Background <p>Stroboscopic visual training (SVT) intermittently occludes vision during skill execution to increase perceptual‑cognitive demand and improve information pickup. Evidence in sport is heterogeneous, and key exposure parameters are inconsistently reported; therefore, we quantified SVT effects and explored whether a duty‑cycle‑adjusted exposure metric relates to performance change.</p> Methods <p>PubMed, Web of Science, Scopus, SPORTDiscus, Embase and Cochrane CENTRAL were searched from inception to 19 August 2025 for controlled studies comparing SVT with non‑stroboscopic controls in athletes or physically active participants. Outcomes were grouped as perceptual‑cognitive, visuomotor reaction time, motor speed/reactive agility, or visual tracking. Random-effects meta-analysis (REML) computed pooled Hedges’ g. Because some studies contributed more than one comparison-level effect size, a multilevel random-effects model with comparisons nested within studies was fitted as a robustness check, and variance partitioning was used to describe the relative contribution of within-study versus between-study heterogeneity. Analyses were stratified by exposure timescale (acute single‑session vs multi‑session programmes). Exposure–response modelling used occlusion‑minutes (training minutes × opaque duty‑cycle fraction) when dose data were available.</p> Results <p>Seventeen studies (513 participants; 22 comparison-level effect sizes) were included. Overall, SVT improved sport-related outcomes (g = 0.79, 95% CI 0.39–1.19; I² = 84.3%; 95% prediction interval − 1.02 to 2.60). A multilevel random-effects model yielded a similar pooled estimate (g = 0.84, 95% CI 0.33–1.34; <i>p</i> = 0.001) and suggested that heterogeneity was predominantly between studies. Benefits were evident for multi-session programmes (15 studies; k = 20; g = 0.88, 95% CI 0.45–1.32) but not for acute exposures (2 studies; k = 2; g = 0.06, 95% CI − 0.26 to 0.38). Pooled effects were largest for perceptual-cognitive tasks (3 studies; k = 3; g = 1.23) and visuomotor reaction time (11 studies; k = 12; g = 0.99), and smaller for motor speed / reactive agility (6 studies; k = 6; g = 0.35). In the multi-session reaction-time subset with dose data (5 studies; k = 5; occlusion-minutes 94.9–325.2), occlusion-minutes were not associated with effect size (linear meta-regression slope = − 0.0016&#xa0;g per occlusion-minute, 95% CI − 0.0058 to 0.0025; <i>p</i> = 0.503). Publication bias analyses did not suggest small-study effects (Egger <i>p</i> = 0.289; trim-and-fill: 0 studies imputed).</p> Conclusions <p>SVT shows a moderate-to-large average benefit across heterogeneous designs, driven primarily by multi-session programmes. Evidence is stronger for near-to-intermediate transfer outcomes than for acute effects or far transfer to representative sport performance. Current data are insufficient to define an optimal SVT dose based on occlusion-minutes alone. Future trials should standardise device reporting, use conceptually grounded transfer outcomes, and include delayed follow-up assessments.</p> Trial registration <p>PROSPERO CRD420251090541.</p>