[EDRM Editor’s Note: The opinions and positions are those of Craig Ball. This article is republished with permission and was first published on January 9, 2025.]
Next month, I’m privileged to be presenting on two topics with United States District Judge Xavier Rodriguez, a dear friend who sits in the Western District of Texas (San Antonio). One of those topics is “Practical Applications for AI.” The longstanding custom for continuing legal education in Texas is that a presenter must offer “high quality written materials” to go with a talk. I’m indebted to this obligation because writing is hard work and without the need to supply original scholarship, I’d probably have produced a fraction of what I’ve published over forty years. A new topic meant a new paper, especially as I was the proponent of the topic in the planning stage–an ask borne of frustration. After two years of AI pushing everything else aside, I was frustrated by the dearth of practical guidance available to trial lawyers–particularly seasoned elders–who want to use AI but fear looking foolish…or worse. So, I took a shot at a practical primer for litigators and am reasonably pleased with the result. Download it here. For some it will be too advanced and for others too basic; but I’m hopeful it hits the sweet spot for many non-technical trial lawyers who don’t want to be left behind.
Despite high-profile instances of lawyers getting into trouble by failing to use LLMs responsibly, there’s a compelling case for using AI in your trial practice now, even if only as a timesaver in document generation and summarization—tasks where AI’s abilities are uncanny and undeniable. But HOW to get started?
The Litigation Section of the State Bar of Texas devoted the Winter 2024 issue of The Advocate magazine to Artificial Intelligence. Every article was well-written and well-informed—several penned by close friends—but no article, not one, was practical in the sense of helping lawyers use AI in their work. That struck me as an unmet need.
As I looked around, I found no articles geared to guiding trial lawyers who want to use LLMs safely and strategically. I wanted to call the article “The Leery Lawyer’s Guide to AI,” but I knew it would be insufficiently comprehensive. Instead, I’ve sought to help readers get started by highlighting important considerations and illustrating a few applications that they can try now with minimal skill, anxiety or expense. LLMs won’t replace professional judgment, but they can frame issues, suggest language, and break down complex doctrines into plain English explanations. In truth, they can do just about anything that a mastery of facts and language can achieve.
But Know This…
LLMs are unlike any tech tool you’ve used before. Most of the digital technology in our lives is characterized by consistency: you put the same things in, and other things come out in a rigid and replicable fashion. Not so with LLMs. Ask ChatGPT the same question multiple times, and you’ll get a somewhat different answer each time. That takes getting used to.
Additionally, there’s no single “right” way to interrogate ChatGPT to be assured of an optimal result. That is, there is no strict programming language or set of keywords calculated to achieve a goal. There are a myriad number of ways to successfully elicit information from ChatGPT, and in stark contrast to the inflexible and unforgiving tech tools of the past, the easiest way to get the results you want is to interact with ChatGPT in a natural, conversational fashion.
Marketplace Models and Versions
The LLM market has boomed, fueled by a trillion dollars in corporate investment with many companies leading the charge in developing and deploying tools for consumer use, each playing a vital role in advancing AI technologies. Here are some market leaders:
- OpenAI: Offers the GPT series, with the latest being GPT-4. These models are accessible through products like ChatGPT and are integrated into various applications.
- Google DeepMind: Developed the Gemini series, with versions like Gemini 1.0 and Gemini 1.5. These models power AI capabilities in Google’s services.
- Anthropic: Provides the Claude series, including Claude 2 and Claude 2.1, designed for conversational AI applications.
- Meta AI: Developed the Llama series, with Llama 2 and Llama 3.1 being notable versions. These models are used in various AI research and applications.
- Microsoft: Introduced the Phi series, such as Phi-2 and Phi-3, which are marketed as “small language models” for efficient AI tasks.
My article focused on OpenAI’s ChatGPT, specifically its ChatGPT-4 product. I don’t feature it because it’s necessarily the best or brightest; it’s simply the one I use most often and pay just $20.00 per month to access. There’s a huge difference in the capabilities and user experience between the free ChatGPT versions and the subscription offerings. If you’re going to integrate an LLM into your practice, even just to kick the tires, opt for the pay-to-play versions. The enhanced performance of ChatGPT-4 justifies the modest investment, particularly when compared to the hourly cost of professional staff or consultants.
