If you use a machine learning policy (usually TED) in your configuration, then Rasa will always try to generalise from the stories in your training data. The more stories you have, the better that generalisation will be. If you don’t want that, then only use the Memoization policy.
See this related thread for more information: How do stories work?