Request a Note: How the Request Function Shapes X's Community Notes System

TL;DR

It is found that requested posts with higher GPT-estimated misleadingness and from authors with greater misinformation exposure are more likely to receive notes, and posts from Republicans are more likely to receive notes than those from Democrats.

Abstract

X's Community Notes is a crowdsourced fact-checking system. To improve its scalability, X introduced ``Request Community Note'' feature, enabling users to solicit fact-checks from contributors on specific posts. Yet, its implications for the system -- what gets checked, by whom, and with what quality -- remain unclear. Using 98,685 requested posts and their associated notes, we evaluate how requests shape the Community Notes system. We find that requested posts with higher GPT-estimated misleadingness and from authors with greater misinformation exposure are more likely to receive notes. Conversely, requested political posts (vs. non-political) are less likely to receive notes. We also observe partisan asymmetries: posts from Republicans are more likely to receive notes than those from Democrats. Although only 12% of requested posts receive request-fostered notes from top contributors, these notes are rated as more helpful and less polarized than others, partly reflecting top contributors' selective fact-checking of misleading posts. Our findings highlight both the limitations and promise of requests for scaling high-quality community-based fact-checking.

Figures (10)

• Figure 1: The estimation results for the logistic regression model predicting the likelihood that a requested post receives a community note. Shown are coefficient estimates with 95% Confidence Intervals (CIs). The coefficients for misinformation exposure score and partisan score are estimated based on the subset of posts containing the corresponding information. The number of words in each post is controlled during estimation but omitted in visualization for better readability. Continuous independent variables---word count, account age, number of followers, and number of followees---are z-standardized before estimation to facilitate interpretation.
• Figure 2: The statistics for the request timing relative note writing. (a) The Complementary Cumulative Distribution Functions (CCDFs) for the hours from post creation to the submission of first request and to the submission of fifth request. (b) The CCDFs for the hours from post creation to the submission of fifth request and to the generation of first community note, respectively. (c) The CCDFs for the hours from post creation to the notes written by top writers versus other contributors. (d) The sankey plot illustrating the estimated proportion of posts for which community notes are likely fostered by requests.
• Figure 3: The estimation results for the logistic regression model predicting the likelihood that a requested post receives a top-writer note after the request threshold, i. e., request-fostered notes. Shown are coefficient estimates with 95% CIs. The coefficients for misinformation exposure score and partisan score are estimated based on the subset of posts containing the corresponding information. The number of words in each post is controlled during estimation but omitted in visualization for better readability. Continuous independent variables---word count, account age, number of followers, and number of followees---are z-standardized before estimation to facilitate interpretation.
• Figure 4: Overview of note evaluations and source domains in community notes. (a) The distribution of estimated note intercepts (helpfulness) and estimated note factors (polarization) from the note selection algorithm. (b) The distributions of estimated note intercepts (helpfulness) and note factors (polarization) from the note selection algorithm between request-related notes and writer-only notes. (c) The distributions of estimated note intercepts (helpfulness) and note factors (polarization) from the note selection algorithm between request-fostered notes and writer-only notes. (d) The distributions of estimated note intercepts (helpfulness) and note factors (polarization) from the note selection algorithm between request-related notes and request-fostered notes. (e) The violin plots showing the distributions of domain bias in community notes across the three categories: writer-only, request-related, and request-fostered. (f) The violin plots showing the distributions of domain quality in community notes across the three categories: writer-only, request-related, and request-fostered.
• Figure 5: The estimation results for the linear regression model predicting note helpfulness. Shown are coefficient estimates with 95% CIs. The number of words in each post is controlled during estimation but omitted in visualization for better readability. Continuous independent variables---word count, account age, number of followers, and number of followees---are z-standardized before estimation to facilitate interpretation.