If I were a law student today ...

If I were a law student today, my approach to studying would be fundamentally different from any generation that came before. The reason is simple: Artificial Intelligence. In 2025, AI is no longer a novelty; it's an essential tool for learning the law and, more importantly, for practicing it. Any student not intensively using AI, where permitted, is putting themselves at a significant disadvantage. They are not only making their studies harder than necessary, but they are also failing to learn the tools that will be central to their careers.

The legal profession is already adopting AI for tasks like research, document review, and drafting. Recent ABA data indicate that roughly one‑third of lawyers were using generative AI for some work function by late 2024, nearly triple the prior year’s share. I suspect the real number is higher; antiquated policies based on exaggerated criticisms of older versions of AI likely make many lawyers unwilling to admit to their use. Many law firms are therefore looking for graduates who are not just familiar with these tools but proficient in using them. To ignore AI in law school today is like a student in the 1990s refusing to use a computer or the internet. It's a choice to be less effective and less prepared.

This guide outlines how I think I would approach law school in the age of AI. I can't say I'm certain of every detail of what I write here today – I haven't been a law student for 42 years and my aging vision from the other side of the podium may not fully reflect today's full reality. Moreover, the world of AI changes so fast that the details of what I include today may be supplanted by technological developments. But I am sure that use of AI is not about finding shortcuts or cheating. It’s about using powerful tools to learn more deeply, manage the overwhelming amount of information, and prepare for the realities of being a lawyer in the 2020s and beyond.

My Core AI Toolkit

First, at least as of the summer of 2025, I would immediately purchase paid subscriptions to the premium versions of the major AI models: ChatGPT Plus (powered by GPT‑4o, about $20/mo) and Google’s Gemini Advanced (about $20/mo or included in the Google One AI Pro bundle). Both plans give access to very large context windows (up to 128k tokens in GPT‑4o and 1M tokens in Gemini Advanced), faster throughput, and multimodal capabilities. $500 a year is a smart investment and a small increment to today's cost of attendance. It's not much more than the price of many study aids students traditionally purchase. Of course, it would be even better if enlightened law schools would try for site licenses that would reduce the marginal cost of use to zero and place all students on an even playing field.

The organizing principle of my entire study system would be Google’s NotebookLM. I would be highly granular in my organization. Rather than one giant notebook for "Torts," I would have a new one for each class session or major topic, titled with a clear convention like "Torts: Class 23," "Criminal Law: Mens Rea," or "Evidence: Week 6." This structure keeps the AI's responses focused and highly relevant to a specific part of the syllabus. It also can act as a kind of file system for materials.

For each class session, I'd scoop up every resource I could get my hands on and load it into the course-specific notebook: casebook PDFs, statutes, my own lecture notes, law-review articles, and any digital materials the professor circulates. This includes exporting casebook chapters when DRM allows, downloading Zoom or AI Companion transcripts of class sessions, and converting any PowerPoint slides to PDF. If I found my material too sparse, I would use NotebookLM's "Discovery" feature to pull in additional relevant sources from the web. Everything goes straight into the appropriate notebook so it becomes a fully grounded expert on that day’s material.

A Strategic Workflow for AI Study

With my toolkit in place, my strategy would be a three-step process: first, organize my information; second, use built-in AI features for broad analysis; and third, build my own custom tools for specialized, repetitive work.

1. Use the "Thinking" Models First

For any serious legal analysis—understanding a case, outlining an argument, or preparing for an exam—I would exclusively use the most advanced models available (e.g., GPT-o3, Gemini 2.5 Pro). These "thinking" models are slower but provide much more accurate and nuanced answers. The faster, "flash" models (e.g. GPT-4o) are fine for quick definitions, but they are not as reliable for the complex reasoning required in law. Waiting a minute for a high quality answer is a first world problem law students should be able to handle

2. Leverage Built-in Features for Quick Analysis and Drafting

Modern AI platforms have powerful built-in features that are perfect for first-pass analysis and document creation.

• In NotebookLM, I'd use the one-click buttons to instantly generate study guides, briefing documents, timelines, and FAQs from the materials I've uploaded. I'd for sure generate synthetic podcasts that would provide a different learning modality (one that suits me well). In Gemini, I would use its Canvas feature to create interactive study tools, like quizzes or even mini-websites that explain a concept.
• In ChatGPT, I would use the chat interface as an iterative writing partner. I can draft a passage, then give it commands like "make this sentence less convoluted" or "adjust the tone to be more formal," using the AI to perform global changes and polish my work.
• Build Custom Tools for Repetitive Tasks, When built-in features aren't enough, I would create my own specialized assistants using OpenAI's CustomGPTs and Google's Gems. These features allow you to give the AI a specific role and a dedicated knowledge base, which is similar to a technique called Retrieval-Augmented Generation (RAG). Essentially, it forces the AI to rely heavily on your documents when answering, making it an expert on your specific subject.

