Sure! One way would be to define one utterance, utter_address say, that outputs the XYZ, and another, e.g. utter_ask_web that asks Do you need the web address?. Then you write your stories in the following way:
Thank you!! For small data it will fine for doing it in each and every story but I have a large data where I wanted bot to recommend relevant question based on previous question?? (Example: Recommendation system in Amazon which shows similar product Hope you got it!!)
When you scale your bot to many stories, you typically collect those stories with Rasa X by letting the bot talk to users (or test subjects). If you have selected a machine learning policy (e.g. EmbeddingPolicy, which is the default I think) for Rasa Core, then it should quite often get it right even in new situations, so most of the time you can just store good examples and then use Rasa X to correct your bot for the bad examples. In this way you can quickly create lots of stories that contain the correct recommendation actions.
If your recommendation only depends on slots set and the last user turn, then you could also implement a custom action instead. This approach is less flexible, however.
You can implement anything in custom actions (see Actions), but we don’t have any special components for a recommender system. You might also be interested in knowledge bases (see Knowledge Base Actions) and the response selector (see Retrieval Actions).