Abstract

Social media platforms have become a central infrastructure for public communication, while simultaneously enabling the rapid diffusion of misinformation. In response, platforms have increasingly adopted participatory approaches to misinformation correction, including Community Notes (CNs) on X, a crowdsourced system that allows users to add contextual annotations to potentially misleading posts. While prior research has examined the accuracy and aggregate performance of CNs, less is known about how evaluation dynamics vary across different domains and narratives. This study addresses this gap by examining immigration-related CNs, a highly contested policy area characterized by persistent political disagreement and complex claims. Using a computational approach, we analyzed a dataset of 17,014 immigration-related CNs. We applied semantic topic modeling using transformer-based sentence embeddings and density-based clustering to identify the thematic structure of these notes and the types of content that X users tend to seek to fact-check. This process resulted in the identification of 15 topics, which were further grouped into four broader thematic categories: 1) Immigration Enforcement and Legal Status; 2) Citizenship, Elections, and Political Authority; 3) Refugees, War, and International Humanitarian Claims; and 4) General Misinformation, Rumors, and Stigmatization. We then examined how the algorithmically determined outcomes that X uses to determine if a note is helpful, not helpful, or needs more ratings vary across 15 topics and 4 broad thematic categories, as well as how quickly notes transition from initial posting to receiving an evaluation status other than needs more ratings. The results show that most immigration-related CNs fall under either the immigration, enforcement and legal status category or the general misinformation category. Evaluation outcomes differ significantly across both topics and thematic categories, although effect sizes are modest, suggesting that topic membership alone does not strongly determine evaluative success. Temporal differences in evaluation speed do not consistently appear at the level of individual topics. However, when topics are aggregated into broader thematic categories, clear differences emerge; that is, notes addressing refugees, war, and international humanitarian claims take substantially longer to move out of the ‘needs more ratings’ stage than notes attempting to correct enforcement-related or general rumour-based misinformation. Future research could examine non-topical features of CNs, such as reference diversity and linguistic clarity, to better explain why some notes attract faster evaluative attention than others.

Download

Share

COinS