Making the LLM node in Flowise more flexible when using RAG?

[paintbot]

How can I make the LLM node in Flowise more flexible when using RAG? No matter what I write in the system prompt, it seems to have little effect on the chain's behavior. For example, if I instruct the second LLM node (SelmaReturn) to ignore the user's question and instead summarize the retrieved documents, it still answers the user's question instead of following my instructions. Is there a way to ensure the system prompt has more influence or to work around this issue?