So, yes, doing the things I describe will cost twenty bucks, but you can do all these things and many more for a month or two, then cancel if you don’t see value in continuing to subscribe.
Getting Started
Getting started with ChatGPT-4 is surprisingly simple, even for lawyers who aren’t particularly tech-savvy but have intellectual curiosity and critical thinking skills honed through years of trial practice. At its core, ChatGPT-4 is a conversational artificial intelligence tool designed to help with tasks like drafting, research, and document analysis. To begin, visit chat.openai.com and create an account using your email address and date of birth. Upgrade your account to a $20.00 per month ChatGPT Plus subscription.
Once logged in, you can enter prompts—questions or instructions—to generate responses. Think of ChatGPT as a hyper-efficient research assistant who can respond instantly to requests for drafting demand letters, summarizing testimony, or brainstorming trial strategies.
To maximize ChatGPT-4’s usefulness, clarity and specificity in prompts are key. The tool thrives on detailed instructions. For example, instead of saying, “Draft a request for production for a baby powder cancer case,” be more specific: e.g., “You are counsel for the plaintiff in a wrongful death case pending in a Harris County Texas Civil District Court. Draft a request for production seeking information from defendant Johnson & Johnson in connection with cervical cancer allegedly caused by exposure to talcum powder containing asbestos. Draft 25 separate requests and be sure to include the language needed to obtain electronically stored information.”
Privacy
As lawyers, we’re entrusted with sensitive, privileged and private information we are obliged to protect. If you’re worried that interactions with ChatGPT may expose data to third parties by training the model with confidential information, be sure to adjust your settings to protect your privacy.
To modify ChatGPT’s settings:
1) Open the ChatGPT app or website.
2) Go to ‘Settings’ or ‘Data Controls.’
3) Toggle off the option labeled ‘Improve the model for everyone’ by sliding the button to the left until it’s grayed out.
4) Click ‘Done” to save your changes.
By default, ChatGPT will keep your chats, but with the privacy setting changed, only those with access to your account or a shared link can get to that content.
ChatGPT lets you share a link to your conversation, and anyone with the link can see that conversation as it existed at the time the link was created. While shared links can streamline collaboration and access to important documents, they also pose risks of unauthorized access, data leaks, and inadvertent sharing. Be cautious when sharing links generated by ChatGPT, as they may be publicly accessible if not properly restricted. To maintain confidentiality, ensure links are shared only with authorized recipients, and avoid posting them in public forums. Review sharing permissions in Settings>Data Controls>Shared Links>Manage
Protecting the privacy of sensitive data doesn’t end with turning off a toggle in ChatGPT’s settings. It’s a good start, but no substitute for a thoughtful, multilayered approach to privacy—a subject beyond the scope here, but not to be glossed over in practice.
Hallucinations
Working with ChatGPT is like teaching a teenager to drive. Sure, they have better vision, hearing, and reflexes; but, at any moment, you may find yourself screaming, RED LIGHT! RED LIGHT! ChatGPT is merely a tool, one only as capable and trustworthy as your ability to direct and monitor its use. Like any power tool, when you fail to respect its risks and limitations, it can hurt you; operated skillfully, it will make your work easier. ChatGPT’s tendency to hallucinate—i.e., fabricate information—requires constant vigilance. ALWAYS cross-check its output, particularly when legal accuracy is critical. If the tool makes up a case or a statute, you’re the one who’ll have to answer for it. I can’t emphasize this enough. Courts do not want to hear that you didn’t verify your citations.
Customizing Settings for Litigation Efficiency
ChatGPT-4 offers customization settings that can make it even more efficient for litigation and trial work. When using the tool, you can adjust settings to enhance productivity:
- Temperature Control: A lower temperature setting (0.0) generates more focused, fact-based responses, while a higher temperature (1.0) generates creative outputs. For litigation tasks like drafting discovery requests or deposition questions, keep the temperature low for precision. ChatGPT can adjust the temperature of its response based on the wording of your prompt: incorporate language advising whether you are seeking, for example, “formal and precise responses” versus, e.g., “creative and expansive responses.”
