Although Problem-Based Learning (PBL) has been implemented to foster critical thinking, evidence regarding the consistency of its effectiveness across instructional formats, subject domains, and geographical contexts in Indonesia remains fragmented. This meta-analysis quantitatively synthesized the effects of PBL on students’ critical thinking while examining implementation type, subject cluster, and geographical region as moderators. A systematic search of Google Scholar was conducted in May–June 2026 for peer-reviewed studies published between 2019 and 2026. Following PRISMA 2020 procedures, 100 records were identified and 22 studies involving 1,513 participants were included. Effect sizes were calculated using Hedges’ g and synthesized with a random-effects model in Comprehensive Meta-Analysis Version 3. The pooled effect was large and statistically significant (g = 1.487, 95% CI [1.018, 1.955], p < .001). However, heterogeneity was very high (I² = 92.90%), and the prediction interval crossed zero (−0.616 to 3.589), indicating contextual variation. Subject cluster and geography did not significantly moderate the effect, suggesting that the observed benefits were not systematically confined to particular disciplines or regions. In contrast, implementation type was a significant moderator, with Integrated PBL producing a larger pooled estimate than Pure PBL. The Assisted PBL estimate was imprecise, non-significant, and based on only two studies, precluding claims of superiority. The funnel plot appeared approximately symmetrical, although visual inspection alone could not rule out publication bias. Overall, implementation quality, scaffolding, facilitation, and contextual alignment appear more decisive than subject area or geography in shaping PBL effectiveness.