Working memory (WM) is a cognitive system that temporarily maintains information needed for reasoning, learning, and comprehension, even under distraction. Because WM has limited capacity, interference-based models such as the SOB-CS model propose that forgetting occurs when new information—like automatically processed distractors—overwrites existing representations. To explain why distraction does not fully erase WM contents, SOB-CS assumes that distractors, although automatically encoded, are then actively removed from the short-term store. This removal process is time-dependent: with more free time available, removal becomes more efficient. Thus, longer intervals after distraction should improve memory when interference is harmful. However, when interference strengthens representations, more opportunities for removal should reduce distraction-related benefits. Our experiments tested whether the previously observed benefit of semantically related distraction—found in both short-term and long-term memory—would diminish when more time for distraction removal is available. In two experiments using a complex-span task, we manipulated (a) whether distractors were semantically related to the targets (related vs. unrelated) and (b) the duration of the unfilled interval following each distractor (100 vs. 1400 ms). In Experiment 1, related distractors were exemplars from the same broad semantic category. In Experiment 2, they were related through non-categorical semantic associations. We predicted that semantically related distractors would enhance performance, reflecting beneficial interference, but that this benefit would be smaller with longer post-distractor intervals, due to distraction being more successfully removed. Our results help determine whether distraction is automatically removed from WM, and with it not only its costs but also its benefits.