Generative AI Tools and Resources for Law Students

This section introduces five categories of generative AI tools that law students and researchers may encounter in academic or professional settings. Whether you are just beginning to explore AI tools or already using them in your research or writing, this guide offers context for understanding how these systems work and what they can (and cannot) do.

As of March 2025, there are more than 200 legal AI tools on the market. This guide is not intended to be comprehensive. Instead, it focuses on tools that (1) are available to UC Davis Law students, (2) offer free access or educational discounts, or (3) are generating significant discussion in the legal industry.

The categories used (foundational, specialized, and agentic) are not fixed. Many tools span multiple categories or shift as new capabilities are added. In particular, so-called agentic features (tools that plan, adapt, or carry out multi-step tasks) are being widely promoted as the next major development in legal AI. While these claims often outpace current capabilities, the concept is shaping how many vendors describe their tools.

Foundation models are large, general-purpose AI systems trained on a wide range of internet and published text. These models, such as ChatGPT, Claude, and Gemini, can generate text, summarize content, translate language, and answer questions across many topics. They are often used in law school to rephrase complex language, brainstorm ideas, or draft early outlines. However, these tools are not connected to legal databases and may produce content that is inaccurate or misleading. Always verify their outputs using trusted sources.

Reasoning and Deep Research Capabilities

Some foundational models now include features described as reasoning or deep research tools. These capabilities extend the basic functionality of a language model by enabling it to perform more complex, multi-step tasks.

Reasoning tools are designed to walk through problems step by step, explain logic, or connect ideas across multiple inputs. While all LLMs can simulate reasoning to some extent, some tools now use structured approaches, such as chain-of-thought prompting or tool-assisted workflows.

Deep research tools allow the model to search external sources, gather relevant information, and synthesize findings into longer, more detailed outputs. These tools often combine real-time search, document retrieval, and summarization in a single process. Examples include Perplexity Pro, Deep Research in ChatGPT, and Deep Research in Gemini.

These features reflect a shift from one-prompt interactions to more autonomous research-style processes. They can be useful in complex or unfamiliar subject areas, but the outputs should still be reviewed and verified against reliable sources.

These AI tools have been customized for legal tasks, typically using two main approaches: fine-tuning and retrieval-augmented generation (RAG). Proprietary generative AI tools like Protégé/Lexis+ AI and CoCounsel use one or both of these strategies. They tend to perform better in legal contexts than general-purpose AI tools and can help with research, drafting, and summarizing. However, their output still requires human review.

Fine-tuning means the model has been retrained on legal information (e.g., case law, statutes, or legal writing) so it “learns” patterns and vocabulary from within the legal field. This process permanently alters the model to make it better suited for legal applications.

RAG, by contrast, does not retrain the model. Instead, it allows the AI to pull relevant information at runtime from a trusted external source (such as a legal research database) before generating a response. This is intended to improve accuracy and reduce hallucinations, especially when the retrieved content comes from authoritative sources of law. 

Agentic models are an emerging class of AI systems designed to perform multi-step tasks with limited autonomy. Unlike generative tools that respond to single prompts, agentic models can plan sequences of actions, adjust based on intermediate results, and interact with tools or documents to pursue a defined goal. The shift is from one-off assistance (e.g., “summarize this case”) to broader task support (e.g., “draft a summary, extract key holdings, and prepare a comparison chart”).

Common features include:

• Task planning: Breaking down a request into logical steps.
• Tool integration: Accessing databases, APIs, or uploaded files while working.
• Iteration and adaptation: Revising or retrying steps when results fall short.
• State tracking: Remembering prior inputs or context to guide current actions.

That said, the term “agent” is being used in wildly inconsistent ways across the tech industry. As this TechCrunch article bluntly points out, there’s no shared definition of what counts as an AI agent. Some tools labeled as agents run a series of scripted steps. Others make dynamic decisions, use multiple tools, or act on user behalf. This ambiguity makes it harder to evaluate what any given product actually does, especially in legal contexts where precision matters.

Legal tech companies like Harvey, Paxton, and LawDroid have started integrating agentic components into tools for regulatory mapping, contract analysis, and cross-jurisdictional research. These tools still require user input and oversight, but they mark a shift in how AI can support legal workflows.

As with many AI developments, the hype around agentic models has grown fast. Some marketing claims outpace the tools' actual capabilities. It’s worth asking what “agentic” really means in context and whether a tool’s features support that label. Still, the technical progress is real, especially for students and researchers exploring automation, legal operations, or AI-assisted legal practice.

Custom-built tools use generative AI technology to serve a specific organization, such as a law firm, court, or legal clinic. These systems are often trained on private data and tailored to match internal workflows or document standards. For example, a firm might develop a tool that drafts contracts using its preferred structure, or a court might deploy an AI system to help manage case assignments. These tools are not typically available to the public and may be difficult to evaluate from the outside.