The video below shows how installing standing instructions reduces an AI model’s inclination to make things up. I also provide a handy starter set of custom instructions for lawyers. First, though, if you want to learn advanced practical AI-assisted workflows for attorneys, sign up for my AI Workflows for Lawyers presentation:
AI Workflows for Lawyers (Presentation)
I can show you legal-AI workflows you haven’t seen before — ones requiring no software beyond a subscription to Claude or ChatGPT.
I use AI to run detailed cite-checks against authoritative sources, exposing weak support so it can be replaced or attacked; map evidence to pleadings and transcripts, checking for inconsistencies and gaps; and edit a document by talking about it. Join me for a live Zoom presentation and Q&A. You’ll leave with a handout of tips, custom instructions, and prompts.
September 17 ($99 pilot-presentation price) and October 15 ($149 regular price) · Noon–1 p.m. Pacific
I’m a practicing litigator who can code. I spent two years researching AI on the UCLA Law faculty, and I track AI developments every day.
Custom Instructions for Lawyers
Custom instructions are standing instructions that an AI model applies to every conversation you have with it and every task it does for you. You can think of them as an invisible part of every prompt that you submit.
The video below is the first in my AI for Lawyers series. It shows you how to adjust your settings and install custom instructions that make it more likely that an AI model will follow your preferences. Strong custom instructions translate into much better results from ChatGPT, Claude, Gemini, and other LLMs. I provide a starter set of custom instructions below the video.
You can download my starter set of custom instructions here.
The video begins with a deliberately bad prompt: “Write a concise brief opposing a motion to dismiss a fraud complaint based on misleading terms in an apartment lease.” I did not tell the model the jurisdiction, which terms were misleading, which type of fraud applied, any facts of the case, or the applicable legal authority. That did not stop it from writing a brief.
If you ask an AI model to write a brief without providing facts or authority that could support the brief, it will often try to follow the instruction anyway. That raises the risk that it will fill the gaps with hallucinations, assumptions, or platitudes. The first step to improving its output is to put custom instructions in place.
In the video, I walk through a starter set that tells the model:
to act as an expert litigation attorney and legal editor for litigators;
to put accuracy and precision before grammar, logic, clarity, and style;
never to add substance absent from supplied materials;
never to change, fix, or reformat a citation or quotation unless asked;
to use supplied materials unless outside research is specifically requested or plainly required;
to say, “I can’t verify that from the provided materials,” when support is missing;
to assign confidence levels to its conclusions;
not to use identifying information without permission; and
to answer only what was asked, preserve substance and style when editing, and flag apparent errors instead of silently rewriting them.
When I run the same bad prompt again, the response improves. Instead of inventing the lease terms and relevant facts, the model creates placeholders and asks for the state, court, specific lease language, and legal basis.
The result is still a terrible brief. AI models write stale prose shot through with banality and cliché. Lawyers should not write their briefs with AI or try to pass off AI-generated text as their own. But as the video shows, my starter set of custom instructions immediately reduced the model’s inclination to make things up and run on assumptions.
Custom instructions make it more likely that an AI model will follow your preferences, but they cannot guarantee it. Follow my ground rules for AI use. Think critically and skeptically about AI output, write your own briefs, and exercise professional judgment.
For more advanced AI workflows, register for one of my presentations and subscribe to my YouTube channel. I’ll continue updating my AI for Lawyers series with demonstration videos.
Elegant Legal Writing is available on Amazon and Audible. Please help me sustain the book’s momentum by rating it five stars and sharing my posts.
Ryan McCarl is a partner of the business litigation firm Rushing McCarl LLP and author of Elegant Legal Writing (Univ. Cal. Press 2024). For more tips about legal writing and argumentation, subscribe to the Elegant Legal Writing blog and follow Ryan on LinkedIn. McCarl’s book is available on Amazon and Audible.
Please share this post with your networks:
Connect with Ryan McCarl: Order the Elegant Legal Writing book | LinkedIn | Rushing McCarl LLP | Linktree | Second Stage (blog) | Bluesky | Substack | Twitter (X) | Threads