Many litigators are curious about ChatGPT but hesitant to try it out using litigation documents because of the work product doctrine and the attorney-client privilege. A worry-free way for litigators to understand what ChatGPT can (and cannot) do for litigation is to experiment using randomly selected documents that have been publicly filed.

One exercise is to try to get ChatGPT to recreate a brief by feeding it background facts, the structure of the brief, and some pointers about how to write like a lawyer.

And no, you should not expect useful results—ChatGPT was not designed to provide them. But this exercise will showcase the potential of generative AI even in its relatively early stages. Below are the steps to try it yourself, with some example output at the end.

Find & Prepare The Inputs

• Find a straightforward motion to dismiss on a random docket. Using a motion to dismiss is a good starting point because the universe of relevant facts is contained within the complaint.

  
  

• Download the complaint and the opening memorandum of law.

  
  

• From the complaint, copy/paste the background sections into Word so you can clean them up a bit. No no need to go crazy—ChatGPT is not that picky. Feel free to include sections like “Parties” and “Jurisdiction” and “Nature of the Suit”.

  
  

• From the memorandum of law, copy and paste the preliminary statement into Word and clean that up too.

Experiment With Prompting

Also in your Word document, start crafting your prompts. Below are some suggestions (in sequential order).

• To Get ChatGPT into Character: You are an attorney at an elite litigation firm. Your goal is to draft a motion to dismiss a a complaint. Your tone should be persuasive and professional. You should never say anything that might be interpreted to harm our position, which I will explain below. Respond with OK.

  
  

• To Give ChatGPT the Facts from the Complaint: The complaint that we want to dismiss is premised on the below BACKGROUND FACTS. [Here, paste in the background sections you pulled from the complaint.] Respond with OK.

  
  

• To Give ChatGPT the Preliminary Statement: Before writing, you need to understand the GENERAL POINTS we are going to make, which is as follows. [Here, paste in the preliminary statement that you pulled from the memorandum of law.] Respond with OK.

  
  

• To Give ChatGPT the Structure of the Argument: You will write two sections, which should be called Section I: [Here, write whatever you’d like] and Section II: [Same]. Respond with OK.

After giving ChatGPT some variations of the prompts above, you will need to give ChatGPT instructions about how to write each of the arguments. Below are some screenshots of an example we tried out.

Before asking ChatGPT to start drafting, you might want to ask it to provide a list of the cases it will incorporate, with relevant quotes.

Reviewing The (Useless) Results

ChatGPT gave us cases and quotes that sounded good—it clearly understood what I was going for. But as we expected, the results were useless:

• Almost all of the quotes were made up

• The case captions often incorporated real party names, but were pulled from other jurisdictions.

• The citations were wrong.

• Most cases were motions to dismiss copyright infringement claims, which is what I wanted, but they did not actually involve the specific issues I was looking for.

In other words, the results were useless. We decided to go forward with the final step, and instructed ChatGPT to prepare the draft anyway. After tweaking some prompts, ChatGPT's draft sounded reasonable, even though it was totally unreliable.

The Takeaway

It is easy for litigators to spot ChatGPT's shortcomings. But a closer look at its off-the-shelf capabilities reveals that it has potential to change litigation. At midpage, we are building a research and drafting platform to make generative AI reliable and usable for attorneys. We'll begin rolling out our first version in the coming weeks—stay tuned.