Scholars highlight the importance of Artificial Intelligences’ (AI) performance for individuals’ willingness to work with AIs. However, research has not yet considered the effect of historical AI performance relative to the user’s performance in an unrelated task. We tested this antecedent with 797 participants in an experiment with a 2 (human or AI as counterpart) x 2 (counterpart with superior or similar performance compared to the participant in an unrelated task) between-subjects design. We find that, firstly, participants preferred to collaborate with an AI compared to a human, independent of the counterpart’s relative performance. Secondly, superior AI performance, in an unrelated task, affected participants’ trust in and willingness to collaborate with this AI negatively. Our work advances theory by adding a novel effect of AI performance on trust and willingness to work with AIs. Further, our findings warn organizations of unintended adverse effects when promoting multi-functional AIs.