- System Instructions: For advanced users, tools like Custom Instructions allow you to direct the model’s behavior. For example, you can instruct: “Always format output as bullet points unless asked otherwise.” This reduces time spent reformatting for legal work. You can add persistent instructions at Settings>Personalization>Custom Instructions
- Memory: OpenAI is rolling out memory features that allow ChatGPT to keep context across sessions, making it better suited for ongoing case work. You can view and change what ChatGPT remembers across chats at Settings>Personalization>Memory>Manage. Getting a feel for memory and context limits is helpful when assessing why you’re not achieving the results you seek when tweaking prompts.
The User Interface and its Token, Upload, and Context Limits
When users interact with ChatGPT-4, they typically do so through the Chat Window (below), a text box with icons in the lower left corner. Users enter prompts and get responses in the Chat Window. The paperclip icon initiates a file upload.[1]
If you’ve experimented with ChatGPT, chances are you’ve been frustrated by limitations stemming from a failure to understand the complex token, upload, and context limitations of the tool, even if you didn’t appreciate why things weren’t working.
Tokens in ChatGPT are units of text used to measure the length of input and output. Tokens can be as small as a single character or as large as a short word, depending on the structure of the text. All inputs and outputs with ChatGPT consume tokens. Each punctuation mark and space consume a token, and longer, compound words consume multiple tokens.
As a rule of thumb, one page roughly equals 375-400 tokens, and 10,000 tokens ≈ 7,500 words ≈ 25 pages.
As I write this early in 2025, these limitations in ChatGPT-4 (in the Plus plan) are:
- Token Limit: In the Plus plan, 32,000 tokens per Chat Window (encompassing both prompts and responses). That is more than 75 pages of text!
- Upload Limit: 20 documents of 512MB or less, but not more than 2,000,000 tokens per file (except spreadsheet and CSV files, where the file cap is closer to 50MB). For images, there’s a limit of 20MB per image. But note that this limit refers to how much data you can upload in a single file for analysis. This doesn’t mean the model will process all 2,000,000 tokens at once. Instead, it can access parts of the file as needed, constrained by the 128,000-token context window discussed below. In my experience, the program often hangs up or “chokes” on large files.
- Context Limit: In the Plus plan, up to 128,000 tokens (~ 300 pages of text). This defines how much information the model can “hold in mind” during a single interaction.
Is this confusing? Totally! Think of the context window as the size of a workspace where you can lay out up to 128,000 tokens of information at once. The file upload limit is more like the size of the entire library (2,000,000 tokens per file). You can upload a large book, but only part of it can be “open” on the workspace at one time. The higher token limit for uploads allows users to send larger files for reference or incremental analysis without needing to break them up manually.
Why Tokens Matter
These limits mean that ChatGPT will fail to upload files that are too large and will truncate longer documents such that ingestion of a longer document will peter out when you bump up against upload and token limits. Moreover, ChatGPT will “forget” prior prompts, responses, and interactions as the context limit is reached. That’s one reason why it can be difficult to refine responses as older prompts and responses are jettisoned in favor of later interactions. Holding onto context entails incorporating summaries and key details in later interactions.
In short, LLMs have strict data size limits—don’t expect them to digest entire case files at once.
The upshot is that the dream of uploading all the depositions and evidence in your case and letting the AI work its magic is not yet a practical reality for most ChatGPT users. But just because the limits prevent ChatGPT from dealing with everything, it can still manage minor miracles with certain things, although that can entail processing longer documents in sections, dividing tasks into multiple prompts and splitting large datasets into manageable chunks to avoid missing data.
Uploading Files
Clicking the paperclip icon in the chat window enables uploading of documents from your computer or from linked online repositories. Turning ChatGPT loose on documents enables it to be trained on forms, summarize and analyze testimony, read medical records and carry out countless other tasks. The tool is extraordinarily adept at extracting intelligible information from a variety of electronic formats; for example, ChatGPT can read Word documents, Excel spreadsheets, PDF files with extractable/selectable text, and comma-separated value (CSV) files exported from databases.
Practical Apps for Litigation and Trial Practice
Drafting Discovery Requests
AI tools excel at translating narrative descriptions into actionable discovery requests. For example, you can describe the type of information or documents you need, and ChatGPT can craft detailed, targeted requests.
Summarizing Voluminous Data Sets
LLMs can sift through testimony or productions to generate summaries or extract critical insights. If you’re looking for specific patterns or themes, AI can save hours of manual review.
Structuring Information
ChatGPT can help you create tables, timelines, or other structured formats for organizing case details. Simply prompt it with the type of structure you need, and it will supply a starting point.
