Legal Prompting Guide

The Observation

Legal professionals are increasingly using generative AI, often without understanding mechanisms or measuring risks. Before delegating reasoning to a machine, it is essential to understand how it works, where it excels, and where it fails.

Part I — Understanding the Risks

LLMs are statistical machines predicting the most probable next word. Five cognitive biases every jurist must know. Documented cases:

• Mata v. Avianca (2023)
• TA Orléans/Grenoble decisions (December 2025) confirming the phenomenon crossed the Atlantic

Part II — Ethics, Deontology and Confidentiality

• Professional secrecy vs. LLM
• Anonymization limits
• Operational legend technique
• Imperative rules for practitioners

Part III — From Bad Prompt to Good

Four pillars:

• Context
• Objective
• Constraints
• Format

Advanced techniques:

• Role prompting
• Few-shot learning
• Chain-of-thought
• Negative prompting
• Meta-prompt: a prompt that generates other prompts

Criminal law comparative examples included throughout.

The Audience

Over 20,000 professionals reached in 4 days on LinkedIn. Nearly half from Ministry of Justice, ENM, or Paris Bar. This is not a technical guide. An attempt to ask the right questions before everyone rushes to wrong answers.