access news in this way. Direct browsing accounts for a small percentage of individual activity; only about one-third of Americans report directly visit news sites (Newman et al., 2018; Shearer, 2018) and the majority prefer to consume entertainment instead (Prior, 2007). Although these estimates are low from the informed citizenry perspective, purposeful news seeking is only one possible avenue through which users access news and other ways of “distributed discovery” (Toff and Nielsen, 2018) abound online.

For one, social media users exist in networked flows of information, in which they can encounter news through friends and via the pages or groups they follow (Klinger and Svensson, 2015; Thorson and Wells, 2016). Platforms like Facebook, Twitter, or YouTube serve as intermediaries between users and news organizations and are prominent information sources (Newman et al., 2018). Social media connections may trump the role of individual motivations that drive direct news access. Surveys suggest that news exposure on social media is prevalent (e.g., 55 percent of Americans say they get some news from social media; Pew, 2019), and especially among users who do not actively seek news (Fletcher and Nielsen, 2018). Addressing the limitations of self-reports of exposure (Prior, 2009), trace data offer less sanguine estimates. Examining URLs shared by partisan Facebook users, Bakshy et al. (2015) find that only 13 percent of these URLs were hard news and Flaxman et al. (2016) show that about one in 300 outbound clicks from Facebook correspond to hard news. Nonetheless, people do encounter news on social networks at least occasionally.

In addition, users can access news through search engines. Search is a key function of the Internet and search engines are widely used and highly trusted for seeking politics. Twenty percent of Americans see search as their main gateway to news (Newman et al., 2018) and some view Google as the only unbiased way to get information (Tripodi, 2018). Users can look for updates on issues or events (e.g., googling where a candidate stands on a policy) or be directed to a news site after typing in nonpolitical keywords (e.g., landing on an article on climate change after searching for electric cars). Yet, although evidence speaks to the role of search in exposure diversity, as noted below, the work on the extent to which real-world individual-level search queries direct users to news is rare.

Although these three avenues (direct visits, social media, and search engines) are sometimes studied, we do not know how they compare with other, less explored paths to news facilitated by digital media. Scant evidence exists on the role of aggregators, webmail, and hyperlinks in news exposure. News aggregators are a growing, albeit still less prevalent, source. Seven percent of Americans report aggregators as their primary way to learn about politics (Newman et al., 2018), with Google News being the most popular aggregator that refers users to news websites (e.g., these referrals may account for more than 30 percent of a site’s traffic; Sherk 2014). Users also encounter news in emails from friends or family or if they subscribe to newsletters from politicians or nongovernmental organizations. Once a person accesses a link in an email, a visit is recorded and reflects mail-driven news exposure. Lastly, the hyperlinked nature of online content allows users to click on links embedded in the text or in a sidebar of articles, leading to more news that contextualize the story (Eveland et al., 2004).

Despite calls to do so (Thorson and Wells, 2016), few studies compare how people arrive at news from all these avenues. Those that do rely on aggregate site-level audience metrics and lack survey information on individual uses (e.g., Flaxman et al., 2016), or compare a limited number of avenues (i.e., direct visits, social media, and search; Möller et al., 2020; Scharkow et al., 2020). This project addresses these shortcomings to ask what percentage of users see news through direct visits, social networks, and search, from aggregators, webmail, and by following hyperlinks in other news, and also which avenue is the most dominant (RQ1).

Ideological News Exposure

News exposure is important, yet it is not an end in, and of, itself. Once users encounter news, the second level of exposure concerns the nature of that content, whether congenial (i.e., supporting individual political leanings), dissimilar (i.e., countering users’ priors), or centrist (i.e., mainstream sources presenting balanced perspectives). Are the different avenues more or less likely to drive exposure to different types of news?

