I didn't notice that the prompt wording implies impersonation, which is incorrect for common usage. Changed the order of tags used to generate the character's response to the user, rather than the vice versa.
Tested on Llama 2 13B chat model provided by TheBloke. Based on these prompt example: https://huggingface.co/TheBloke/Llama-2-13B-chat-GPTQ#prompt-template-llama-2-chat