Gen AI for Legal Use Cases: Specific Over the General?

As an industry grounded in natural language, the legal industry is ripe for transformation by generative AI. A recent LexisNexis study revealed that most lawyers agree AI will impact the legal profession and almost 50% believe it will significantly transform the practice of law.[i]

With so many emerging solutions, law firms are in experimentation mode, seeking tools that will have the highest impact for their practice. Many of these solutions focus on a specific problem, like legal research, contract analysis, or discovery. While others provide broader functionality, allowing users to experiment with their own problems on their own terms. The latter category holds great promise for lawyers with an interest in this new technology, who want to make AI work for their workflows.

Putting this theory to the test, I have been speaking with lawyers across a range of practice groups about their interactions with generative AI tools and the problems they think it could solve. I then enlisted Claude (Anthropic’s generative AI assistant) and GPT (OpenAI’s chatbot) to test some solutions.

The results . . . these LLMs are remarkably proficient at some routine legal tasks! Equipping lawyers with the ability to experiment with LLMs at work could produce huge value for law firms and automate some of the more mundane tasks that consume lawyers time.

AI Legal Drafting Companions

As generative AI tools like GPT are trained on immense quantities of human-generated text, they are highly skilled at analyzing and emulating human language. One study found that Chat GPT reduced the time taken to complete professional writing tasks by 40%, while boosting quality by 18%.[ii] These tools can act as author or editor and can be told what writing style or tone to adopt. While they will not always produce output that can be copy-pasted into a final work product, they offer a productive way to iterate on and refine written work.

These tools are particularly proficient at routine legal writing tasks. In the below example, I asked GPT and Claude to weigh in with some objections to a set of discovery requests. The responses closely tracked typical objections raised by lawyers, picking up that the requests were over-broad and explaining why.

AI tools can also provide powerful support when it comes to brainstorming ideas or navigating new tasks. This holds especially true in a world that embraces remote work; having an AI assistant can be a crucial resource where junior associates can no longer reliably find a mentor in the next office. These tools can provide ideas, answer questions, or suggest a structure for written work. In the next example, I asked Claude how I might go about drafting a motion requesting extension of a deadline.

Here, Claude not only provides a checklist list of points to cover, but underscores what the focal point of the motion should be. For a junior associate, this guidance could immensely improve a first draft and cut down the versions passed back and forth with their senior colleagues.

Data Extraction From Legal Documents

One capital markets lawyer I spoke with suggested that generative AI could be an effective aide for financial analysis. This can entail combing through hundreds of pages of financial documents to extract key terms. She recounted a recent late night spent on this exercise, piecing together a detailed table to analyze the financial status of an entity her client was considering transacting with. She thought AI could help with this process as surfacing the key terms from these documents is more about attention to detail than legal expertise. To test this theory, I asked Claude to summarize the key terms of several financial documents.

A note on this process: at around 150 pages, each of the documents exceeded the amount of information Claude (or GPT) can process in a single prompt. As I was accessing Claude through Credal, I could include the full documents with my queries as Credal worked behind the scenes to divide documents into segments Claude could handle. Another way to get around this issue is to use shorter excerpts from a document. This is particularly helpful if you know which part of a document has the information you want to extract.

My experiment went as follows:

At this point, I asked Claude to fix the formatting, as the document title was listed in each cell, rather than as a heading.

The result was impressive! In just five minutes, I had a table summarizing the key terms of each document, which I could paste into excel. I reviewed the responses—an essential step when working with generative AI systems, and good practice generally. I asked Claude to elaborate on the “subsidiaries” listed as guarantors to the June 10 indenture, as I felt that response was incomplete, and it responded with a full and accurate list. I then tasked it with providing some more complex information, which it accomplished in seconds.

There are many and varied ways to approach this exercise. Here, I managed to quickly pull information from a set of lengthy documents using simple prompts. By using generative AI for this task, the lawyer’s job becomes reviewing the output, following up with additional queries, and supplementing the information where necessary. This can save considerable time and free up lawyers to focus on analysis and strategy.

