Misunderstanding the harms of online misinformation

Summary

Budak, Nyhan, Rothschild, Thorson, and Watts argue that public discourse about online misinformation has drifted away from what empirical research actually shows. They identify three persistent misperceptions — about the prevalence of problematic content, the causal role of algorithms, and the link between social media and broad societal harms — that pervade commentary by journalists, public intellectuals, and policymakers. The paper urges a recalibration of debate and policy so that interventions target the actual, more concentrated and heterogeneous patterns of misinformation exposure and effect documented in the literature.

Key Contributions

• Names and articulates three recurring misperceptions in public commentary about online misinformation.
• Foregrounds the gap between alarmist popular narratives and the empirical consensus in communication and political science research.
• Offers a corrective framework for journalists, policymakers, and researchers to characterize misinformation harms more accurately.

Methods

The paper is a synthetic review and critical analysis. The authors survey the empirical literature on online misinformation exposure and effects and juxtapose its findings against representative claims in public discourse to surface systematic mismatches.

Findings

• Exposure to false or problematic content is concentrated within a small, atypical subset of users rather than being widespread on average.
• Algorithmic recommendation plays a smaller role in misinformation exposure than commonly asserted; user choice and social network structure matter as much or more.
• Causal claims linking social media use to large-scale societal outcomes (polarization, democratic decline, etc.) are weaker and more contested than public narratives imply.
• Policy and journalistic responses calibrated to inflated estimates of harm risk being misdirected.

Connections

This paper is the canonical statement of the “limited and concentrated effects” position and connects directly to empirical work showing that misinformation exposure and sharing are skewed toward small subpopulations, e.g. Gonzalez-Bailon2024-rq and Mosleh2024-op, as well as work re-examining algorithmic responsibility such as DeVerna2025-dl. It also speaks to critiques and reframings of the misinformation-research field itself, including Marwick2025-ov and contributions on how researchers and platforms construct the problem like Starbird2025-jj and Donovan2025-ws.