As artificial intelligence (AI) becomes increasingly prevalent in legal practice, lawyers must learn to effectively harness its potential. This article focuses on practical strategies for using AI, with a particular emphasis on prompt engineering and specific use cases across various legal tasks.
Identifying AI Tools for Legal Practice
Several types of AI tools are particularly useful for lawyers:
- Legal research platforms (e.g., LexisNexis, Westlaw, Casetext)
- Contract analysis tools (e.g., Kira Systems, Luminance, eBrevia)
- E-discovery software (e.g., Relativity, Everlaw, Disco)
- General-purpose AI (e.g., ChatGPT, Claude, GPT-4)
- Predictive analytics tools (e.g., Lex Machina, Premonition)
- Document automation software (e.g., HotDocs, Contract Express)
Each of these tools requires different approaches to prompting and use. Understanding the strengths and limitations of each tool is crucial for effective utilization.
Mastering Prompt Engineering
Prompt engineering is the art of crafting effective inputs for AI systems to generate useful outputs. Here are key principles and examples:
- Be specific and clear. Example: Instead of “Find cases about discrimination,” use “Find federal appellate court cases from the last 5 years addressing racial discrimination in hiring practices within the tech industry.”
- Provide context. Example: “Given a software license agreement for a cloud-based service, identify clauses related to data privacy, security, and service level agreements. Consider GDPR compliance implications.”
- Specify the desired output format. Example: “Summarize the key holdings in bullet points, including case names and citations. Then provide a brief paragraph synthesizing the overall trend in the court’s reasoning.”
- Use legal terminology accurately. Example: “Analyze the elements of negligence as applied in medical malpractice cases in California, with particular attention to the standard of care for specialists versus general practitioners.”
- Break complex tasks into steps. Example: “First, identify all provisions in this contract related to intellectual property. Then, compare these provisions to standard clauses in our template agreement, highlighting any significant deviations.”
Practical Examples of AI Use in Legal Tasks
Jumping straight into practical examples of how to get things done, below are a few examples of how effective prompts (and follow-up prompts) should be structured in the various types of tools identified above. Remembering the five principles above, you can use the examples below and modify them to whatever your needs might be.
Legal Research Tool: Westlaw Edge
Initial prompt: “Find cases from the Second Circuit Court of Appeals in the last 3 years that discuss the application of the fair use doctrine to digital sampling in music copyright cases. Provide a brief summary of each case’s holding, and any circuit splits noted. Then, compare these holdings to the most recent Supreme Court decision on fair use.” Follow-up prompt: “Based on these cases, draft a short memo (250 words) outlining the current state of fair use doctrine as applied to digital sampling in the Second Circuit.”
Contract Analysis Tool: Kira Systems
Initial prompt: “Review this merger agreement and extract all clauses related to representations and warranties. Highlight any unusual or non-standard language. Then, compare these clauses to our standard merger agreement template and identify any material differences.” Follow-up prompt: “For each material difference identified, provide a brief explanation of potential risks or benefits associated with the non-standard language.”
E-Discovery Tool: Relativity
Initial prompt: “Identify all emails in the dataset that mention both ‘Project Alpha’ and ‘budget overrun’ within a 5-word proximity. Exclude any emails marked as privileged. Then, create a timeline of these communications, noting the key participants and any significant changes in tone or content over time.” Follow-up prompt: “From the timeline created, identify the top three most active participants in these discussions and summarize their main points or concerns.”
Legal Writing Assistance Tool: ChatGPT
Initial prompt: “Draft an introductory paragraph for a motion for summary judgment in a breach of contract case. The key issues are failure to deliver goods on time and quality of goods not meeting specifications. Use a formal, persuasive tone appropriate for a federal court filing. Include a brief overview of the facts and a clear statement of the relief sought.” Follow-up prompt: “Now, outline the main arguments for the body of the motion, including relevant case law citations from the Southern District of New York.”
Case Outcome Prediction Tool: Lex Machina
Initial prompt: “Analyze patent infringement cases in the Eastern District of Texas over the last 5 years. What percentage were decided in favor of the plaintiff? What is the average time to trial? Break down the success rates by technology sector (e.g., software, pharmaceuticals, telecommunications).” Follow-up prompt: “Based on this analysis, what strategies appear to be most successful for defendants in this jurisdiction? Provide specific examples from cases with favorable outcomes for defendants.”
Document Automation Tool: HotDocs
Initial prompt: “Create a template for a standard non-disclosure agreement that includes variable fields for party names, effective date, duration of confidentiality, and specific protected information. Include conditional clauses for mutual versus one-way confidentiality obligations.” Follow-up prompt: “Add a clause to the template addressing the return or destruction of confidential information upon termination, with options for different methods of confirmation.”
Optimizing AI Results
When dealing with excessively voluminous results, remember to narrow down your search using iterative refinement. Start with a broad prompt and progressively narrow it based on initial results. For example:
- Initial: “Find cases about data breach liability.”
- Refined: “Find cases about data breach liability for cloud service providers in federal courts since 2018.”
- Further refined: “Find cases about data breach liability for cloud service providers in federal courts since 2018, focusing on the standard of care required for encrypting stored data.”
- Final refinement: “Among the cases found, identify those where the court discussed specific encryption standards or best practices that influenced the liability determination.”
Once you master each tool, you should use various AI tools in conjunction for more comprehensive results. For instance:
- Use a general AI like ChatGPT to draft a research plan on a novel legal issue.
- Execute the plan using a specialized legal research tool like Westlaw or LexisNexis.
- Use Casetext’s CARA A.I. to find additional relevant cases that might have been missed.
- Employ another AI to summarize and synthesize the findings.
- Then, use a document automation tool to create a report template based on the research.
Finally, it can never be stressed enough that you are still the licensed professional and both your law license and your professional ethics are on the line when producing work product in your name. As such, all AI output should be reviewed, the citations should be cross-checked and verified, and no AI-generated work should ever be regarded as a final product until such a review is complete. For example:
- After using AI to summarize case law, manually check all cases to ensure correct interpretation (and their actual existence).
- When using AI for contract analysis, have a junior associate verify the AI’s findings and escalate any uncertainties to a senior lawyer.
- For AI-assisted legal writing, review the document for tone, persuasiveness, and adherence to local court rules.