Why is near_text differs from near_vector results for the same encoded text?

I have noticed that if I were to encode outside of weaviate and then perform a near vector search on a class encoded by the same model. I can get vastly different results than if I just preform a near text search on the class using the text. I am wondering why this is?

Are you passing more then one concept to the nearText query? Can you give an example? and which vectorizer are you using?