NemoClaw Knowledge Wiki

Optimizing AI for Legal Work Custom Instructions for Professional Output

3 min read

Clip title: How to Set Up ChatGPT, Claude & Gemini for Legal Work Author / channel: Liam Barnes URL: https://www.youtube.com/watch?v=BP6x_FRwZ3w

Summary

This video introduces a five-minute setup process that significantly improves the quality of AI output for legal professionals, moving it from “mediocre” to “partner-level work product.” The core problem identified is that general AI platforms (like ChatGPT, Claude, Gemini, Perplexity, Copilot) are designed to be safe and generic. They lack context about the user’s professional role, communication style, jurisdiction, and specific preferences, leading to hedged answers, disclaimers, and the need for constant re-explanation. The presenter demonstrates how custom instructions can rectify this across various AI tools.

The solution involves configuring two layers of instructions: Global Instructions and Project Instructions. Global Instructions, which apply to every conversation, define “Who You Are” and encompass four key elements: your role (e.g., a litigation partner at a mid-size firm in Texas, handling commercial and employment matters), your preferred communication style (e.g., direct, conclusion-first, plain language, no legalese), specific behavior rules (e.g., never telling the user to consult a lawyer, citing specific statutes and rule numbers, flagging jurisdiction differences), and what the AI should not do (e.g., no generic opening phrases, no hedging with “it depends” without explanation, no legal advice disclaimers). Implementing these greatly reduces generic output.

The second layer, Project Instructions, focuses on “What You’re Working On.” This involves creating dedicated workspaces for specific clients or matters where relevant documents can be uploaded and engagement-specific instructions can be set. This ensures the AI has full context for that particular project without needing repetitive uploads. The presenter cautions that for new sessions within a project, it’s often wise to explicitly prompt the AI to explore the uploaded documents. Regarding data privacy, the presenter notes that custom global instructions do not contain client data and are safe, much like a LinkedIn profile. For actual client work, he advises using paid, business-tier AI platforms (like ChatGPT Team or Claude Pro/Business/Enterprise) that do not train on user data, and always using anonymized scenarios for any sensitive or confidential information.

In conclusion, by spending just a few minutes configuring these global and project-level custom instructions across their preferred AI platforms, legal professionals can dramatically enhance the relevance, directness, and quality of AI-generated content. This systematic approach eliminates approximately 90% of the repetitive context-setting typically required, enabling lawyers to receive tailored, actionable responses that are aligned with their professional standards and specific needs. The video ends by offering a free template at LegalAIROI.com to help lawyers set up these custom instructions efficiently for various practice areas.

Graph View