Careers

Diversity, Equity & Inclusion

Experience

Innovation

Interactive Maps

Locations

About

Pages

Meet our team of innovators
Ask us how we do it

Professionals

Insights

Services

Thompson Hine LLP, a full-service business law firm with approximately 400 lawyers in 8 offices, was ranked number 1 in the category “Most innovative North American law firms: New working models” by The Financial Times and was 1 of 7 firms shortlisted for The American Lawyer’s inaugural Legal Services Innovation Award. The firm has been recognized by the National Law Journal as a Legal Technology Trailblazer and featured by Bloomberg for its innovative service delivery models, which drive accuracy and predictability. The firm’s commitment to innovation is embodied in Thompson Hine SmartPaTHTM – a smarter way to work – predictable, efficient and aligned with client goals. For more information, please visit ThompsonHine.com and ThompsonHine.com/SmartPaTH.

Home | Thompson Hine LLP

Atlanta

Chicago

Cincinnati

Cleveland

Columbus

Dayton

Los Angeles

Minneapolis

New York

Washington, D.C.

The advent of artificial intelligence, and in particular the public introduction of <a>generative artificial intelligence (“generative AI”)</a> over the last 18 months, presents perhaps the greatest opportunity in the last half-century to fundamentally improve efficiency, thoroughness and polish in a transactional lawyer’s work product. What’s more, generative AI might accomplish all of that with impressive speed in the coming few years. Yet, despite the promise of exponential and expedited progress, faster is not always better, and for the unwary, <a>generative AI</a> can cause more problems than it solves. One critical step to lawyers finding themselves on the right side of generative AI use is mastering “prompt engineering,” which is the process—the art!—of creating the prompts entered into a generative AI platform so that it can return a response or draft that is most usable for your purposes. Think of it intuitively as framing the questions the right way for the generative AI to answer them—like running an efficient Google search. The better your input, the better the generative AI output.

contentpilot/introduction

The art of prompt engineering is very much in its early stages, but below are a few general tips to help you make the most of your prompts with whatever generative AI program you use.

core/paragraph

1. Treat It Like It’s a Colleague. This tip is a really general one, but it helps frame the perspective with which you should approach prompting—that you are asking questions, or giving instructions, rather than necessarily focusing on getting the particular output. In other words, don’t treat generative AI like it’s Google; instead, ask questions in full sentences as if the generative AI were a person. For example, instead of “purchase agreement indemnification sample provision,” which you might use if you were Googling the concept of indemnification provisions, expand the prompt to “Draft a purchase agreement indemnification that involves the seller indemnifying the buyer for breaches of representations and warranties as well as covenants.”

2. Choose Being Detailed Over Being Concise. This point may be an opinion that others would disagree with, but we tend to recommend being detailed—erring on the side of more detail rather than less—even if your lawyerly instincts tell you to be brief. That includes telling the generative AI which sources to use (“focus on the ABA deal study”) or exclude (“exclude the 10-Ks of any companies with a market cap greater than $2 billion”), which statutes to analyze (“only focus on the Ohio Revised Code”) or anything else that might come to mind. Indeed, one of the generative AI sources we use caps your prompts at 100,000 characters (the equivalent of about 670 Twitter/X posts)—so don’t be afraid to be detailed.

3.&nbsp;Specify the Audience (and, Possibly, Tone). If you’re asking the generative AI to provide sample correspondence—say, an email or a component of a memorandum or report—then it helps to tell the generative AI who you’re writing it for. Perhaps use something like, “I’m drafting a memo advising a client on piercing the corporate veil and want to help clarify for the client the risks,” or “I’m preparing an email to sellers’ counsel to propose a settlement offer regarding an indemnification matter.” You can even go so far as to specify the tone or the approach—to the latter example, you could add, “I want the settlement offer to be friendly but firm, and express that we are offering this as a last resort before going to litigation.” Adding that tonal clarification and building on the rules in (1) and (2) can provide a really useful response you can further hone. Finally, specifying the audience can help the generative AI make its explanations more intuitive—you can ask it to explain concepts to you as if you’re a seasoned lawyer, a first-year associate or just a smart 10-year-old. Our experience has been that generative AI is exceptional at using analogies and breaking down complex concepts, so the prompts can help be an excellent teaching tool.

