Recent research revealed that predicting one’s memory performance during learning (judgments of learning, JOLs) can alter memory performance (JOL reactivity). In particular, eliciting JOLs is found to enhance cued-recall performance for related word pairs (i.e., positive reactivity) and/or impair recall performance for unrelated word pairs (i.e., negative reactivity). It has been proposed that positive reactivity might occur because participants who make JOLs use more effective learning strategies than participants who do make JOLs. In two preregistered experiments (N = 130 undergraduates each), we examined (1) whether participants in a JOL group and a no-JOL control group differ in reported learning strategies and (2) whether these differences underlie JOL reactivity. In Experiment 1, participants studied weakly related and unrelated word pairs and reported their learning strategies for each pair after completing the cued-recall test. Experiment 2 conceptually replicated Experiment 1 by using a list of strongly related and unrelated word pairs and integrating the assessment of learning strategies in the cued-recall test. Both experiments revealed positive reactivity for related word pairs and negative reactivity for unrelated ones. Further, participants in the JOL group used less mental imagery and more retrieval practice for studying unrelated word pairs than participants in the control group. These differences in learning strategies partially accounted for negative reactivity for unrelated word pairs. Taken together, our findings indicate that altered learning strategies in the JOL group contribute to negative reactivity, but not to positive reactivity.