Does Web Search Personalization Polarize Political Information During U.S. Elections?

Matter and Hodler (2024) ask whether Google’s web search personalization creates ideological segregation in election-related search results during the 2020 U.S. election. They study this using a large-scale field experiment with 150 synthetic users (“bots”) assigned different partisan browsing habits and locations across 25 U.S. cities, issuing identical election-related Google searches over several months. They find substantial personalization in search results, driven mainly by location rather than individual ideology. Search results in Republican-leaning cities are about 0.05 standard deviations more conservative than in Democratic-leaning cities, rising to roughly 0.12 standard deviations for local news. Individual partisan browsing has limited effects, mainly for Democratic users.

Why This Article Was Selected for The Policy Scientist

This article addresses a policy issue of broad and growing importance: how algorithmic information intermediaries shape exposure to political information in democratic societies. As search engines increasingly mediate access to news, understanding whether personalization contributes to ideological segregation is central to debates about polarization, platform governance, and election integrity. The study is timely given heightened scrutiny of digital platforms following recent elections in the United States and abroad. Matter and Hodler, who have an established research record on media, institutions, and political economy, make a valuable contribution by moving beyond observational data. Their long-run bot-based field experiment offers unusually strong causal leverage, high-quality data, and transparent design. While the U.S. context is distinctive, the mechanisms plausibly generalize to other advanced democracies with similar media ecosystems.

Full Citation and Link to Article

Matter, U., & Hodler, R. (2025). Web Search Personalization during the US 2020 Election. American Economic Review: Insights, 7(4), 516–533. https://doi.org/10.1257/aeri.20240115

Central Research Question

This article asks whether Google’s web search personalization meaningfully shapes the ideological composition of election-related search results and, in doing so, contributes to ideological segregation during a major democratic event. The authors focus on the 2020 U.S. presidential election and its immediate aftermath, a period marked by unusually high political salience and intense public reliance on online information. The central question is not whether search results differ across users—an outcome that is largely expected—but whether those differences systematically align with political ideology in ways that could plausibly matter for political knowledge, exposure, and polarization. In particular, the study distinguishes between personalization driven by users’ inferred partisan preferences and personalization driven by geographic location, asking which mechanism dominates and under what conditions.

Previous Literature

The article builds on a large literature examining ideological segregation, selective exposure, and the role of digital platforms in shaping political information environments. Earlier work by Gentzkow and Shapiro and others documented ideological segregation in media consumption and raised concerns about online “echo chambers.” Subsequent studies examined social media platforms such as Facebook and Twitter, often finding asymmetries in exposure or engagement but mixed evidence on polarization effects. Research on search engines has been more limited, in part because of data constraints. Prior studies relied on short-term audits, small-scale experiments, or observational data donated by users, which limited causal interpretation. Matter and Hodler extend this literature by focusing on Google Search, a dominant information intermediary, and by adopting a design that allows cleaner causal attribution. Their work directly engages with concerns raised by Epstein and Robertson and others while offering a more systematic and transparent approach to measuring personalization over time.

Data

The study uses data generated through a long-run field experiment involving 150 synthetic internet users, or “bots,” designed to closely mimic real human browsing behavior. These bots were randomly assigned partisan browsing preferences—Democratic, Republican, or non-partisan—and placed in 25 U.S. cities with varying partisan leanings. Over a period spanning late October 2020 through early February 2021, the bots engaged in routine browsing, partisan website visits, and non-partisan searches, thereby building distinct browsing histories. Crucially, all bots also conducted identical election-related Google searches on a daily basis. The resulting dataset includes more than 25,000 election-related searches and captures the full first page of organic search results. This design yields unusually rich data on how identical queries produce different outputs across users, locations, and time.

Methods

The authors employ a bot-based field experiment that combines randomized assignment with sustained exposure, allowing for strong causal inference. By holding search queries constant while varying prior browsing behavior and location, they isolate the effects of personalization mechanisms embedded in Google’s search algorithm. The analysis uses established similarity metrics, such as the Jaccard Index and rank-biased overlap, to quantify differences in search results across users. To assess ideological content, the authors construct a Search Result Ideology Score that aggregates multiple independent domain-level ideology measures and weights results by their prominence on the search page. Regression models with fixed effects and clustered standard errors are used to estimate how user partisanship and city-level political context affect ideological exposure. While not a randomized controlled trial in the classical sense, the design closely approximates experimental conditions and substantially improves on purely observational approaches common in this literature.

Findings/Size Effects

The study finds clear evidence that Google search results are personalized, but the ideological implications of that personalization depend on the source. Location-based personalization plays a dominant role. Election-related search results in Republican-leaning cities are, on average, about 0.05 standard deviations more conservative than those in Democratic-leaning cities, with effects reaching approximately 0.12 standard deviations when focusing specifically on local news domains. In contrast, individual partisan browsing behavior has limited effects. Democratic users see slightly more liberal national news content than non-partisan users in the same cities, but Republican users do not see comparably more conservative content. The authors attribute this asymmetry to differences in the structure of the media ecosystem, particularly the relative prominence and reach of partisan news sources. Overall, ideological segregation arises more from geography and local media markets than from individualized ideological targeting.

Conclusion

This article makes an important contribution by clarifying how search engine personalization operates in practice during a high-stakes political moment. Its primary insight is that ideological segregation in search results is real but largely driven by location rather than inferred individual ideology. The study’s experimental design, scale, and transparency represent a methodological advance in the analysis of digital information intermediaries. While the findings are rooted in the U.S. context, the underlying mechanisms plausibly extend to other democracies with geographically differentiated media systems. The use of a long-run field experiment provides stronger causal leverage than most prior work, though future research could further strengthen inference by integrating experimental variation in content exposure or downstream behavioral outcomes. Overall, the article deepens understanding of how algorithmic systems interact with existing media structures to shape political information environments.