We’re developing a RAG engine and aim for the responses to be as natural as possible. However, the responses always start with “Based on the provided context…”. After numerous trials and errors with system and user prompts, the outcome remained unchanged.
Have you experienced something similar and managed to resolve it?
Happy for every suggestion 
This question was previously asked but was closed without a resolution:
based-on-the-information-provided/377159
I think it has to do more with how you are injecting the knowledge.
I would try it with an assistant role. Then give a prefix “reminder to self: this is knowledge I previously retrieved to help answer the latest question, but I won’t mention these automated mechanisms exist when answering questions” within the role that adds them. You can also try positioning, either before or after the history of past chats, and see which is less mentioned and still used.
Can you share some prompts ? Will be easier to provide some feedback