How to use LLMs like LLAMA-2 for NER tasks?

I want to extract countries and organizations from texts and wonder if there are best practices how to do that?

Would you just define a prompt like “Provide a list of all countries mentioned in the following text: …” ?

LLAMA-2 for instance provides good answers but is not very consistent and sometimes adds additional text. Are there any better approaches to use LLAMA-2 for NER tasks to get more reliable and structured results?