Search Term Generation and Query Assessment
Turning natural language prompts into search terms or Boolean queries is another area where ChatGPT shines. It can also analyze the effectiveness of existing search terms, refining them for better results.
Drafting Pleadings and Arguments
Whether it’s crafting an opening statement or refining a motion, ChatGPT can suggest language, organize arguments, and polish drafts to save time.
Demonstrative Evidence
LLMs can generate visuals or outlines for demonstrative evidence. While they won’t create admissible exhibits, they can help conceptualize and draft what you need.
Ten Examples with Prompts
ChatGPT-4 can be a game-changer in litigation and trial preparation, going beyond simple document drafting. Now that we’ve discussed what it can do at a high level, here are some practical, effective ways to deploy it, and remember, these are just examples to get you started, not “magic words;” don’t be afraid to frame the prompt to fit your case:
1. Drafting Deposition Questions:
- Prompt Example: “Generate a list of deposition questions for a corporate designee in a products liability case involving a defective medical device. Focus on chain of custody, design specifications, and failure reports.”
- Application: Build a solid outline for questioning key witnesses based on liability theories.
2. Discovery and ESI Protocols:
- Prompt Example: “Draft a meet-and-confer letter outlining an ESI protocol for a federal employment discrimination case. Address data sources, search terms, metadata fields, and custodians.”
- Application: Save time drafting initial versions of formal discovery documents and ESI agreements. Tip: you might try uploading my free primer on ESI Protocols to ChatGPT along with your prompt: http://www.craigball.com/ESIProtocol.pdf
3. Trial Preparation:
- Prompt Example: “Create a cross-examination outline for a medical expert testifying about causation in a personal injury case. Highlight potential bias, financial incentives, and prior inconsistent statements.”
- Application: Develop targeted cross-examination questions and issue-spot for trial strategy.
4. Case Summaries for Motion Work:
- Prompt Example: “Summarize the key holdings and reasoning from Zubulake v. UBS Warburg focusing on preservation and spoliation standards in federal eDiscovery.”
- Application: Generate concise case law summaries for inclusion in motions and briefs.
5. Privilege Logs and Document Review:
- Prompt Example: “Generate a privilege log template for a federal securities fraud case including columns for document date, author, recipients, privilege type, and a brief description of withheld content.”
- Application: Assist with large-scale document reviews and privilege log generation.
6. Voir Dire Preparation:
- Prompt Example: “Generate a list of voir dire questions for a jury selection in a wrongful death case involving a truck collision. Focus on attitudes toward commercial drivers and damages claims.”
- Application: Develop probing questions to identify bias and juror attitudes during selection.
7. Opening and Closing Argument Assistance:
- Prompt Example: “Draft an opening statement for a breach of contract case where the defendant failed to deliver construction materials, causing project delays and financial loss.”
- Application: Generate persuasive language and themes for trial presentation.
8. Crafting Jury Instructions:
- Prompt Example: “Generate plain language jury instructions for a negligence case under Texas law. Include elements of duty, breach, causation, and damages.”
- Application: Assist in drafting jury instructions that meet clarity and accuracy standards.
9. Case Chronologies:
- Prompt Example: “Create a timeline summarizing key events in a breach of fiduciary duty case involving financial mismanagement by a corporate officer.”
- Application: Organize case facts into a clear, chronological format for internal use or presentation.
10. Building Better Prompts:
- Prompt Example: “Help me draft a prompt for ChatGPT-4 that will extract data from an uploaded set of interrogatories and generate an Excel spreadsheet breaking out for each question the sources of information that should be searched to formulate a response and appropriate objections pursuant to the Texas Rules of Civil Procedure.”
- Application: Use ChatGPT to achieve a defined goal by outlining the objectives and letting the tool come up with the best prompt for the task. You’ll be amazed how adept ChatGPT is at talking to itself!
Ten Tips for Improved Prompts
The following are ways you can improve prompts to better realize your goals, but you don’t need to slavishly follow these tips to achieve satisfactory results. That said, if you’re not getting what you want, refine your approach as suggested below.
1. Precision Matters: Frame with Factual and Legal Specificity
- Avoid generalities—refine prompts with detailed facts and legal context.
