The semantic relatedness effect, a memory advantage of semantically related items (e.g., “penguin, giraffe, goat”), is well established in the literature on working memory (WM). Nevertheless, it remains unclear what mechanisms are responsible for this effect. Although an influential account ascribes it to the cue-dependent retrieval process (e.g., “animal” works as a cue for “penguin, giraffe, goat”), this account has not yet been fully investigated. This is partly because the influence of cues cannot be directly tested in typical studies using common categories (e.g., “animal” is likely to be generated and used by participants, but the generation and use of cues are uncontrollable for the experimenter). The present study, by introducing ad hoc categories and cueing ad hoc category labels, directly tested the influence of cues. Specifically, seemingly unrelated items (e.g., “bone, fly, car”) were presented with or without the corresponding ad hoc category label (e.g., “things that dogs chase”). Four experiments demonstrated that providing ad hoc category labels affected WM performance. Importantly, providing the labels improved item memory in WM (Experiments 2 and 3). This supported the retrieval-cue account. Nevertheless, the effect was small (Experiments 2 and 3) and was not found in an experiment (Experiment 1). In contrast, providing the labels had a substantial and systematic effect on long-term memory, suggesting that the manipulation of providing the labels, per se, was successful. The current study’s implications for research on WM and ad hoc categories were also discussed.