A large language model (LLM) is an AI system trained on large volumes of text data to predict and generate human-like text; it serves as the core engine underlying most legal AI tools for research, drafting, and document analysis.
Last reviewed: 2026/05/19
Fine-tuning is the process of further training a pre-trained large language model on a domain-specific dataset to improve its performance on tasks in that domain, such as legal document analysis, contract drafting, or jurisdiction-specific research.
Tech / ModelHallucination in legal AI refers to instances where an AI model generates factually incorrect, fabricated, or unsupported output — such as nonexistent case citations, invented statutes, or inaccurate summaries of legal holdings — presented with apparent confidence.
Tech / ModelRetrieval-Augmented Generation (RAG) is an AI architecture that combines a retrieval system — which fetches relevant documents from a specified corpus — with a generative language model that produces answers grounded in those retrieved documents, rather than relying solely on the model's training data.
Tech / ModelTraining data is the corpus of text and examples used to train a large language model, establishing its capabilities, knowledge, and limitations; the quality, recency, and composition of training data directly affects the model's reliability for legal tasks.
The most expensive legal AI in the market — Am Law 100 firms only.
Thomson Reuters' GPT-backed legal research and drafting with Westlaw integration (relaunched as CoCounsel Legal, 2025).
AI-powered legal research with citation-validated answers from Westlaw.
Conversational legal research with real-time Shepard's citation validation.
AI contract drafting and review inside Microsoft Word for transactional lawyers.
Move from this definition to role-based legal AI shortlists and the selection criteria that matter for each type of legal team.
Am Law 200 and global firm workflows: accuracy at scale, security compliance, and matter-level auditability.
Reviews built for 2–20 attorney firms: collaborative workflows, mid-range budgets, limited IT overhead.
60 legal AI tools vetted for the solo lawyer: tight budget, no IT team, billable-hours pressure.
Last reviewed: 2026/05/19. Definitions are written by the LawyerAI Editorial team. Commercial relationships are disclosed and do not determine editorial scores or conclusions. See our Sponsorship & Affiliate Disclosure.
A large language model (LLM) is an AI system trained on large volumes of text data to predict and generate human-like text; it serves as the core engine underlying most legal AI tools for research, drafting, and document analysis.
LLMs are the technical foundation of the current generation of legal AI tools. Understanding what an LLM is — and what it is not — helps lawyers calibrate their reliance on AI-assisted work product.
An LLM does not "know" the law the way a lawyer does. It learns statistical patterns from training text, which enables it to generate text that resembles authoritative legal analysis. It does not reason from principles; it predicts likely text. This distinction matters when an LLM produces a confident-sounding answer about a legal question: the confidence reflects pattern-matching, not verified accuracy.
For lawyers, the practical implications are several. LLMs perform well on tasks where the correct output resembles patterns in the training data — drafting standard commercial clauses, summarizing documents with clear structure, explaining well-established legal doctrines. They perform less reliably on tasks requiring precise factual recall (exact citation text), novel legal reasoning, or jurisdiction-specific analysis not well represented in training data.
The legal AI market is built largely on top of foundational LLMs (GPT-4 family, Claude, Gemini, and others) with varying amounts of legal specialization, fine-tuning, and retrieval augmentation layered on top. The base model matters, but the legal-specific engineering applied to it often matters more for task-specific performance.
Most legal AI vendors do not train their own foundational LLMs — the compute and data requirements are prohibitive. Instead, they build on top of foundation models from Anthropic, OpenAI, Google, Meta, or others, applying legal-specific fine-tuning, prompt engineering, and retrieval augmentation to improve legal task performance.
Harvey AI, for example, is built on top of OpenAI's models with legal-specific tuning and integration capabilities. Cocounsel applies GPT-4 architecture with Casetext's legal research infrastructure. Westlaw Precision AI and Lexis+ AI integrate foundation models with their respective legal content databases through RAG architecture.
The degree to which a tool discloses its underlying LLM and model architecture varies. Some vendors are transparent about the base model; others treat this as proprietary. Understanding the architecture helps lawyers assess hallucination risk and data privacy implications — knowing which vendor processes the data that passes through the LLM.
No LLM produces perfect legal output. All require attorney verification before the output is used in client work.