Wikipedia mobile-web top-50 turnover decreased in 9 of 10 language editions in 2016–2025 because fewer articles re-entered the lists; first-time entries did not decrease

Question: Lorenz-Spreen et al. (2019) reported faster collective attention on several platforms. Earlier SCI studies found slower, not faster, attention in English Wikipedia daily top lists, and no clear trend in most other editions with all-access lists. We asked if top-list turnover changed in ten editions, and which part of the list change causes the trend. Data: Wikimedia Pageviews API daily top-1000 lists (agent = user), all-access and mobile-web, for en, de, fr, es, ja, ru, it, pt, zh and pl Wikipedia, every available day 2015-07-01 to 2026-09-30 (about 80 million rows). Method: we removed non-article pages and 'resident' articles (in the top 100 on 25% or more of the days in a year; these include long-lasting probable automated pages). Then we calculated daily top-k turnover (k = 10, 50, 100), entrants (first-time or returning within 365 days), stay lengths, and the median daily absolute log change of views. We fitted linear trends to 10 annual values per edition and pooled them with random effects. Results: mobile-web top-50 turnover decreased in 9 of 10 editions (p < 0.05 each; 4 after Holm correction). The pooled trend was -0.71 percentage points per year (95% CI -0.92 to -0.50). Japanese did not change (-0.01, CI -0.39 to +0.37). For the top 10 and top 50, no edition increased in any list variant. All-access lists gave weaker decreases. Returning entrants per day decreased from 17.6 to 14.7 (mean of editions; decrease in all 10 editions, significant in 9). First-time entrants did not decrease (11.6 to 12.3; pooled +0.09 per year, CI +0.04 to +0.14). The total days in the top 50 per article had no pooled trend, but stays were longer and more continuous. Daily view volatility of top-50 articles decreased in 7 editions and did not change in Japanese. Its trend agreed with the turnover trend across editions (Spearman 0.81, p = 0.005). In a fixed-effects model, it accounted for 77% of the mean turnover decrease. The decrease was largest in 2016-2020, and it continued after the April 2020 change in bot labels (-0.51 pp per year, CI -0.76 to -0.26). Limitations: top 1000 only, daily resolution, Wikimedia 'user' traffic, possible residual automated traffic (no app-consistency filter), descriptive decomposition, and no causal test.

collective attention · Wikipedia · pageviews · social acceleration · top lists · turnover · multilingual · attention dynamics · bot traffic · computational social science