What actually changes?
Prompting changes instructions and examples in the context. RAG retrieves reference material. Fine-tuning changes model parameters using a training dataset. These methods can be combined rather than treated as competing products.
Teach a format, retrieve a fact
Use instructions for the response format, retrieval for current product policies, and consider fine-tuning when many representative examples are needed to teach a consistent task.
LoRA trains smaller adapter matrices while keeping the base weights frozen. It reduces the number of trainable parameters; it does not make dataset quality unimportant.
Training needs evaluation
Fine-tuning can overfit, introduce unwanted behaviour or weaken other capabilities. Keep evaluation examples separate from training examples. Changing weights is not a dependable substitute for looking up frequently changing facts. Compare the adapted system with a simpler prompting baseline before accepting the extra maintenance.