In terms of direct visits, the selective exposure framework suggests that citizens should seek outlets that support their beliefs (see Stroud, 2011). Surveys find that partisans use news outlets consistent with their beliefs (Stroud, 2010) and experiments show that people often select pro-attitudinal messages (Knobloch-Westerwick and Meng, 2011). At the same time, however, experimental participants do select balanced content when given the opportunity (Brenes Peralta et al., 2016; Feldman et al., 2013; Garrett and Stroud, 2014; Winter et al., 2016), and online trace data show that most users visit mainstream news (Flaxman et al., 2016; Guess, 2021). Given the greater reliability of trace data, we expect that direct visits should result in exposure to centrist news sites to a greater extent than to congenial and—especially—dissimilar exposure.

When it comes to “distributed discovery” of news (Toff and Nielsen, 2018), social media offer opportunities to connect with like-minded others, and thus online network homophily may reinforce individual motivation for selectivity (Bakshy et al., 2015; Barberá et al., 2015). These networks may then act as information disseminators that reduce exposure to diversity. Online behavioral data indeed show that Twitter users discuss politics with in-partisans (Barberá et al., 2015; Colleoni et al., 2014; Himelboim et al., 2013) and that Facebook users are exposed to congenial news shared by their network (Bakshy et al., 2012). However, because social media facilitate connections to both strong and weak ties, they hold the potential for diverse and dissimilar exposure (Beam et al., 2018; Lu and Lee, 2018).

Turning to other avenues, search and aggregators may limit political diversity due to ranking and recommendation algorithms, which personalize information, recommendations, and search output in ways that align with individuals’ interests, attitudes, and past behavior (Pariser, 2011; Vrijenhoek et al., 2020). At the same time, the search may lead to dissimilar exposure because users see search engines as fair and unbiased (Fallows, 2005; Dutton et al., 2017; Logg et al., 2019) and use rank order, not partisan leaning, as the dominant criterion in selecting news (Nechushtai and Lewis, 2019; Puschmann, 2019). In addition, although Google News is synchronized with users’ activity across all Google applications (e.g., search, email, location, and browsing), allowing for unprecedented personalization, aggregators do not seem to promote congenial exposure. Instead, Google News yields largely the same recommendations to users with different political leanings and in distinct states, suggesting that aggregators may promote more centrist news diets by selecting news from a few mainstream conglomerates (Haim et al., 2018; Nechushtai and Lewis, 2019).

Research on which avenue leads to congenial, dissimilar, or centrist exposure is rare but does suggest some patterns. Flaxman et al. (2016) show that social media and search are associated with greater diversity than aggregators and direct visits. Relying on surveys combined with trace data from Spanish users, Cardenal et al. (2019) find that search limits congenial exposure and direct navigation enhances it. Many questions remain, however. No work examines whether news received via email or accessed through hyperlinks enhances diversity, and previous work pairing survey and trace data does not distinguish between centrist and dissimilar forms of diverse exposure (Cardenal et al., 2019). As importantly, no studies test the avenues for news and diverse exposure side by side, a gap we fill. We investigate (RQ2) which avenue leads to the greatest levels of exposure to (1) dissimilar, (2) centrist, and (3) congenial news.

Method

We need data on individual political predispositions and online behavior (e.g., which news sites people visit; what URLs they click on following a search query). We combine a survey of American adults with web traffic data from our respondents’ internet browser histories. This approach is advantageous because we examine actual exposure and—when combined with self-reported political predispositions—we know whether this exposure is congenial or uncongenial. The data are available at the Harvard Dataverse repository at https://doi.org/10.7910/DVN/PNOH5I and all additional details, analyses, robustness checks, and others are in Supplemental Information file.

We recruited participants using Facebook ads targeting adults in the US between March 28 and April 30, 2018. That year was President Trump’s second year in office, midterm elections took place in the fall, and the issues of immigration, mass shootings, and the Mueller investigation into Russian interference in the 2016 presidential election dominated the news. To test our questions, we needed to rely on social media users and Facebook was the most used platform in the US (Gramlich, 2019). The ads appeared in the right-hand column of 266,827 users’ pages and 3,735 users clicked on them (1.4 percent click-through rate). Participants were directed to a website that invited them to a survey and provided a link to Web Historian, an open-source extension for Google Chrome, that accesses respondents’ browser history stored on their computers, displays it to them using visualizations (e.g., network graph of websites visited, word cloud of used search terms, searchable table of browser history, see Supplemental Information A), and allows them to submit it to researchers for analysis (Menchen-Trevino, 2016).

