Years ago, I had the opportunity to work with LexisNexis on building “LexisNexis™ Portal powered by Plumtree” – an enterprise portal solution designed to create comprehensive knowledge management systems for law firms. Even then, one of our biggest challenges was achieving high accuracy and recall for search queries across vast collections of legal documents. It’s fascinating to see how this same fundamental challenge – finding the right information quickly and accurately – remains at the heart of today’s AI-powered search with RAG (Retrieval-Augmented Generation) technology.
Given this background, when I set out to research how enterprise companies are implementing RAG solutions, I was naturally curious to see how LexisNexis has evolved their approach in what could be considered the next generation of legal knowledge management. And what better way to understand the current state of the KM art than by comparing their solution to their primary competitor – Thomson Reuters’ Westlaw service?
That early challenge of achieving accurate legal document search has evolved into something I never could have imagined back in the Plumtree days. The legal profession has always been built on precedent, but in 2025, two of its most trusted institutions are busy rewriting the rules of legal research itself. Thomson Reuters’ Westlaw and LexisNexis aren’t just adding AI features to their platforms—they’re fundamentally re-imagining what it means to conduct legal research in the age of artificial intelligence.
The transformation is remarkable. If you’re still thinking about “AI-powered search” or even the basic RAG implementations we see today in many enterprise settings—you’re already behind. Both companies have leapfrogged beyond basic Retrieval-Augmented Generation into something far more sophisticated: agentic AI systems that don’t just find information, but actually conduct research the way skilled attorneys do. It’s a big leap from the search accuracy challenges we grappled with in the early knowledge management systems or even those built 2-5 years ago..
Beyond Search: The Rise of AI Legal Researchers
The transformation is striking. Just 2-3 years ago, legal AI meant glorified search engines that could pull relevant cases and statutes. Today, both Westlaw and LexisNexis are deploying AI agents that plan multi-step research strategies, follow leads between cases, and deliver comprehensive reports that mirror the work of experienced associates.
“This is not just another RAG tool or search interface,” explains Omar Bari, engineering lead from Thomson Reuters Labs, describing their new Deep Research system. “It’s a rethinking of how AI can perform within complex, high-stakes professional workflows.”
The numbers tell the story of this transformation. Where lawyers previously spent 10-20 hours on complex legal research, these new AI systems are completing sophisticated analyses in approximately 10 minutes—while maintaining the depth and rigor that legal work demands.
Two Paths to AI Excellence
What’s fascinating is how differently the two legal research giants are approaching this challenge. Their strategies reveal two distinct philosophies about the future of legal AI.
Westlaw’s Deep Research: The Specialist Approach
Thomson Reuters has bet big on creating AI that thinks like a lawyer. Their Deep Research system, launched in August 2025 as part of Westlaw Advantage, represents what they call “the legal industry’s first professional-grade agentic AI research capability” (see “Thomson Reuters Launches CoCounsel Legal with Agentic AI and Deep Research Capabilities, Along with A New and ‘Final’ Version of Westlaw”).
The system works more like a skilled research associate than a search engine. When given a complex legal question—say, analyzing discrimination claims where a company hired someone of the same race and gender as a rejected applicant—Deep Research doesn’t just return relevant cases. Instead, it:
- Generates a multi-step research plan that users can review
- Executes searches across Westlaw’s 20+ billion documents
- Follows citations and legal reasoning chains between cases
- Adapts its strategy based on what it discovers
- Delivers comprehensive reports with arguments on both sides
The technical sophistication is impressive. The system orchestrates specialized AI agents, each optimized for different types of legal documents. It maintains memory across long research sessions and uses Westlaw’s proprietary tools—KeyCite, Key Numbers, Precision Research—as building blocks for AI-driven analysis.
Perhaps most importantly, Westlaw has designed the system around what they call a “generation and verification loop.” The AI generates comprehensive research, but provides lawyers with the tools they need to quickly verify every finding through direct links to source material and highlighted excerpts.
LexisNexis’s GraphRAG: The Knowledge Web Approach
LexisNexis has taken a fundamentally different path, building what they call GraphRAG—a system that leverages knowledge graphs to understand the relationships between legal concepts, cases, and citations.
