Web Search Personalization During the US 2020 Election

We study the impact of web search personalization on ideological segregation in search results. We deploy 150 synthetic internet users with randomized partisan preferences across 25 US cities. These users generate authentic browsing histories and are active during the US 2020 election and its aftermath. Daily experiments in which the users enter identical election-related queries provide evidence for ideological segregation in search results across locations with different partisan leanings but not across users with different partisan browsing habits. We discuss the important role of contextual factors, in particular the national and local (online) media landscape, for understanding these results.