Skip to Main Content

Canadian Legal Research Manual

This reference work was created by the Lederman Law Library to support Queen’s students learning legal research skills.

Creating Prompts for Legal GenAI

prompt is the set of instructions you use to direct a generative AI tool. 

Think of it like the search terms you enter into Google—the words you use affect the webpages you see on your results list. The same is true for an interaction with genAI. The quality of the response will vary depending on the quality of the instructions you provide. 

The process of developing instructions for a genAI system to interpret and respond to is called prompt engineering. For legal research tasks, thoughtful prompts can help you avoid mistakes such as generating information about the incorrect jurisdiction or generic and unfocused responses.

This page provides information on some considerations for effective prompt engineering with legal research systems.

Note: Even with carefully developed prompts, genAI can provide unhelpful, ungrounded, and incorrect information. Don't forget to verify and evaluate all AI-generated information before relying on it for legal research. 

►See Critically Assessing AI-generated Content for more information.

Before You Start Developing Prompts

Before you start writing prompts, consider the following:

  • Do some reading first. In general, more background information leads to better prompts. It can be hard to develop effective prompts if you are new to an area of the law—you may not know what questions to ask or be able to catch hallucinations. Start your research by consulting a reliable secondary source.
  • Understand the privacy and security risks. GenAI tools have varied privacy and data security settings and options. Do not enter confidential or privileged information into a system unless you are confident that adequate security measures are in place. 
  • Consult the product documentation. No two genAI tools are alike, so prompting advice for one system may not work well in another. Look for product documentation on the specific tool you are using for optimal performance.
  • Provide citations for sources. When referencing sources in your prompts, always provide a full citation (instead of the style of cause for a case or the title of a statute). This improves the likelihood that the system will generate a summary of the correct document. 
  • Insist that the tool consider specific source types. Legal genAI does not always reference multiple source types in answering a question. For instance, it may only summarize cases on a topic, even though there is applicable legislation. If you notice that a type of primary source is absent from a response, you may need to redirect the system's focus.

The 5Ps Prompting Method

The 5Ps is a method for developing prompts using five categories. The following explanation of the 5Ps includes examples applicable to legal research.

This method is recommended by Lexis in use of Lexis+ AI. See Lexis' Prompting Best Practices tip sheet.

Note: not all systems are responsive to all 5Ps. For example, some are designed for very specific outputs (e.g. case summaries) and will not respond well to instructions that ask for other types of products. Often the only way to discover what works best for prompting a specific genAI tool is to work with it repeatedly and often. 

 

Iterate and Evaluate

Like any research, iteration is key to developing useful prompts. Don't forget to document your approach so you can easily retrieve or reuse your prompts at a later date.

►See Tips (Before You Start) for help organizing your research. 

Once you have generated text that you think will be useful in your research, you can proceed to evaluating and verifying it. 

Canadian Tools

These Canadian tools are available either freely online or via the Law Library:

Other Methods

There are numerous other frameworks used for prompting, generally outside of the law-specific context. For example:

Leo Lo, "The CLEAR path" (2023) 49:4 J Academic Librarianship 102720.

With CLEAR, the user should be:

  • Concise (be specific)
  • Logical (maintain a logical flow)
  • Explicit (e.g. about format)
  • Adaptive (be flexible)
  • Reflective (continue to adjust)