My custom GPT keeps misinterpreting instructions – what advanced prompt engineering am I missing?

I've built a custom GPT for internal documentation, but despite extensive fine-tuning of the custom instructions, it frequently deviates from my specific formatting and persona requirements. I've tried rephrasing prompts and adding negative constraints, but it still struggles with consistency. Is there a more advanced prompt engineering technique or a common pitfall I might be overlooking to ensure strict adherence? Really need some expert insights to get this working reliably.