Bench & Bar January/February 2025

EFFECTIVE LEGAL WRITING

PROMPTING with PURPOSE: HOW LAWYERS CAN GUIDE G en AI THROUGH RHETORICAL WARENESS BY SUSAN TANNER ASSISTANT PROFESSOR OF LAW

I know what you’re thinking - oh great, another article about Generative AI (GenAI). As someone who travels the country giving talks on GenAI in legal practice, I understand the fatigue. We’re all a little burnt out on the GenAI hype train. But I hope to provide a bit of a different perspective: I spent seven years getting my Ph.D. at Carnegie Mellon University using computational linguistics and machine learning to study patterns in legal language. Now, I watch ChatGPT do in an hour what it would have taken me months to do. That’s both humbling and illuminating. But this rather unique background has given me a different take on why some lawyers love these AI tools while others find them deeply frustrating. Here’s what I’ve noticed when I talk to prac titioners about GenAI: most skepticism about GenAI in legal practice comes from misunderstanding what these tools are actu ally good at. I watch brilliant lawyers grow frustrated when AI fails at legal research or complex reasoning - but that’s like being angry at a hammer for not being a good

WHERE G en AI REALLY SHINES: ESTABLISHED GENRES Legal writing, like all writing, can be orga nized into distinct genres - categories of documents that share common purposes, structures, and conventions. A genre isn’t just a format; it’s a recognized way of com municating within the legal community. When we say GenAI excels at understand ing genres, we mean it can recognize and reproduce these established patterns of professional communication. Letters offer a good example of how GenAI understands and applies genre conventions. Take demand letters: their effectiveness comes from following a clear, established structure that the legal community rec ognizes and respects. GenAI has learned this structure through exposure to count less examples. The progression is logical: you establish representation, describe the problem, outline legal basis, make specific demands, set timelines, and explain conse quences. What makes GenAI particularly useful here is that it doesn’t just know this structure - it can adapt it to different

screwdriver. The key insight from my years of studying language patterns is this: Large Language Models aren’t research or analysis machines - they’re writing machines. HOW G en AI REALLY “UNDERSTANDS” LANGUAGE Let’s be clear about what’s happening under the hood. When you’re working with an GenAI writing assistant, you’re not work ing with a legal mind. You’re working with a sophisticated pattern-matching system that has seen millions of legal documents. These systems learn by analyzing vast collections of text, identifying patterns in how differ ent types of documents are structured, what language they use, and how they flow. Think about how you learned to write legal documents. You probably studied examples, noticed patterns, and gradually internalized the conventions of different document types. AI does something similar, but at a massive scale. It’s seen millions of examples of each type of legal document, learning the patterns that make a motion persuasive or a contract clear. In other words, GenAI is a genre machine.

30 january/february 2025