After reviewing their browsing data, participants saw an informed consent form and could participate in the survey with or without submitting the browsing data. All survey participants were paid and those who uploaded browsing data could win one of five $100 Amazon gift cards. Among the 764 participants who completed the survey, 736 opted to upload their browsing data, 711 did so successfully, and 636 provided sufficient data for this analysis, as detailed below and shown in Table 1. The median participant provided data that spanned the ninety-day time period preceding the collection date, browsed on seventy six of the ninety days, and visited 432 different domains.

OPEN IN VIEWER

Table 1.

Number of Participants Who Participated in the Listed Ways.

	Participants
Successfully uploaded browsing data	711
Did not report a political leaning (ideology, party, or Trump job approval)	21
Had news browsing records with an identifiable source	653
Reported a political leaning and browsed a news site with an identifiable sourcea	636

a

Four participants reported no political leaning and did not have news browsing records with an identifiable source. Supplemental Information H further characterizes the browsing data of those fifty-eight participants without news source data.

Participant ages ranged from 18 to over 65 (median, 35-39). Education levels ranged from less than high school to a graduate degree, with a four-year college degree being the median. In total, 74.5 percent of respondents were women and 79 percent were White. The sample was geographically diverse and included respondents from 48 states and the District of Columbia. We acknowledge that our recruitment strategy and the resulting sample represent a limitation. Both the sampling strategy and the motivational thresholds to participate (i.e., sharing one’s web history), make the sample not representative of the US population, let alone internationally. This limitation is common to most work that uses data from online samples willing to share their online traces, in that no such work can claim representativeness. Individuals who take part in online studies and who share their online data are simply distinct from the general population. In Supplemental Information C, we compare the demographics of all study participants to the US population and to Facebook users, finding that our sample was more female, better educated, more left-leaning, and underrepresented Blacks and those over 65 years old. Yet, we find correspondence between the rankings of news sites accessed by our sample and those of Amazon’s Alexa, as noted in the discussion. Hence, while this limitation should be kept in mind, we believe the advantages provided by being able to test all the news avenues in a single study justify the effort.

Results

We first report findings about the volume of participants’ news consumption, which is low for many and very high for some. These patterns in the aggregate (per visit) are important for understanding what typically happens when people consume news online but do not reflect the behavior of the average person, who has a low to moderate volume of news consumption. This distinction between the aggregate visit level and the person level is important throughout our findings. The aggregate patterns shape the products of news organizations through tracked metrics (Whipple and Shermak, 2018), and the person-level data reflect individual behaviors more directly. We first present the aggregate data and person-level descriptives before examining dissimilar, centrist, and congenial exposure with multilevel models (MLM) that allow us to analyze each visit along with the individual-level differences. By using MLM, the visit level data is retained for the outcome variable, while random intercepts for individual exposure account for the variance in diversity caused by personal routines. In short, we retain nuance in the data while avoiding that individuals with very prolific or scarce news habits skew our estimates for the effects of each avenue.

Online news exposure is not frequent (Flaxman et al., 2016). Participants had a median of 65.5 unique visits to news pages over 90 days, less than one such visit per day, with the bottom quartile having a median number of visits below 19. ⁸ As could be expected, visits were largely concentrated among a small subset of avid news readers. As Figure 1 shows, the distribution of the number of news visits is highly skewed, with 74 people having more than 500 unique visits, and 64 people having fewer than 5.

OPEN IN VIEWER

Figure 1.

Histogram of total news visits per participant during the ninety-day time period studied ( n = 636 , range 1–8, 214).

Avenues for News Exposure

Among these visits, in aggregate, accessing news through hyperlinks in news was the main source of news consumption volume 47 percent (77,799), yet—because one needs to be reading news to access more news in this way—direct access is the most common way to initially arrive on news, 34 percent (55,932). Figure 2 depicts the aggregate patterns, which reflect primarily the behavior of the heavy users, whose news browsing makes up the bulk of the aggregate data.

