Learning how to make decisions from experience is often studied using probabilistic choice tasks, as in reward learning or risky gambles (e.g., option A gives a reward with a probability of 75%, option B with 25%, to-be-learned in repeated trial-and-feedback trials). An observed phenomenon in such tasks is negative or wavy recency. This entails that learners expect that the likelihood of encountering a rare event (reward for option B) increases with the number of consecutively observed common events (reward for option A). In classical conditioning, it has been speculated that negative recency is moderated by the magnitude of the reward associated with the rare event. To test this hypothesis in the context of decisions from experience, we meta-analytically re-analyzed 90 studies on lottery decisions (risky gambles: safe vs. risky option). In these, participants received descriptions of the options’ payoffs and probabilities (e.g., safe option: 15 points vs. risky option: [25% 50 points or 75% 5 points]). In the first trials, they made decisions only based on the descriptions (DE) without receiving feedback. Subsequently, they received full feedback (FF), i.e., gained experience. In DE trials, participants seemed to overweight rare events in their decisions, which was moderated by whether the risky option was associated with a rare loss (increased safe choice) compared to a rare win. Crucially, this effect completely vanished in FF trials (i.e., DE-FF-gap). In FF trials, negative recency was strongly evident and moderated by reward magnitude. This speaks for feedback-based experience altering cognitive processing, possibly mitigating risk preferences.