Bench & Bar July/August 2025

LAW PRACTICE MANAGEMENT

NON-CLIENT-SPECIFIC USING AI FOR RESEARCH BY JEFF SALLEE

I t can be difficult to keep up with any technology, but the changes in AI are fre quent and substantial. This technology is growing much faster than we anticipated. New versions and capabilities are deployed a few times a year. Entirely new systems are also introduced frequently. The Artificial Intelligence Task Force has been focused on the ethical use of AI in the legal profession. We have focused particularly on the need to address confidentiality and the ethical pitfalls of using AI improperly. The pur pose of this article is to demonstrate how ChatGPT can be used in your law practice as a research tool without being tied to spe cific case details. One of the core changes in AI over the past year is the introduction of LLMs that can perform searches on the Internet to respond to prompts. We are no longer limited to the data used to train the current version of the AI model. The public-facing LLMs con tained that could have been uploaded a year or longer before the AI system went public.

EXAMPLE OF NON-CLIENT SPECIFIC RESEARCH In April, I was asked to determine the cri teria to be used to decide if an electronic signature company was secure and rep utable. I’ve been monitoring electronic signature vendors even before the ESIGN Act 1 was enacted. Back then, the primary players were DocuSign and Adobe’s Ado beSign. This is not the article to compare and contrast the kinds of signatures. I will focus on the AI aspect of this task. I could have dedicated several hours to refresh my knowledge of the current vendor and their offerings. I would have researched each of these vendors on the web and documented the data that met the criteria established by our vendor selected committee. I would then have researched another source to see if there were known Information Security vulnerabilities or incidents related to these vendors. What I did instead was to open ChatGPT and ask the AI to create the deci sion criteria and a table listing the top 25 vendors and include all the data I would have searched manually. The AI delivered the results in about a minute. The only revision I made was to limit security and

litigation incidents to the last 5 years. I then created a second request to list the top 10 vendors that had the most com plaints and negative press coverage over the same period. I spent about 30 minutes spot checking the data, then passed that along to the vendor committee for review. Another way I use ChatGPT is to create my own GPTs within our company-restricted version of ChatGPT. These GPTs are pre determined advanced prompts you can use repeatedly without needing to rewrite the entire prompt again. Before this, the best option for users was to manually create massively complex prompts and save them as documents. One could open the docu ment, change some target text to meet the new need, then paste it as a new prompt in the AI. Now, one can start with a well-de veloped prompt, or you can complete a form that prompts you for information to include in the GPT. You can preview the results and tweak the GPT until it meets your needs, then save it. A nice feature with your personal GPTs is that you can share PERSONAL (AND SHAREABLE) GPTS

40 july/august 2025