Corrective feedback on perceptual decisions is essential for refining decision-making strategies, but its interaction with other decision components, like previous stimuli and choices, complicates understanding its influence on subsequent decisions. Traditionally, "reinforcement learning" has been applied to this problem, assuming that individuals treat corrective feedback as rewards, similar to reward-based decision-making. This approach posits that feedback helps individuals learn the value of their choices. However, we propose an alternative hypothesis: humans interpret corrective feedback as evidence about the actual state of the world, not as rewards for their choices. To test this, we implemented both the "feedback-as-reward" and "feedback-as-evidence" models on a unified learning platform. Our findings show that the "feedback-as-evidence" model better explains how feedback, along with past stimuli and choices, shapes decision-making strategies. This suggests that during perceptual decision-making, humans use corrective feedback to learn about their environment rather than merely evaluating the worth of their past choices. This perspective shifts the understanding of how feedback influences learning and decision-making, emphasizing an evidence-based approach over a reward-based one.