Combating Online Harms: The Importance of Rigorous Scientific Testing for Proposed Interventions

Key claims and findings

1. Misinformation follows power-law distributions. Most people receive low doses of misinformation; a small population receives very high doses. Exposure is not uniformly distributed across the public.

2. Ideological asymmetry in exposure. People on the political right received higher exposure to misinformation on social media than people on the left during the periods studied. This asymmetry complicates "unbiased" content moderation policies.

3. Misinformation is a small part of most people's information diet. Despite media attention, misinformation constitutes a tiny fraction of what most individuals consume. Some individuals consume more, but the population average is very low.

4. Online harms extend beyond false information. A true news headline ("Doctor gets COVID shot and dies two weeks later") caused greater vaccine hesitancy than typical misinformation because readers inferred false causation. The harm came from misleading-but-true framing, not falsity.

5. Educational interventions: searching online can backfire. Encouraging people to search for corroborating evidence before assessing veracity increases belief in false news when the internet contains "data voids"—clusters of false or low-quality information about a topic. This directly challenges conventional digital literacy advice.

6. Data voids create search-engine problems. For certain misinformation topics, corroborating true information is sparse online, while false information dominates search results. Searching then reinforces false beliefs.

7. Source credibility labels reduce misinformation belief—with caveats. When people encounter false news with labels indicating the source's unreliability, they are less likely to believe it. However, the effect size varies by political ideology and other moderators.

8. Algorithmic ranking shapes exposure at scale. The US 2020 Facebook and Instagram study, tracking millions of users' feeds, found that algorithmic ranking—how the platform decides which content appears in users' feeds—significantly affects exposure to political content across the ideological spectrum.

9. Just because an intervention sounds good doesn't mean it works. The central takeaway: many conceptually appealing interventions (prompting people to search, adding source labels, algorithmic changes) have complex, sometimes counterintuitive effects in practice.

10. Rigorous, pre-registered testing is essential. Before platforms adopt an intervention or researchers recommend policy changes, proposals must be tested experimentally with transparent methods, reported hypotheses, and realistic scale.

11. Misinformation research has become politically fraught. Platform moderation efforts targeting misinformation, especially around the 2020 election and pandemic, led to accusations of political bias. This politicization eventually resulted in federal restrictions on misinformation research funding, constraining the field.

12. Generative AI will reshape the information environment. As LLMs and synthetic media become widespread, maintaining data voids and combating online harms will require new strategies and continued empirical testing.

Research design and interventions studied

Tucker frames the talk around four types of interventions:

• Educational interventions — act on individual cognition: digital literacy, searching, fact-checking, source credibility labels, warnings
• Technological interventions — algorithmic changes: ranking, recommendation, content filtering
• Social interventions — peer or community feedback, human-provided corrections
• Psychological interventions — accuracy nudges, pre-bunking, priming social norms

The three empirical studies presented are:

1. Online searching study (lead author: Kevin Alett; published Nature 2024; conducted at Center for the Study of Social Media and Politics, NYU) - Five separate experiments measuring whether searching for corroborating evidence improves or harms accuracy in assessing false news - Articles tested immediately after publication and weeks later - Theory: data voids lead to backfire effects - Finding: searching increases belief in false news when false information dominates search results

2. Source credibility study (conducted at Center for the Study of Social Media and Politics) - Measured effect of source-credibility labels/rankings on misinformation belief - Varied by political ideology and other individual characteristics

3. US 2020 Facebook and Instagram election study (6-year longitudinal project; co-chaired by Tucker; involves millions of users) - Studied the causal effect of algorithmic ranking on exposure to political content - Randomized algorithmic conditions across users - Findings on how ranking shaped feed composition and political exposure

Context

This lecture was part of the 2026 Bruce R. Andrews Lecture series at Dickinson College, initiated by student project managers at the Clarke Forum for Contemporary Issues. The Andrews fund honors a deceased professor of political science and funds vibrant discussions of politics and public life. Tucker also co-directed NYU's Jordan Center for Advanced Study of Russia and the Center for Social Media, AI, and Politics, co-edited Social Media and Democracy: The State of the Field, and chairs the American Political Science Association's Presidential Task Force on AI and Politics.

Notes

This 2026 lecture represents the current state of empirical misinformation research after a decade of investigation. Tucker's central argument—that interventions must be rigorously tested before deployment—is increasingly important as platforms and policymakers seek to address online harms. The emphasis on "data voids" and search-engine backfire effects challenges mainstream digital-literacy messaging and highlights the importance of environmental factors (what information actually exists online) in determining intervention success.

The politicization of misinformation research is also historically significant: the Trump administration's restrictions on NSF-funded misinformation research in early 2026 effectively constrained scientific investigation of a critical public-interest topic, illustrating tensions between research, policy, and partisan politics. Tucker's discussion of this chilling effect is timely and candid.

Finally, the three studies exemplify a shift in the field from observational data toward randomized experimentation and from small-scale lab studies toward platform-scale interventions involving millions of users. This methodological maturation increases confidence in causal claims about what interventions actually do in practice.