Algorithmic visibility of child-directed content: Exposure and persistence dynamics in YouTube trending lists across 113 countries

Abstract

The algorithmic curation of content on digital platforms raises important questions about which content categories achieve systematic visibility and under what structural conditions. This study examines explicitly child-directed and child-centered content in YouTube trending lists across 113 countries. Drawing on 5,505,720 trending-video observations collected between 26 October 2023 and 5 July 2026 and aggregated into 110,847 country-day units, the analysis uses a conservative lower-bound metadata classifier and two-way fixed-effects panel models. A 256-video simple random validation yielded TP = 1, FP = 0, FN = 1, and TN = 254 (precision = 1.00, recall = 0.50, F1 = 0.67, classifier-manual κ = 0.665), but these estimates are imprecise because only two manually positive cases occurred. An additional blinded stratified challenge set of 196 videos yielded 37 true positives, 9 false positives, 15 false negatives, and 135 true negatives; positive confirmation among classifier-positive cases was 80.4%. The final classifier identified 7707 records (0.14%). Child-directed share showed strong aggregate autoregressive persistence (β = 0.768, p < .001); this coefficient concerns persistence in the country-level share, not retention of the same videos. Exposure scale was positively associated with child-directed share, whereas approval and conversational rates provided no robust positive support. PPML and sensitivity specifications preserved the central persistence and exposure findings.