The rise of generative artificial intelligence, especially large language models (LLMs), is subversively changing the legal industry. This paper explores how prompt engineering can be leveraged to control LLMs for advanced legal knowledge generation, moving beyond simple information retrieval to the creation of new legal insights. It discusses how LLMs, through cross-domain knowledge integration and counterfactual reasoning, can generate novel legal arguments. The paper introduces various prompt engineering techniques, from basic commands to advanced, structured instructions embedding legal frameworks like the “legal syllogism”. A case study on contract risk analysis demonstrates how the quality of prompts directly impacts the quality of LLM-generated output. Finally, the paper addresses the significant challenges and ethical dilemmas, including model “hallucinations”, inherent data bias, and professional responsibility in the age of AI. It advocates for a human-centric, collaborative model where legal professionals act as responsible operators, guiding AI to enhance legal practice while upholding fairness and justice.