We report 17 experiments in which participants predicted versus did not predict their future memory during learning (judgments of learning, JOLs). In all experiments, making JOLs increased the difference in cued-recall performance between related and unrelated word pairs. Making JOLs produced numerically better memory for related pairs (positive reactivity) and numerically worse memory for unrelated pairs (negative reactivity) in almost all experiments, but either effect was reliable in just half of the experiments. Small-scale meta-analyses revealed that both positive and negative reactivity were of small-to-moderate size. Individual experiments that showed positive reactivity tended not to show negative reactivity, and vice versa. These findings indicate that negative JOL reactivity for unrelated pairs is similarly large and robust as positive reactivity for related pairs. They favor the cue-strengthening hypothesis with dual-task costs over other theories and raise the practically relevant possibility that monitoring could have detrimental effects on learning in educational settings.