OPEN IN VIEWER

Figure 2.

Count of total visits to news domains ( n =136,923) for each source used to access news web pages, colored by ideological congruence between the participant and the outlet’s ideology score indicated by color: (a) total visits per source and (b) percent of visits per source.

Figure 3 illustrates the data per person, or individual-level browsing, such that if a person ever used the avenue to access news they are counted once. The figure depicts different patterns. Although, as above, the majority (87 percent or 553) of participants visit news via hyperlinks in news, the avenues for primary access to news change. Search engines (78 percent or 494 use them at least once to access news) and social media (73 percent or 467 access news this way) play a large role, followed by direct access (71 percent or 453 people). Although webmail is less frequent, its share is nevertheless notable (32 percent or 209 users). In contrast, aggregators (9 percent, or 56) are substantially less frequent as news avenue.

OPEN IN VIEWER

Figure 3.

Count of people who used each source to access news web pages, colored by ideological congruence between the participant and the outlet’s ideology score indicated by color ( n = 636 ): (a) visits to news web pages per person and (b) percent of news web page visits to a source per person.

Addressing our RQ1, redirects via hyperlinks in other news are the dominant avenue for news exposure, yet—again—because one needs to be reading news to take advantage of its hyperlinked nature, we consider it to be a “secondary” avenue. On an individual level, hyperlinks are closely followed by search engines, social media, and direct access. ⁹

Ideological Congruence of Exposure

The next question regards the second step in the process, namely whether news challenges users’ views, reaffirms them, or is largely centrist. In terms of the aggregate diet (163,826 visits), a small majority of visits, 53 percent (86,028), were to centrist sites. This is consistent with past work, according to which mainstream sites dominate among the population. Due to the political composition of our sample, exposure to left-leaning domains was roughly two times higher (52,128) than exposure to right-leaning ones (25,670). In addressing our RQ2, we first present aggregate data on the volume of visits, followed by person-level results, and finally multilevel models predicting each of these types of exposures. Figure 2 shows the aggregate visits per source by ideological congruence and Figure 3 provides an overview of how many users accessed dissimilar, centrist, or congenial news via each source on an individual level. ¹⁰

Dissimilar Exposure

In aggregate (Figure 2a), direct navigation (50 percent) and hyperlinks (27 percent) form the majority of dissimilar exposure. ¹¹ Turning to the individual level (Figure 3a), a different pattern emerges because many participants consume news infrequently. Most people visited multiple sites so each individual could access sites assigned to their own side only, both sides, or their own side’s sites plus centrist sites, or only centrist sites, or only centrist sites and the other political side, or (in very few cases) only dissimilar sites (see legend, Figure 3). Because those in the “both sides,” “other side and center,” and “other side only” categories encounter some uncongenial content, we count them as having been exposed to some dissimilar and diverse news. Figure 3b reflects these “mixes” of exposures. Search engines are the key avenue that exposes users to news from across the political spectrum, with 76 percent (376) of search users ( n =494), or 59 percent of all participants, accessing some dissimilar news via search. For social media news users ( n = 467 ), dissimilar exposure drops, but remains a majority share with 69 percent (321), 50 percent of all participants. For those who access news directly ( n = 453 ), the figure drops to 43 percent (194), or 31 percent of all participants. These individual patterns suggest that search engines and social media are drivers of more diverse exposure, not engineers of like-minded echo chambers or algorithmically curated filter bubbles.

However, both the individual and aggregate descriptive data are imperfect; the former because so much is contributed by the most active users, and the individual level because it does not take into account the proportion of exposure (e.g., a person with 1,000 congenial exposures and one dissimilar would be in the “both sides” category). To offer more stringent evidence, we estimate a multilevel logistic regression model with random intercepts. ¹² The data structure is nested; each visit (level 1) belongs to a person (level 2), allowing us to combine the visit- and person-level information examined separately above. We predict dissimilar exposure from the tested avenues, with direct access being the reference (i.e., positive coefficients indicate that an avenue predicts a greater likelihood of dissimilar exposure than direct visits). We also include individual ideology as a covariate to test whether the left or right is exposed to more congenial versus dissimilar news. Figure 4 plots the estimates from this model, depicting the odds ratio of dissimilar exposure with red triangles across the avenues.