Summarizing Legal Developments

Summarizing legal developments can be a time-consuming and often time-pressured task. Lawyers are often tasked with promptly analyzing an unexpected legal opinion or the latest developments in the law. In some cases the information to be condensed can be significant. For example, following a major deadline in a multi-party litigation, clients will want to understand the arguments raised by the numerous parties to the case and will look to their lawyers to explain the voluminous briefing. In such cases, lawyers must divert their attention to quickly understand and explain the information for their clients.

Sometimes it can feel that these tasks need to be completed faster than humanly possible. Thankfully, AI tools can do just that. They can quickly condense large quantities of information and generate a first draft faster than any human could. By using these tools, rather than starting with a blank page, lawyers can have a leg-up on the process in the form of a well-organized first draft.

In this example, having read the opinion, I prompted GPT to provide additional relevant information that I felt was important to highlight. I could then edit this information and add my own analysis. A productive process for working with GPT to produce a summary looks something like this:

  1. Read and understand the document;
  2. Ask the AI tool to generate a summary;
  3. Ask further questions in areas you think could be better explained, have been left out, or are inaccurate;
  4. Edit the output to reflect your own voice and style; and
  5. Add your expert analysis.

Crucially, this process will conserve lawyers’ time for analysis and strategy, where they can add the most value for clients.

The Verdict

These tools showed impressive ability to handle the tasks I assigned to them. As these examples show, they could shave hours off routine legal tasks and improve work product. In all cases, the use of AI tools should be paired with the attorney’s own judgment, expertise, and knowledge of their cases. Of course, attorneys should never delegate tasks to these tools entirely or reproduce the responses without verifying their accuracy and ensuring compliance with applicable rules. But the downsides of thoughtful and secure use of these tools are few, while the potential benefits are significant.

That said, they don’t have magical abilities. I didn’t manage to solve all the ideas put to me by the lawyers I spoke with and it took some practice to frame useful prompts, understand the models, and get to the right output. Spending some time up-front sharpening your skills is a great investment if you are aiming to use these tools to speed up your workflows. For me, keeping track of my previous requests and reviewing my audit logs was a great reminder of where I was going right and what I was doing wrong. Questioning, correcting, and requesting additional output from these models can also help to improve output.

The concepts explored here—drafting, data extraction, and summarization—have applications far beyond the four corners of this blog post. There are vastly more ways to leverage AI tools in legal work, some of which I hope to return to in a later post.

Additional Benefits of General AI tools for Law Firms

Introducing a general AI tool into a law firm can do more than expand the ability to use AI for legal tasks. It could prove beneficial for other professionals, such as secretaries, marketing teams, and human resource teams.  

It can also help to diversify the firm’s AI strategy. AI innovation is happening fast and it’s not yet clear which LLM provider will emerge on top. OpenAI grabbed the public’s attention in late 2022 with ChatGPT and cultivated strong brand awareness. But other players including Anthropic’s Claude, Meta’s Llama, and Google’s Bard are very much in the race. Google has also threatened further disruption with the announcement of Gemini, its forthcoming multi-modal AI which leverages Google’s extensive in-house research teams and has access to training data from various Google services like Google Scholar and Google Books. Amid the race for AI dominance, banking on a single provider (or tools built upon a single provider) is not a safe bet. Exploring different models or adopting a system like Credal, that integrates with multiple AI models, will allow law firms to remain nimble and seize upon the latest advances.  

Experience with these tools and finely honed prompting skills will soon be required skills for junior lawyers. By encouraging use and experimentation under appropriate conditions, law firms can take steps to future-proof their business.


[i].  LexisNexis, International Legal Generative AI Report, Detailed Survey Findings (August 22, 2023) (the study was conducted between March and July, 2023.)

[ii].  S. Noy and W. Zhang, Experimental evidence on the productivity effects of generative artificial intelligence, Science (2023).

Building blocks towards secure AI apps

Credal gives you everything you need to supercharge your business using generative AI, securely.

Ready to dive in?

Get Started