The ephemeral reward task consists of giving an animal a choice between two distinctive stimuli, A and B (e.g., black and white), on each of which is placed a bit of food. If the animal chooses the food on A, it gets that reinforcer, but the other stimulus, B, is removed, and the trial is over. If it chooses the food on B, however, it gets that food and the stimulus A remains, so it can have that food as well. Thus, choice of stimulus B gives the animal two reinforcers rather than one. Wrasse (cleaner fish) easily learn to choose optimally, whereas surprisingly, most non-human primates do not. Parrots, however, appear to learn this task as easily as the fish. To test the hypothesis that animals that choose with their mouth can learn it, we tested pigeons and found that they show no evidence of optimal learning with this task (with either the manual presentation of the stimuli or the operant presentation of the stimuli). Similarly, rats show no evidence of optimal learning. However, if a 20-s delay (fixed-interval schedule) is inserted between stimulus choice and reinforcement, both pigeons and rats learn to perform optimally. The ephemeral reward task appears to promote impulsive choice in several species, but with this task (and others), inserting a delay between choice and reinforcement promotes more careful choice, leading to optimal performance.