OPEN IN VIEWER

Figure 4.

Plot of odds ratios for hierarchical linear models.

Consistent with the descriptive findings, search engines and social media are significantly more likely to expose people to dissimilar news than direct access (1.57 and 1.25 times more, respectively), while hyperlinks are significantly less likely, with odds of just 0.53. The model also shows that conservative ideology is a significant predictor of dissimilar exposure, with the conservative participants being over four times more likely to access left-leaning news than liberals were to access right-leaning sources. ¹³

Addressing RQ2a, the most stringent multivariate model shows that search engine-sourced news use is associated with dissimilar exposure more than other avenues to news (see the red triangles in Figure 4).

Centrist News Exposure

Our RQ2b inquired about which avenue exposes people to centrist news. In terms of the aggregate volume of visits, direct access was the largest primary avenue from which users were directed to centrist news (28 percent), while hyperlinks within news sites were the largest avenue overall (59 percent) (see Figure 2a). These sites are visited frequently, and their design—with numerous within-site hyperlinks—successfully generates additional visits. At the individual level (Figure 3), hyperlinks also dominate (80 percent of participants accessed centrist content in this way), but search (66 percent) and social media (58 percent) play a more dominant role than direct access (50 percent) at the individual level.

Turning to the MLM predicting centrist exposure, we find a significant difference between direct access and that based on hyperlinks, with the latter being over twice as likely (2.1) to lead to centrist news. Search engines did not significantly differ from direct access, while social media were less likely to yield centrist visits, with an odds ratio of 0.85 (see the gray circles in Figure 4). This model does not find differences between those on the right and the left, suggesting they were equally likely to visit centrist news (see Supplemental Information J).

Congenial Exposure

Lastly, although congenial exposure may seem like the inverse of dissimilar exposure, they are empirically distinct. Each avenue can (and does) lead to partisan sites from both sides as well as centrist outlets. In the aggregate, direct visits are the largest source of congenial exposure 39 percent, with hyperlinks a close second at 38 percent, followed by social media at 12 percent, and search at 8 percent, and with webmail and aggregators providing a small amount of congenial content. However, at an individual level, search and social media expose the most people to some congenial news sites (87 percent and 86 percent of their users, respectively), indicating their potential for both counter- as well as pro-attitudinal exposure. Direct visits and hyperlinks expose 77 percent and 78 percent, respectively, to dissimilar news, as such visits are predominantly to centrist sites (whereas search or social media yield more partisan exposure; see also Flaxman et al. 2016).

Another MLM was estimated to better understand the visit- and individual levels of data together for congenial news exposure (see the blue diamond shapes in Figure 4). Only social media (1.12) and webmail (1.81) had higher odds of producing such exposure than direct access, with search significantly lowering the odds (0.83). Also, conservatives were significantly less likely than liberals to access congenial news, with an odds ratio of 0.35, suggesting that right-leaning participants were more likely to visit dissimilar news and also less likely to visit congenial news domains.

Addressing RQ2c, webmail and social media encourage the most congenial exposure relative to direct access according to the hierarchical linear model.

Considering our definition of diverse exposure as that which is distant from one’s political stance, coefficients for congenial exposure might be a suitable indicator of the overall diversity of each avenue when compared to direct access. In particular, interpreting the odds ratios for congenial exposure together with Figure 4 sheds light on the types of diversity promoted by each avenue. For instance, while congenial exposure from webmail comes at a cost of both centrist and dissimilar exposure, making it less diverse by any measure, social media increase congenial exposure at the expense of centrist news consumption. In contrast, while the odds ratios for congenial exposure for hyperlinks and search suggest that these avenues direct to diverse content, the type of diversity they promote is noticeably different, with hyperlinks promoting diversity through centrist exposure and search doing so through dissimilar exposure when compared to direct access. This distinction between the type of diversity encountered can only be captured by distinguishing between the three types of news in relation to one’s political stance, modelling them, and comparing the three models side-by-side.

