Just like in human infants, ostensive verbal utterances can transform human actions into a natural teaching scenario for dogs. However, functional selection created ‘independent’ and ‘cooperative’ dog breeds with different dependence on human signals. We hypothesize that this could affect dogs’ sensitivity towards verbal communication. We tested independent and cooperative breeds in the two-choice ‘A-not-B paradigm’. The experimenter used either ostensive or neutral intonation speech while hiding the target. Based on the target’s position, the trial order was A-A-B-B-A. Perseverative ‘A-not-B’ errors in Trial 3 are interpreted as learning the rule to look for the reward at location ‘A’. From the near 100% success rate in Trials 1 and 2, each groups’ performance dropped to chance level in Trial 3, except for cooperative dogs in the neutral speech condition. Independent dogs in the neutral speech condition paid the least attention to the experimenter. We conclude that perseverative errors can be either the consequence of rule-learning elicited by ostensive intonation or reverting to the ‘win–stay’ strategy, when independent dogs lost interest in watching where the experimenter exactly hid the reward. Functional selection could influence dogs’ general attentiveness towards human communication; thus, neutral speech may have an underestimated relevance for cooperative dogs.