Should lawyers learn prompt engineering? Being a prompt pro is not a prerequisite for using generative artificial intelligence (AI). But knowing the basics gives lawyers an advantage.
To unlock the benefits of prompt engineering, there are two things to be aware of. Firstly, how the models use AI to retrieve and compose information, and secondly, how to write effective prompts.
The rest, you already have as a legal expert. By using plain language and your subject matter expertise to write an effective prompt, you’ll obtain more accurate results. After all, AI must be overseen by (human) lawyers to ensure accuracy.
There’s also a third best kept secret: intuitive technology. Innovative tools like CoCounsel are inbuilt with prompt engineering suggestions to make the user experience seamless for lawyers, meaning no prompt engineering background is needed. Here’s how to write effective prompts based on the logic behind it.
What is prompt engineering?
Prompt engineering is the practice of formulating target queries to extract relevant results from the AI tool being used by the lawyer or solicitor. A prompt is merely a sentence, written in natural language, that provides the tool with the task required as well as any additional information. The practice is iterative.
Professional engineers test, evaluate, and optimise results by experimenting with phrasing, terminology and instructions. Prompt engineering practices include ethical considerations, which means reviewing results and iterating your prompts to remove bias or misinformation.
A clever prompt will help to generate responses that are accurate and specific to the user’s needs. It will assist with mitigating inaccuracies or so-called hallucinations.
What should lawyers know about prompt engineering?
Generative AI is redefining how lawyer’s work. An AI legal assitant will help hasten tasks including legal research, drafting, and analysis. With the heavy lifting taken care of, you’ll be free to spend more time with clients, on strategy, and adding value to the business. But first, you must know how to prompt.
Lawyers should incorporate their legal expertise into the logic of prompt engineering to generate better results. But to gain this competitive advantage as a legal expert, you need to know some prompt engineering basics. A baseline understanding of how to frame a prompt will help you gain valuable results from vast databases, maintain legal nuance, streamline your day and improve the precision and quality of your work.
How should lawyers write prompts for generative AI tools?
It’s all about clear communication. The easiest way to craft a prompt is to simply think of how you would effectively brief a human.
- What is a clear outline of the task?
- What are they expected to provide?
- What guidance would they need?
- Are they clear on context?
- Are there any caveats?
The more accurate the prompt, the more accurate the results.
5 tips for writing effective legal prompts
1. Use precise language
Phrase your prompt without ambiguity. Go for a simple sentence structure and be concise. Avoid being too formal, so ditch the passive and write in the active voice. It helps to think how you would explain something to someone conversationally. This also means being inclusive, so spell out abbreviations and avoid using colloquial language. Review these helpful before and after examples for how to fine-tune the phrasing of your prompts.
2. Provide context
For more complex prompts, provide any background information that would help to define your query and account more clearly for legal nuance. Think of the tool as an actor that needs to familiarise themself with a role. Start by providing the tool with an introductory statement that sets the scene and explains the role it needs to play.
3. Be specific
Clearly state the task and the expected format of the output. Provide examples if you can. This will help to prevent the technology from filling in any gaps, which could lead to imprecise results.
4. Tweak your prompts
To refine the result, you’ll need to iterate prompts. Think of it as a two-way conversation. Some back and forth to reach a shared understanding is to be expected. An iterative process will ensure a more accurate result.
5. Be responsible
Monitor your results for biases and be mindful of ethical considerations. Responsible AI relies on human oversight of the results, and this is where you as a legal professional can apply your critical training in the field. Would you like it to explain its logic to you or cite all the sources it used. Ask it to.
Related reading: How is AI regulated in Australia? What lawyers should know.
5 good and bad AI prompt examples
Looking at these examples will show you how to use the tools effectively in your daily work. We’ve used CoCounsel (part of Thomson Reuters) as the example of the prompting recipient:
Prompt 1: “I am representing a global conglomerate, Hypothetical Example Holdings, who are looking to divest a part of their business. We’re in the vendor due diligence process now. How can CoCounsel help me?”
Why it works: It provides context and consults the tool first for strategic advice. By learning how the tool can help, you can improve your prompts on the topic with new insight.
Prompt 2: “Please review this contract to determine whether it complies with the following policies: (1) Any anti-assignment provisions should be mutual and should authorise assignment in a merger, acquisition or other change of control. (2) Governing law should be New South Wales.”
Why it works: It’s specific. It specifies the task and the jurisdiction for CoCounsel to examine.
Prompt 3: “Can you please help me draft a letter? Address the letter to Jane Doe, the lawyer for Example Tech, the seller under this contract. I represent the buyer, Example Holdings.
“The letter needs to say that I am sending back an updated draft of the agreement and explain why. Please summarise the ways in which the current agreement conflicts with my client’s policies, per CoCounsel’s contract policy compliance report, above.”
Why it works: It is providing context and detail.
Prompts 4 and 5: “List me the top five contracts in this database”, or “Give me a legal contract that I can use for my multinational clients.”
Why they don’t work: Neither example is specific enough to generate useful results. Additionally, the expected output (e.g. ‘a contract’) is not sufficiently explained.
Should lawyers train as AI prompt engineers?
Knowing how to write effective prompts is crucial to getting what you need from the tool. However, you don’t need to spend much time upskilling.
Remember, you are only as good as your tools. Use generative AI technology that is built for legal professionals like you. Leverage AI that speaks the language of lawyers, so you don’t have to be a pro at prompt engineering.
About CoCounsel, the AI legal assistant
CoCounsel Core, the world’s leading AI legal assistant, uses machine learning to train the tool in prompt engineering principles. For the most accurate results, the tools translate your input into a more complex query for the technology to process. You can converse with CoCounsel using plain, conversational language and your lawyerly knowledge.
Additionally, CoCounsel uses processes like Retrieval Augmented Generation (RAGs) to help mitigate hallucinations, making the technology safer for legal work, and lessening the burden of prompt engineering.
What is Retrieval Augmented Generation (RAG)?
Free, generative AI available online typically uses large language models (LLMs) trained on unverified information. The tools look for patterns to feed their predictions, which can result in factually incorrect generalisations (or false case citations, made infamous in this New York legal case).
Conversely, CoCounsel Core is built on the largest LLM ever created and draws from trusted content. It uses the powerful generative AI technology of ChatGPT4 to extract accurate and reliable responses informed by trusted legal databases (including Westlaw and Practical Law). Using a model called Retrieval Augmented Generation (RAG), the technology only bases its responses on the vetted material it has been trained on, thereby mitigating inaccuracies.
Among its many offerings, CoCounsel Core can draw from its sources to help review texts, prepare depositions, search databases, compose legal memos, generate timelines and create summaries of large bodies of texts.
In summary, here is our take. Lawyers don’t need to become experts in prompt engineering, but knowing the basics will help them get more comfortable using generative AI.
