Towards a Legal Prompt Engineering Strategy for Identifying Rationes Decidendi

Large language models (‘LLMs’) have become increasingly prevalent in legal practice and scholarship. However, their utility for case law analysis, has remained limited. We propose a legal prompt engineering strategy for using LLMs, specifically OpenAI’s Generative Pre-Trained Transformer 4 (‘GPT-4’), to identify case law rationes decidendi. The strategy draws on Branting’s computational model of ratio decidendi which we translated into a series of prompts. The strategy distinguishes two conceptions of ratio: a narrow view focussed on material facts and a broad view centred on legal principles. Our strategy incorporates targeted prompt engineering techniques improving GPT-4’s ability to step through a multi-stage reasoning process. We evaluate the strategy using the landmark case of Project Blue Sky Inc v Australian Broadcasting Authority, which was chosen for its potentially divergent ratio interpretations. Results indicate significant improvement over a baseline approach, particularly in identifying the narrow ratio. Limitations remain in identifying the broad ratio due to varied legal opinions on the appropriate scope and application of legal principles. Our study suggests that while legal prompt engineering can substantially enhance LLMs’ capabilities in legal analysis, fundamental differences between LLM reasoning and traditional legal interpretation persist.