If I understand your question correctly, you have a couple of options:
You can specify the utterance used in the collect step with the utter parameter, see more information here.
You can specify a custom action to generate the collect step utterance. This allows fully dynamic prompting in collect steps. More on this option here.