Retrieval-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.
Last reviewed: 2026/05/19
An embedding is a numerical vector representation of text — such as a word, sentence, or document — produced by a machine learning model, enabling AI systems to measure semantic similarity between texts and retrieve relevant information.
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 / ModelA 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.
Tech / ModelVector search is a retrieval method that finds documents semantically similar to a query by comparing numerical vector representations (embeddings) rather than exact keyword matches, enabling natural language queries to surface conceptually relevant results.
AI-powered legal research with citation-validated answers from Westlaw.
Conversational legal research with real-time Shepard's citation validation.
Thomson Reuters' GPT-backed legal research and drafting with Westlaw integration (relaunched as CoCounsel Legal, 2025).
AI legal research pioneer (CARA AI); standalone retired 2025, its technology now powers Thomson Reuters CoCounsel.
The most expensive legal AI in the market — Am Law 100 firms only.
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.
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.
Retrieval-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.
RAG is the primary technical mechanism that distinguishes legal AI tools capable of grounded legal research from general-purpose chatbots that generate text from training data alone. For lawyers, understanding this distinction matters because it directly affects how much to trust a tool's output.
A general-purpose LLM answering a legal question draws on patterns learned during training — which may be months or years out of date, and which may have included unreliable sources alongside authoritative ones. A RAG-based legal research tool, by contrast, retrieves from a specific, curated, and continuously updated legal database (case law, statutes, regulations) and generates its response based on those retrieved documents.
The practical difference: a RAG-based tool citing a specific case should be able to show you exactly which document it retrieved. A non-RAG tool presenting the same citation may have generated it from statistical patterns, making fabrication more likely.
For lawyers evaluating legal AI tools, asking how the tool sources its answers — and whether it can cite the specific retrieved documents — is a key due diligence question. RAG does not eliminate hallucination, but it significantly reduces the risk by grounding generation in verifiable sources.
Most leading legal research tools use RAG architectures to varying degrees. Westlaw Precision AI grounds its AI answers in Thomson Reuters' continuously updated legal database, with citations linking back to specific cases and statutes. Lexis+ AI operates similarly on the LexisNexis content corpus, providing source attribution with responses.
The retrieval component of RAG relies on embedding models that convert text into numerical vectors and vector search to identify the most semantically relevant documents in the corpus for a given query. The quality of the retrieval step directly affects the quality of the generated answer — if the retrieval fails to surface the most relevant cases, the generated answer may be incomplete or misleading even if it accurately reflects the retrieved (but imperfectly chosen) sources.
CoCounsel and similar tools have invested heavily in both retrieval quality and model grounding to reduce hallucination rates in legal research contexts.
Tools that do not use RAG — applying a general-purpose LLM directly to legal questions — are more prone to hallucination and less suitable for tasks requiring citation accuracy.