Immediate judgments of learning (JOLs), predictions of one’s future memory performance during learning, are often used to investigate how people monitor their learning and memory. Previous research on immediate JOL reactivity shows that soliciting immediate JOLs can change memory. It is not well-understood, however, under which conditions making JOLs improves, impairs, or does not affect memory, and which processes contribute to JOL reactivity. The present study provides a comprehensive meta-analysis of immediate JOL reactivity based on 344 effect sizes from 175 experiments reported in 49 records with a sample of 15,079 adults. Overall, making JOLs improved memory performance (g = 0.22, 95% CI [0.17, 0.27]). While the subset of effect sizes based on related and unrelated word pairs yielded virtually identical results overall (g = 0.23), examining related and unrelated pairs separately revealed that JOL reactivity differed considerably depending on relatedness: Related pairs revealed positive reactivity (g = 0.44), whereas unrelated pairs revealed negative reactivity (g = –0.09). Moderator analyses indicated that the size and direction of JOL reactivity varied depending on experimental characteristics, such as memory tests, manipulating making JOLs within or between participants, or the language in which experiments were conducted. Further, results revealed a strong publication bias favoring positive reactivity over negative reactivity. This study confirms that making JOLs can alter memory but also highlights that JOL reactivity is diverse in size and direction. New theories are needed for explaining when and why making JOLs has positive or negative effects on memory performance.