This paper describes SmartTur+ECO, a conversational recommender system for tourist attractions. It extracts the basic trip data and elicits user preferences through a multi-round interaction with the user guided by a chatbot. In our case, the main advantages of using a conversational recommender system are that we are facing a user-cold start problem because no user history is available, and regarding tourism, it is very likely for most of the users to be a first-time visit to the destination. The preference elicitation process is based on an exploit-explore approach that leverages the similarity between POIs, represented as sentence embeddings generated from the POI description. Finally, the most suitable attractions are recommended and organized in a multi-day plan that considers opening hours and the location of these attractions.