In metamemory research, it is commonly assumed that learners spontaneously monitor their learning progress. To assess monitoring, participants are asked to provide judgments of learning (JOLs) as they memorize information, usually on a percentage scale. Recent evidence that making JOLs has a reactive effect on memory performance, however, suggests that the processes involved in spontaneous monitoring are different from those that are involved when participants are requested to make and report JOLs. The present research examined two hypotheses regarding the difference between spontaneous monitoring and making JOLs. According to the overt-response hypothesis, JOL reactivity emerges from the requirement to report the JOLs that are spontaneously done covertly. To test this hypothesis, participants studied related and unrelated word pairs and made JOLs either overtly or covertly or did not make any JOLs. According to the numerical-response hypothesis, JOL reactivity emerges from the requirement to translate qualitative, non-numerical spontaneous assessments of learning into numerical JOLs. To test this hypothesis, participants made overt JOLs on a percentage scale, a Likert scale, a yes/no scale, or did not make JOLs. The results yielded JOL reactivity regardless of the response format or the response scale except when participants were under-engaged in making covert JOLs. These findings are inconsistent with both hypotheses and allude instead to the possibility that JOL reactivity emerges when monitoring does not spontaneously occur and is initiated by the JOL prompt.