Here are excellent tutorials on how to create them:

• How to Create a CustomGPT
• How to Create a Gemini Gem

I've built several of these, including a 1L Case Briefer, a Black Letter Law generator, a law student sentence editor, an essay exam generator, a legal multiple choice question generator, a "prepare like a law professor" system, and a cranky system that subjects whatever you write to relentless criticism. Building these tools not only saves time but also forces a deeper understanding of the material. I'd even trade the tools I build with friends in my study group who had also seen the light, creating a shared library of powerful study aids.

Staying Smart: Avoiding AI's Traps

Minimizing Hallucinations

AI models (like the humans on whose work they were trained) can "hallucinate"—that is, make things up. This is a well-documented problem, and, much to the delight of those predisposed to hate AI while exempting humans from similar scrutiny, lawyers have been sanctioned for submitting briefs with fake case citations generated by AI. My strategy to avoid AI error follows a three-step ladder:

1. Gemini’s Google Button: Before relying on any Gemini answer, I always tap the built-in "G" button. Gemini launches a live Search that colour-codes text it can corroborate (green) or dispute (orange), giving me a fast reality check with virtually no extra effort.
2. Cross-Model Check: Feed the question and answer from one AI into another and ask it to look for mistakes. While it’s possible both AIs could make the same error, this cross-check eliminates many of them.
3. Traditional Verification: For any high-stakes matter (like a submission to a court or client), I would verify any remaining questions in Westlaw or Lexis.

These systems are unlikely to reduce errors to zero. But used intelligently, AI is likely to generate fewer errors than the humans for which they serve as a partial substitute.

Avoiding AI Dependence

The point of law school in 2025 is not to have a digital archive of AI's thoughts; it is to get the information and analytical framework into your own head. We are not quite yet at the point where AI lawyers battle each other in front of an AI judge. To address contemporary necessities, then, I would force myself to work without AI regularly.

• Write essays and practice exams without AI assistance. This is non-negotiable. You need to be able to perform under exam conditions without a digital assistant. Some clients may be willing to suffer a "hold on, let me put it into ChatGPT" response to a question, but few will be willing to tolerate many such responses without wondering exactly why they are paying the lawyer.
• Use the AI's voice features for active learning. Instead of just typing prompts, I would use ChatGPT’s voice mode or Gemini Live to have conversations with the AI. Explaining concepts back to the AI in my own words is a powerful way to simulate a Socratic dialogue and solidify understanding. As I will discuss in a future blog post, it's also a great way to develop oral skills still needed in legal practice such as motion practice, depositions, and negotiations.

A Note on Law School Policies and Exams

In 2025, any law school professor who bans the use of AI for studying is doing their students a disservice. It is an obstacle to learning and leaves students unprepared for a profession where AI is already becoming standard and tools such as Harvey, Vincent are entering broad use and legacy products such as Westlaw and Lexis are embracing AI as well. As a student, I would politely but firmly lobby for policies that embrace these tools for learning while maintaining academic integrity.

Speaking of integrity, it is irresponsible for faculty to give open-book exams without proctoring. Honor codes are important. However, law-school grading is so high-stakes—where the jump from the top 11 percent to the top 10 percent of the class can translate into hundreds of thousands of dollars in lifetime earnings—that the temptation to use AI for cheating turns the mixture into a toxic one. Relying on an honor code alone in this environment is naive.

A Note on Fun

There's no crime in having occasional fun in law school. So after a few hours of slogging through secured transactions, I might use AIs like Suno or Udio to generate memory-enhancing songs to help me memorize material (see the fabulous work of a former student here). Or I might create artwork to capture the essence of key cases. Here's one a former student produced in connection with a health law course.

Or if I felt like spending some money or credits, I might even use Google's veo3 to produce a little video. Here's one illustrating the rather horrifying scenario known in contracts as "Williston's Tramp."

Conclusion

If I were a law student today, I would be an AI power user. I would subscribe to the best tools, build my own custom assistants, and integrate them into every part of my study process. This isn't about being lazy; it's about being smart, efficient, and engaged. A conversation with a very informed and patient (artificial) colleague is exactly the sort of individualized Socratic dialogue that legal education was supposed to generate but that in 2025 seldom does. The goal is to use AI to augment my own intelligence to focus on the difficult and interesting parts of learning to "think like a lawyer." The students who master these tools will not only do better in school but will also enter the legal profession, be it in biglaw, small law, or public interest, with a significant advantage.

Here's a NotebookLM-generated podcast derived from this blog post. (By the way, see if you can spot the hallucination it contains 😀)