In 2026, mobile broadband internet may be crossing an important threshold: there are now more active mobile broadband subscriptions worldwide than people. This technological transformation has profound political, economic, and social implications, not least because mobile broadband provides the main infrastructure for modern social media. The expansion of social media platforms has been extraordinary. As of April 2026, platforms such as Facebook, Twitter and later X, TikTok, Instagram, and Reddit had roughly 5.8 billion users worldwide, equivalent to about 70% of the world’s population and 95% of internet users. In many countries, social media has become a major source of political news and information. In the US, more than half of the population gets at least some news from social media.

In environments where traditional media are censored, captured, or heavily restricted, social media can broaden access to information, facilitate political mobilisation, and strengthen government accountability, in some cases even contributing to democratic change (Miner 2015, Guriev et al. 2021, Donati 2023).

However, a growing body of evidence suggests that social media may also contribute to the spread of misinformation (see Allcott and Gentzkow 2017 and Vosoughi et al. 2018, as well as the surveys by Zhuravskaya et al. 2020 and Aridor et al. 2024). The advertising-based business model of modern social media platforms tends to reward attention-grabbing and inflammatory content, which is often misleading or outright false (Tufekci 2018, Haidt and Rose-Stockwell 2019, Beknazar-Yuzbashev et al. 2025). Recent advances in artificial intelligence have further increased these concerns by making misinformation easier and cheaper to produce while also increasing its apparent credibility (Campante et al. 2025). Previous research has shown that misinformation can be highly persuasive and difficult to counter through fact-checking interventions that expose people to true information (Barrera et al. 2020, Nyhan et al. 2020). Limiting the spread of misinformation online is therefore an important policy objective.

How can misinformation on social media be countered? One widely discussed approach is fact-checking. Julia Cagé et al. (2025) show that third-party fact-checking of Facebook posts significantly reduces engagement with content labelled as false and lowers subsequent user activity. Henry et al. (2022) demonstrate that merely making users aware that fact-checks are available substantially reduces the sharing of misinformation, regardless of whether users actually consult them.

However, as we showed in an experiment conducted during the 2022 US midterm elections (Guriev et al. 2023), fact-checking presents an important trade-off: while it slows the circulation of false news, it also reduces the sharing of true news. This creates tensions with the incentives of social media platforms, whose business models rely heavily on user engagement.

Our findings instead point to the effectiveness of softer, content-neutral interventions, such as nudges and priming (Pennycook and Rand 2022;,Kozyreva et al. 2024). In particular, in our experiment the most effective intervention was a simple reminder: “Please think carefully before you retweet. Remember that there is a significant amount of false news circulating on social media.” Analogous to health warnings on cigarette packs or alcohol bottles, this short message was the only intervention that significantly reduced the sharing of false news without affecting the sharing of true news.

The 2022 experiment was based on four news items selected by the researchers, two pro-Republican and two pro-Democrat, with one true and one false item in each group. These items were presented to participants, who could choose whether to share one of them.

Have recent advances in AI and changes in the US political environment altered this picture? During the 2024 US presidential campaign, we conducted another large-scale experiment involving 10,000 voting-age users of the X social media platform and 40 news items rather than four (Guriev et al. 2026). Participants were exposed to a broad set of systematically collected true and false news stories that had actually circulated during the campaign and were based on campaign speeches and media reports associated with both Republican and Democratic politicians. 

The results largely confirmed our earlier findings from 2022, namely, that the nudge performs best, but also yielded important new insights. The broader range of political content and the larger sample allowed us to examine heterogeneity in responses across both users and types of misinformation.

First, we found that while information associated with both Republican and Democratic candidate included both true and false content during the campaign, false content favouring the Republican candidate, Donald J. Trump, was far more abundant than false content favouring the Democratic candidate, Kamala Harris: for every pro-Democratic false story, we identified roughly ten pro-Republican false stories.

Even more importantly, participants had difficulty navigating the information environment. When asked to assess the veracity of news stories in an incentivised setting, they often struggled to distinguish true stories from false ones. Figure 1 presents the distribution of veracity assessments assigned by participants to true and false news stories circulating during the 2024 US presidential campaign. The average perceived veracity of true stories was 63.9 on a scale from 0 (“certainly false”) to 100 (“certainly true”), while the average perceived veracity of false stories was 46.9 on the same scale.

