Daily top-10 to top-100 lists of English, German and French Wikipedia did not turn over faster from 2016 to 2025, and English churn decreased by 8–14%

Question: Did collective attention on Wikipedia become faster after 2017? Lorenz-Spreen et al. (2019) found faster attention dynamics in Twitter and other media, but not clearly in English Wikipedia for 2012–2017. Approach: We downloaded the Wikimedia daily lists of the 1000 most-viewed pages (user agent) for English, German and French Wikipedia, all-access and mobile-web, 2015-07-01 to 2026-10-21 (4123–4131 days for each list). We removed non-article pages and ranked the articles again. For each calendar year from 2016 to 2025 and for list sizes 10, 50, 100 and 500, we calculated daily churn, the mean days in the list for each distinct article (the Lorenz-Spreen turnover measure), one-day stays, and the shape of peaks. A third list variant kept only the all-access articles that were also in the mobile-web list on the same day. This variant is a control for automated desktop traffic. Results: English top-50 churn decreased from 0.447 to 0.390 (2016–17 to 2024–25; Theil–Sen slope −1.95%/year, 95% interval −3.10 to −1.41). Top-10 to top-100 churn decreased by 8–14% in all three English variants, and the share of one-day stays decreased by 9–19%. German and French top-10 to top-100 churn did not change clearly (top 50: ratios 0.95 and 0.99). Only the German and French top-500 lists turned over a little faster (+0.4 to +1.5%/year). In all 36 combinations of edition, variant and list size, the 95% lower limit of the residence slope was above −3.7%/year. Twitter top-50 residence decreased by about 13%/year from 2013 to 2016. In all three editions, views fell more slowly after top-50 peaks (next-day fall ratio 0.78 in English, 0.95–0.96 in German and French). Long-staying articles did not explain the results: churn of the slots that these articles did not hold had no significantly positive slope in 27 of 27 tests. Limits: The data have ten yearly values and contain some unidentified automated traffic. The lists show only the top 1000, so some peak measures are bounds. App views are not included. The study describes the change but does not identify its cause. The raw data, code and result tables are included or retained.

collective attention · Wikipedia · pageviews · attention dynamics · turnover · popularity · top lists · social acceleration · computational social science · replication