one of the first things that most LLM chatbot writers get wrong is not prefixing their instructions with clear instructions to not talk like a typical chatbot.

it's possible to have a preamble on all requests that specify that you want it to, for example, always type in lower case except for names, sometimes, and acronyms, sometimes, and write like a salty grumpy intellectual and say things that may or may not be verifiable.

then you could have the chatbot look like me.