Discussion

Most work on news exposure focuses on whether citizens seek news (finding that they rarely do) and/or whether they engage with dissimilar perspectives (finding mixed evidence contingent on the methods used). Yet, looking at the overall extent of news or diverse exposure—rather than at the avenues to such exposure—overlooks the fact that the online environment changes the media ecology and the ways in which people encounter news. We aimed to reflect the abundance of avenues to information online by examining direct access to news sites, hyperlinks in other news, social media, search, news aggregators, and webmail. We focused on which avenue leads to the greatest news exposure and also to dissimilar, centrist, and congenial exposure.

Our data, which combine individual self-reports with online traces, offer nuance to the debate about the effects of online environment on news consumption. First, we find that hyperlinks in other news are the largest avenue for more news consumption, both per visit and per person. Naturally, one needs to be reading a news article already in order to be redirected to another article, and—when accounting for initial news exposure in the aggregate—it was direct navigation that was the dominant news avenue. More importantly, at the individual level, search and social media were the key drivers of initial news exposure, and—furthermore—these two avenues mattered most to those who are least likely to access news directly, as shown in Supplemental Information I. Citizens not only stumble upon news on social media but also use search to either seek politics or land on news as a result of search output, underscoring the potential of these widely criticized avenues to informed citizenry. Research has extensively focused on incidental exposure on social media, whereas search engines have been largely overlooked. This imbalance, we argue, should be ameliorated.

Second, our paper makes a contribution by distinguishing between diverse exposure through centrist and dissimilar content. If we look at diversity as dissimilarity, search engines are the key source driving access to news from across the political aisle. Over three-fourths of those who use search (and over half of our sample), land on some dissimilar news following search. Furthermore, search also lowers exposure to congenial news relative to direct visits. These findings clearly show that search engines are performing important democratic functions, driving and diversifying news exposure. Social media platforms also have this potential. They facilitate connections with diverse weak ties and communication with like-minded friends and family, thereby driving exposure to both dissimilar and congenial news. The finding that social media drive congenial and dissimilar exposure, and not centrist exposure, also points to their potential polarizing nature: users may be more likely to encounter links to partisan outlets from both sides than to less extreme news in their Facebook feed or on Twitter (as confirmed in the additional analyses in Supplemental Information L). In short, although there is some discussion that social media enable homogeneous networks and that algorithms in search engines minimize users’ chances to encounter dissimilar news, we suggest that both avenues are central to informed citizenry and encourage researchers to further study search engines as key news agents that diversify exposure. In turn, if we see centrist exposure as desirable due to its ideological content diversity, news aggregators are a promising avenue to fulfill this normative ambition. In light of our findings, future work should look beyond aggregate or avenue diversity metrics and investigate the type of diverse exposure being promoted by each news avenue.

We note that the contributions of these avenues may be greater than what we show because we cannot speak to the domains the users saw, only to the pages they actually visited. Inasmuch as users see news in their social media feed (e.g., headlines or news videos) or read a headline that appears as a search result, we are underestimating exposure and instead capture a more meaningful engagement (i.e., accessing news by clicking on the URL). Although some surveys aim to examine this passive exposure, our study—and those that rely on online traces in general—speaks to the content people select, potentially underestimating the volume of (dissimilar) news encountered thorough social media and search.

Third, the finding regarding the importance of email as a news source extends past work, which overlooks this avenue. In our data, nearly four times as many people visited news received via email—from newsletters or their contacts—than use aggregators (again, the actual figure may be much higher as many people read headlines displayed in an email, without clicking on the outbound link). This news, moreover, is disproportionately like minded: webmail was most likely to result in congenial exposure. We argue that webmail should be considered a distinct category with different affordances. Although some activities can be done via email and social media (e.g., seeing friends’ posts vs. receiving friends’ emails; following a news organization vs. subscribing to its newsletter), email not only has distinct targeted nature but also is more successful than social media in generating subscriptions to news organizations (Stroud et al., 2018).

