People play an active role in much of their everyday learning, making sampling decisions about the stimuli they choose to learn about. Recent work incorporating genuine autonomy into an evaluative conditioning paradigm demonstrated that these sampling decisions not only shape the information available, but also provide evaluative information themselves: Sampling a stimulus more often predicts increased liking of the stimulus, regardless of whether it is consistently paired with positive or negative images. The current experiments employed an instruction-based version of the paradigm, asking participants to imagine that they chose to interact a variable number of times with stimuli that were consistently paired with positive or negative images. Mirroring the effects in the original paradigm, we found that the effect of (imagined) paired valence replicated in this version of the paradigm, such that stimuli stated to have been paired with positive versus negative images were evaluated more positively. In contrast, we did not find a similar effect of sampling frequency on evaluations. That is, people did not reliably exhibit positive shifts in evaluations for those stimuli they were asked to imagine choosing more frequently. Nevertheless, follow-up analyses revealed that people recalled a higher number of interactions for those stimuli they showed a more positive evaluative shift toward. Together, these results suggests that while people tend to infer a connection between their attitude toward a stimulus and how frequently they chose to interact with it, genuine autonomy in those decisions plays a crucial role in their positive impact on evaluation.