Figure 1 Perceived veracity of true and false news circulating during the 2024 presidential campaign

Moreover, participants often either indicated uncertainty by assigning a score of 50 or expressed confidence in an incorrect answer, making the distributions of veracity assessments appear almost symmetric. One possible explanation is that false campaign messages were intentionally formulated in a misleading way to resemble true information. This difficulty distinguishing true from false news underlies our second main result. This contrasts sharply with perceived partisanship of campaign news: unlike veracity assessments, the distributions of partisanship assessments are strongly one-sided for posts from both pro-Democratic and pro-Republican outlets (Figure 2).

Figure 2 Perceived partisanship of news stories from pro-Democratic and pro-Republican outlets circulating during the 2024 US presidential campaign 

Mimicking the structure of the previous experiment, participants were presented with random subsets of these news stories such that each participant saw two posts from pro-Republican outlets and two from pro-Democratic outlets, with one true and one false item in each group. Because news items were randomly assigned to participants, we can examine heterogeneity in treatment effects depending on the type of news participants were exposed to.

In the control group (i.e. in the absence of any intervention), respondents assigned higher average veracity scores to many false posts than to many true posts. Only 55% of true posts in the 2024 experiment received average veracity scores above the highest score assigned to any false post in the sample. We classify these posts as “uncontroversial true” in the heterogeneity analysis, as the average respondent would not confuse them with even the most believable false posts. In Figure 3, we present examples of controversial and uncontroversial true news, as well as believable false news, drawn from our sample.

Figure 3 Example of controversial and uncontroversial campaign news

While priming is the most effective intervention in curtailing false news and amplifying truth across all sets of news, it is particularly effective when participants are exposed to uncontroversial true content. In this case, concerns about appearing to be a credible source of information push participants toward sharing more uncontroversial true news and less false news to a significantly greater extent than when no uncontroversial true content is available to anchor this choice. By contrast, when participants face a confusing information environment in which they cannot easily distinguish true from false content, they are more likely to disengage and reduce sharing overall. Even in this setting, however, priming reduces overall sharing primarily by reducing the sharing of false news, thereby improving discernment.

The magnitudes are striking. Among participants exposed to uncontroversial true content, priming significantly increased the sharing of true news by 4 percentage points (from 24% in the control group) and significantly reduced the sharing of false news by 8 percentage points (from 23%). In contrast, among participants presented with controversial true content, priming had no significant effect on the sharing of true news and reduced the sharing of false news by 5 percentage points. The difference in treatment effects across the two subsamples is statistically significant. Interestingly, fact-checking (although outperformed by a simple nudge) is also more effective in improving discernment when participants are exposed to uncontroversial true content.

The finding that exposure to uncontroversial truth helps social media users better distinguish false from true news, and that interventions such as priming are particularly effective in this environment, is our third main result. It has important policy implications.

Anti-misinformation policies do not operate in a vacuum; their effectiveness depends on the surrounding information environment. Interventions work best when users are exposed to credible, relatively uncontroversial true information that helps anchor their judgments. Such content often originates from mainstream news organisations, either directly or through social media. By contrast, when users face a confusing information environment in which true and false content are difficult to distinguish, they are more likely to disengage and reduce sharing overall.

Thus, fighting misinformation is not only about slowing the spread of falsehoods; it is also about ensuring the supply and visibility of credible information. The demand for and supply of mainstream news are therefore essential for combating misinformation. Encouragingly, Campante et al. (2025) show that exposure to AI-generated disinformation increases users’ demand for mainstream news.

However, the supply of mainstream news on social media depends on platform algorithms. As shown in Gauthier et al. (2026a), the X algorithm changes users’ information environment in ways that make it substantially harder to navigate: the algorithm demotes accounts of traditional news organisations and promotes those of political activists (defined as regular users who post extensively about politics and who cannot be classified as media organisations, governments, or other organisations). Posts from news media appear 15.5 percentage points less often (a decline of 58.1%) in the algorithmic X feed relative to the chronological feed, whereas posts from political activists appear 5.9 percentage points more often (an increase of 27.4%) (see also Gauthier et al. 2026b).