Our results are not free from limitations. We relied on a convenience sample that includes Facebook users who shared their online data and thus systematically differ from those who are not comfortable submitting their online traces for research. Importantly, our sample reflects the general online population in its news browsing behavior (see Supplemental Information E showing a correspondence between our news site rankings and those of Amazon’s Alexa rankings). Any study that relies on online samples and especially on samples that voluntarily share their online data (ours included) cannot make point estimates about the population. In addition, the patterns of news exposure, the importance of the tested avenues, and the nature of diverse or dissimilar exposure are very distinct in media systems outside the US, and we cannot speak to these systems. For instance, internet access is much lower in the global south than in Western countries, suggesting limited relevance of our findings to these contexts. Also, Europeans may use some avenues less (i.e., the 2019 EU copyright directive that prohibits aggregators from linking to news outlets without prior authorization), but also encounter news (Shehata and Stromback, 2011) and diverse viewpoints (Castro-Herrero et al., 2018) more uniformly than Americans thanks to strong public service broadcasting. Also, in most other contexts, ideological congruence cannot be as neatly determined as it can be in the two-party American system, and—as such—our approach to diversity and dissimilarity may not apply beyond the US. These limitations underscore the need for comparative analyses of the contributions of the tested avenues to (diverse/dissimilar) news exposure.

We also acknowledge that we cannot account for the overarching communication and information ecology of our participants. We cannot capture people’s exposure on multiple browsers, computers, or mobile devices, or speak to news exposure on other media or from interpersonal networks. Researchers cannot access all the information about all the (news) outlets people see, and likely never will. If scholars had access to Facebook newsfeed combined with mobile and desktop data, we would still be missing private message services like Messenger and WhatsApp, which are popular for sharing news (Waterson, 2018). Even if a researcher could access all of the digital traces a person produces, offline exposure and interpersonal communication are still crucial, and no trace-data-based work can account for these sources. Our project focused on systematically comparing the various avenues for online news, and so the lack of data from mobile and multiple machines is a key limitation.

In addition, our focus on domains overlooks content-level exposure. On the one hand, news and political exposure are not synonymous, as major news sites offer recipes or travel stories (i.e., we may have overestimated news exposure). On the other hand, because we examined news outlets only, we did not consider political exposure outside news domains. Inasmuch as people encounter politics in nonnews sources (e.g., an article about abortion in Women’s Health) and see current affairs when they go online for romance or gaming (Wojcieszak and Mutz, 2009), we may have underestimated political exposure. Focus on the content seen would offer more nuanced evidence.

Lastly, examining the patterns of exposure through our comprehensive list of avenues, while a valid goal in itself, is the first step only. Researchers should assess whether the various avenues influence the effects of news exposure and of centrist, dissimilar, and congenial exposure on relevant cognitions, attitudes, and behaviors, such as knowledge and participation, afective polarization, policy preferences, among others. In a related vein, it is of interest to examine whether users perceive the various news sources as diverse and dissimilar; after all perceptions matter to news media use and their effects (see Wojcieszak and Price 2012a; 2012b); and also whether users are more likely to accept counter-attitudinal news when delivered through specific avenues (e.g., search engines, which may be seen as ideologically blind).

Despite these limitations, our findings enhance our understanding of contemporary news exposure. The numerous functionalities of the online environment differently influence how citizens encounter news and what news they see. Citizens may engage with and process information differently depending on where it is accessed (Möller et al., 2020), and the unique characteristics of various avenues clearly influence exposure to diverse and dissimilar news. Our project offers a comprehensive portrayal of the complex online news ecosystem and suggests important directions for future work. Only when we theorize and study these various avenues together will we be able to understand their unique contributions to informed and tolerant citizenry. Considering multiple aspects of the rapidly evolving online information environment shows how it changes both the volume and type of contemporary news consumption.

Supplemental Material