<p>The mere act of providing a judgment of learning (JOL) influences subsequent memory performance. Such JOL reactivity depends on features of the learned material, as shown by numerous studies demonstrating that providing JOLs tends to enhance memory for related word pairs but not unrelated word pairs. Consequently, the mechanisms underlying existing accounts of positive reactivity predict the effect should generalize across semantically related word pairs broadly. In contrast, we propose that providing JOLs prompts elaboration, but JOL reactivity is conditional on the amount of test-relevant elaboration that cues afford. That is, positive reactivity for related word pairs is not due to their semantic relatedness per se, but rather, occurs because semantically related pairs tend to afford deeper elaboration. Across three preregistered experiments (<i>N</i> = 375), we test these competing mechanistic assumptions by examining whether positive reactivity arises for synonyms–a word pair type that is characterized by high semantic relatedness but only affords shallow elaboration on that relation due to high similarity. Despite being highly related, we find no JOL reactivity for synonym pairs, significantly differing from the positive reactivity observed for distinct-related word pairs. That is, JOL reactivity does not uniformly generalize across all semantically related word pairs. Further, supporting an afforded elaboration mechanism, we find that a novel, continuous metric of the amount of elaboration word pairs afford moderates JOL reactivity. These results emphasize the need for refinement of major accounts of JOL reactivity and offer a pathway for transferring positive JOL reactivity to real-world learning.</p>