Taken together, the findings in Gauthier et al. (2026a) and Guriev et al. (2026) point to the need for regulatory oversight of social media platforms. First, platforms should be required to provide transparency about how their algorithms rank different types of content. Second, policymakers could consider minimum visibility standards for content produced by professional news media, similar to obligations that exist for television broadcasters. Finally, public support for quality journalism and media literacy becomes even more important in the age of social media, AI, and algorithmic feeds.

In an information environment increasingly shaped by algorithms and AI-generated content, preserving the visibility of credible news sources may be as important as correcting false information itself.

References 

Allcott, H and M Gentzkow (2017), “Social Media and Fake News in the 2016 Election”, Journal of Economic Perspectives 31(2): 211–236.

Aridor, G, R Jiménez-Durán, R Levy and L Song (2024), “The Economics of Social Media”, Journal of Economic Literature 62(4): 1422–1474.

Barrera, O, S Guriev, E Henry and E Zhuravskaya (2020), “Facts, Alternative Facts, and Fact Checking in Times of Post-Truth Politics”, Journal of Public Economics 182: 104123.

Beknazar-Yuzbashev, G, R Jiménez-Durán, J McCrosky and M Stalinski (2025), “Toxic Content and User Engagement on Social Media: Evidence from a Field Experiment”, The Warwick Economics Research Paper Series (TWERPS) 1543, University of Warwick, Department of Economics.

Campante, F, R Durante, F Hagemeister and A Sen (2025), “AI misinformation and the value of trusted news”, VoxEU.org, 16 September.

Cagé, J, N Gallo, M Hengel, E Henry and Y Huang (2025), “Fact-checking reduces the circulation of misinformation – we should not get rid of it”, VoxEU, 21 December.

Donati, D (2023), “Mobile Internet Access and Political Outcomes: Evidence from South Africa”, Journal of Development Economics 162: 103073.

Gauthier, G, R Hodler, P Widmer and E Zhuravskaya (2026a), “The Political Effects of X’s Feed Algorithm”, Nature 652: 416–423. 

Gauthier, G, R Hodler, P Widmer and E Zhuravskaya (2026b), “How X’s algorithm shifts political attitudes”, VoxEU.org, 26 February.

Guriev, S, N Melnikov and E Zhuravskaya (2021), “3G Internet and Confidence in Government”, Quarterly Journal of Economics 136(4): 2533–2613.

Guriev, S, E Henry, T Marquis and E Zhuravskaya (2023), “Evaluating anti-misinformation policies on social media”, VoxEU.org, 10 December.

Guriev, S, E Henry, T Marquis and E Zhuravskaya (2026), “Curtailing False News, Amplifying Truth”, accepted at Econometrica. 

Haidt, J and T Rose-Stockwell (2019), “The Dark Psychology of Social Networks: Why It Feels Like Everything Is Going Haywire”, The Atlantic. 

Henry, E, E Zhuravskaya and S Guriev (2022), “Checking and Sharing Alt-Facts”, American Economic Journal: Economic Policy 14(3): 55–86.

Kozyreva, A, P Lorenz-Spreen, S M Herzog et al. (2024), “Toolbox of Individual-Level Interventions Against Online Misinformation”, Nature Human Behaviour 8(6): 1044–1052.

Miner, L (2015), “The Unintended Consequences of Internet Diffusion: Evidence from Malaysia”, Journal of Public Economics 132: 66–78.

Nyhan, B, E Porter, J Reifler and T J Wood (2020), “Taking Fact-Checks Literally but Not Seriously? The Effects of Journalistic Fact-Checking on Factual Beliefs and Candidate Favorability”, Political Behavior 42(3): 939–960.

Pennycook, G and D Rand (2022), “Accuracy Prompts Are a Replicable and Generalizable Approach for Reducing the Spread of Misinformation”, Nature Communications 13: 2333.

Vosoughi, S, D Roy and S Aral (2018), “The Spread of True and False Information Online”, Science 359: 1146–1151.

Zhuravskaya, E, M Petrova and R Enikolopov (2020), “Political Effects of the Internet and Social Media”, Annual Review of Economics 12: 415–438.