“Draft voir dire questions tailored for a medical malpractice case involving a delayed cancer diagnosis where the plaintiff alleges a breach of the standard of care in Texas.”- Specificity clarifies intent and reduces irrelevant results.
2. Direct ChatGPT to Perform Defined Legal Tasks
- Frame prompts in terms of practical litigation tasks.
“Generate a Rule 26(f) conference checklist for a complex products liability case emphasizing ESI preservation obligations and proportionality considerations.”- Task-focused prompts ensure outputs align with litigation strategies.
3. Break Complex Requests Down into Logical Steps
- Divide multifaceted tasks into clear components.
“First, list the elements of defamation under Texas law. Next, draft a motion to dismiss based on failure to meet the public figure standard.”- Stepwise instructions prevent the model from oversimplifying complex legal issues.
4. Instruct the Model to Assume a Role
- Direct ChatGPT to take on a specific legal perspective for better focus.
“You are a federal magistrate drafting a discovery order where both parties dispute the scope of ESI production in a commercial contract dispute.”- Role-specific prompts improve relevance and formality.
5. Request Multiple Variations for Strategic Flexibility
- When creativity or optionality matters, request diverse outputs.
“Provide three alternative opening statements for a wrongful death trial: one emphasizing sympathy, another liability and the third focusing on damages.”- Variation expands strategic possibilities for trial presentations.
6. Use Iterative Refinement: Request Successive Adjustments
- Fine-tune outputs by requesting modified versions.
“Revise the following deposition outline to focus more on the defendant’s conflicting statements during prior testimony.”- Iterative prompting helps refine output toward a trial lawyer’s evolving needs.
7. Focus on Legal Frameworks and Case Law Alignment
- Encourage responses rooted in legal standards rather than speculation.
“Summarize the elements of promissory estoppel under Texas law, citing Restatement (Second) of Contracts principles where applicable.”- Reducing open-ended phrasing helps avoid speculative results.
8. Clarify the Desired Level of Detail and Tone
- Tailor the depth and tone based on audience and context.
“Explain the concept of comparative fault in premises liability cases for a CLE presentation—formal but accessible to non-litigators.”- Clear tone instructions ensure the content fits the intended audience.
9. Use “If-Then” Prompts to Explore Conditional Reasoning
- For nuanced analysis, frame prompts with conditional logic.
“If a plaintiff fails to disclose an expert within the deadline under Rule 26, explain both the exclusionary risks and potential grounds for leniency.”- This encourages balanced exploration of legal scenarios.
10. Explicitly Request Citations and Source Validation
- Prompt ChatGPT to acknowledge information reliability limits.
“Explain the principles of spoliation sanctions under Zubulake and detail any potential conflicts with the FRCP’s 2015 amendments. Note where independent case law review may be necessary.”- Encourages awareness of when verification is critical in litigation.
Understanding Limitations and Ethical Guardrails
While powerful, ChatGPT-4 is no substitute for legal expertise. It does not access live court databases like Lexis, Westlaw or PACER, and as discussed, it can generate incorrect or incomplete information (“hallucinations”). Always read the cited authority and verify case citations and factual assertions before use. Consider it a junior associate: fast, insightful, but requiring close supervision.
Final Thoughts for Trial Lawyers
As a law professor, longtime trial lawyer, and special master, I’ve found ChatGPT-4 immensely useful for early-stage drafting, ideation, and breaking through writer’s block. I concede that AI and LLMs are no substitute for a skilled lawyer. They won’t replace the art of advocacy. But they can sharpen your practice by automating routine tasks, helping you explore arguments, and improving document clarity. Whether you’re drafting deposition questions, summarizing key cases, or preparing ESI protocols, ChatGPT-4 can be a force multiplier—a powerful tool to amplify your abilities—especially when you learn to guide it with well-tailored prompts.
Embrace AI as a tool for efficiency, and you’ll soon find it becomes an indispensable part of your litigation arsenal; but remember: every tool needs a craftsman, and no amount of technology can fully replace the judgment and insight of a seasoned trial lawyer.
Read the original article here.
Notes
[1] That black circle on the lower right of the Chat Window opens Voice Mode, allowing you to converse with ChatGPT as you would with the world’s best-informed colleague who is unfailingly polite and helpful. Prepare to be awestruck!
Assisted by GAI and LLM Technologies per EDRM GAI and LLM Policy.