4. Offer Context and Surrounding Facts. Much as a colleague appreciates being given instructions in context rather than in a vacuum, generative AI does too (well, it improves the output—we can’t speak to generative AI “appreciating” the instructions). You’ll want to be sensitive to whether to include party names or other identifying facts (depending on your organization’s AI use policy and open source/closed source considerations raised below), but otherwise providing the context in which the request is being made can be useful. For example, in the context of a settlement offer, you may wish to include the proposed terms of the settlement in the prior rounds of negotiations. Doing so can help the generative AI frame the subsequent correspondence, both tonally and in terms of issues.

5. Specify How Much Output You Want. Especially for items like correspondence or memoranda, you can tell the generative AI exactly how much you want it to write. For example, to the indemnification agreement example in (1) above, you might say, “Draft it in two paragraphs, segmenting out between representations and warranties in subsection (a) and covenants in subsection (b).” If you do not specify the format for a response, what you get may be meandering, too brief, too formal, too informal, or otherwise require more work for you to tailor. You can ask, for example, for a response in an informal email format using six bullet points to organize the information. If you’re asking the generative AI to help craft arguments for why you’re taking a position in purchase agreement negotiations, rather than simply “provide five arguments for why the undisclosed liabilities representation pasted below is appropriate,” you might add “but limit each argument to one or two sentences as I don’t need too much detail.”

6. Refine Your Prompt (Constantly). Read the prompt you’ve drafted again and issue spot for confusing or vague language. Clean it up and then execute the search. Even adding a sentence or two of detail can meaningfully improve the quality of the response—or alternatively, removing superfluous details can have the same effect. Again, treat generative AI like a colleague, an exceptionally ambitious first-year associate, whom you both review and train. And, of course, make sure that the output you get both makes sense and is accurate (search <a>Mata v. Avianca</a> if you wish to read about one such example of inaccuracy).

Two final reminders about prompts: First, before you enter any prompts into generative AI, make sure you check your company’s generative AI use policy. Generative AI policies help clarify what generative AI platforms employees can use as well as what types of information they can put into the platforms, which helps protect your company’s data. If your company doesn’t yet have one, but people are starting to use generative AI at your company, then put simply, you should have a policy. Our firm has lawyers who have extensive experience developing corporate AI policies who can help you draft and implement an appropriate policy that can change and grow with your organization and the constantly transforming generative AI landscape. Second, be aware of whether the generative AI you’re using is “closed source” (i.e., entirely contained within your internal servers) or “open source” (i.e., akin to running a Google search). If your prompt involves confidential information, you’ll want to be abundantly sure that you’re using a closed-source platform—if it’s an open-source platform, you risk sharing your confidential information with the world (or otherwise getting hacked).

The tips above can get you started on your journey to becoming a more efficient M&amp;A practitioner as you bear in mind both the limits and the possibilities of generative AI. As we (and you) get better at prompting, there’s no doubt that generative AI can make us better lawyers. Happy prompting!

https://admin.thompsonhine.com/wp-content/uploads/2024/07/Thompson-Hine-Newsletter-Business-Law-Update-July-2024.pdf

core/file

This article may be reproduced, in whole or in part, with the prior permission of Thompson Hine LLP and acknowledgement of its source and copyright. This publication is intended to inform clients about legal matters of current interest. It is not intended as legal advice. Readers should not act upon the information contained in it without professional counsel.

This document may be considered attorney advertising in some jurisdictions.

Prompt Attention Required: Best Practices for Using Generative AI by Transactional Lawyers

Services