In-house attorneys know that legal interns can be valuable collaborators on projects like first drafts, document reviews and research. They can give in-house counsel the bandwidth to focus on high-level legal strategy.
The same can be said for AI. Even better, new legal-specific AI solutions can act more like associates with legal knowledge and work experience.
But just like an intern or associate might struggle with vague instructions, AI tools rely on well-crafted prompts for the most useful outcome.
Prompting 101
The concept of “garbage in/garbage out” or “GIGO,” is a computer science term that was coined in the 1950s and is still applicable today. If you provide AI with an imprecise or ambiguous prompt, you’ll get unhelpful results in return. However, with a few simple techniques, in-house counsel can turn generic prompts into “artisan” prompts designed to enhance efficiency.
1. Use formatting and symbols to guide AI:
AI models like OpenAI have learned to “read” code because training data includes open source computer code. That means AI trained on code understands symbols, like hashtags, as they are used in specific ways to code.
For example, # can denote headers and help AI organize its responses. This applies to other symbols used in coding, like square brackets [] to indicate placeholders or provide context, or mathematical symbols, including equal =, plus + and minus -.
In addition, social media content has also been used as AI training data. Hashtags can draw AI’s attention to important words or phrases in a prompt, just like in social media.
2. Be specific; use numbers to sharpen the output
Specificity is crucial in AI prompting. For an outside-the-office example, notice the difference between:
“Give me recipes for easy weeknight meals,” and
“Give me 7 recipes for easy weeknight slow cooker meals. Include exact measurements and cooking instructions.”
A straightforward way to return more specific and actionable results is to use numbers in prompts. For example, asking for five bullet points for a client memo or analysis of each of the 13 factors in the DuPont test for likelihood of confusion in trademark law will provide a better starting point for advising your company.
3. Provide context; tell AI who you are:
AI prompts should not be created in a vacuum. If legal teams provide AI tools with information on context and tone, the results will better suit their needs.
In the weeknight meal recipe example, the results will be more useful if the prompt specifies that the recipes must include ground turkey, feed four people, including small children and accommodate a dairy allergy.
In the legal context, in-house teams can provide examples of prior memos and contracts and information about the audience for the advice. Make sure to use a legal-specific AI tool that has security features that will keep your information safe and confidential.
One overlooked aspect of providing context is letting AI know that you are a lawyer. If your prompt requests legal information, AI tools like ChatGPT are likely to return results with an “I am not a lawyer” disclaimer response. Avoid this by informing it that you are a lawyer and are qualified to provide legal advice, or use a legal-specific AI tool that is engineered on the backend with instructions about being a lawyer
Putting it into action
Below are three AI prompts, beginning with a vague prompt, to illustrate how simple tweaks can greatly improve output:
Weak: What is the law on the statute of limitations in copyright cases?
Better: What is the current position of the U.S. Supreme Court regarding application of the discovery rule versus the injury/accrual rule on the statute of limitations in copyright cases?
Best:
# Task: Summarize the U.S. Supreme Court’s decision in Warner Chappell Music v. Nealy, providing 5 bullet points on the application of the discovery rule versus the injury/accrual rule. The tone should be professional.
## Opening paragraph (title: “Executive Summary”): Give me a general assessment of the potential risks of the decision in the publishing industry for C-suite readers.
## Bullet points: List 5 key takeaways after the opening paragraph.
Quality in, quality out
AI is rapidly becoming a valuable tool for in-house legal teams, particularly for these use cases:
- replicable, high-frequency tasks such as contract review and drafting or preparing first drafts of simple client communications;
- tasks that require synthesis of high volumes of data such as research, regulatory review and benchmarking.
Unlike an intern, AI can sift through vast amounts of legal data in seconds. But like an intern, it needs the right information and guidance to succeed. By simply refining your prompts, you can turn AI into a powerful, time-saving legal assistant.
For more insights on becoming a pro prompter, see GC AI’s Prompting 101 course.