At the heart of their approach is the integration of Shepard’s Knowledge Graph (see “LexisNexis announces new capabilities for Lexis+ AI including RAG enhancements“), which maps the intricate web of how legal authorities relate to each other. This isn’t just about finding similar cases; it’s about understanding the deep connections that make legal arguments persuasive.
Using the Knowledge Graph is conceptually very similar to improving quality of legal content retrieval by categorizing corpus with multiple legal taxonomies e.g.:
- Practice areas (contracts, torts, criminal law)
- Document types (cases, statutes, regulations)
- Jurisdictions (federal, state, local)
- Legal concepts (liability, damages, jurisdiction)
Both create explicit connections between related legal concepts and authorities. But Shepard’s Knowledge Graph goes beyond traditional taxonomies by adding:
- Citation Relationships – the graph explicitly maps how cases cite each other, creating a web of legal authority that traditional taxonomies don’t capture.
- Dynamic Connections – in addition to static categories, it includes relationships like: which cases overrule others, which statutes are interpreted by which cases, or how legal principles evolve across jurisdictions
- Metadata Integration – as LexisNexis describes it, the knowledge graph incorporates “legal taxonomies, and additional metadata that enhance the quality and context of responses.”
Moving on, their Protégé AI assistant, launched to general availability in January 2025, combines three types of AI:
- Extractive AI for finding relevant information
- Generative AI for creating new content
- Agentic AI for autonomous task completion
But perhaps LexisNexis’s most interesting innovation is their dual-mode approach. In August 2025, they launched Protégé General AI, which lets users toggle between specialized legal AI (grounded in LexisNexis content) and general-purpose AI models like GPT-5 and Claude Sonnet 4 – all within a secure, enterprise environment.
This addresses a real problem: lawyers are already using consumer AI tools for brainstorming and drafting, often exposing confidential client information. LexisNexis provides a secure workspace where attorneys can access the latest AI models while maintaining privilege and confidentiality.
Table below captures both shared foundations and key differences between LexisNexis and Westlaw RAG implementations:
| Category | Shared Foundations | LexisNexis (Distinctives) | Westlaw (Distinctives) |
| Corpus | Both ground LLMs in closed, proprietary legal content (cases, statutes, regulations, secondary sources). | Corpus integrated with Practical Guidance + Shepard’s Knowledge Graph. | Corpus enriched by West Key Number System and editorial headnotes. |
| Retrieval method | Both use hybrid retrieval: lexical (BM25/keyword) + semantic (embeddings). | Dual-track retrieval pipeline (parallel lexical + semantic), re-ranked with Shepard’s authority signals + recency boost. | Editorial-first hybrid: retrieval steered by Key Numbers and head notes; semantic similarity augments but does not dominate. |
| Authority signals | Both integrate citators to elevate “good law” and suppress bad law. | Shepard’s in-flow: “At Risk” warnings, Authority summaries, and treatment narratives in the AI chat. | KeyCite: granular signals including “Overruled in Part,” visible via linked citations but outside the chat flow. |
| Generation | Both constrain LLMs to generate from retrieved, validated sources with linked citations. | Emphasis on in-flow validation UX, citations validated and surfaced inline. | Answers link to authorities; validation requires viewing KeyCite reports. |
| AI assistant | Both wrap retrieval + generation in AI assistants for research and drafting. | Protégé in Lexis+ AI: conversational assistant supporting research, drafting, summarization, uploads. | AI-Assisted Research + CoCounsel / Deep Research orchestration for multi-step queries |
| Hallucination profile | Both show non-zero hallucinations despite grounding. | 2024 study: hallucination ≈ 17%. Lower, but still significant. | 2024 study: hallucination ≈ 33%, highest among tested systems. |
| Knowledge graph/editorial backbone | Both enrich retrieval context. | Uses Shepard’s Knowledge Graph for authority-aware ranking and context assembly. | Uses West Key Number taxonomy to anchor lexical retrieval and ensure topical coverage |
What emerges from this analysis is that both companies are succeeding, but in different ways. Westlaw appears to have the edge in deep, specialized legal research capabilities—their Deep Research system represents genuine innovation in how AI conducts legal analysis. LexisNexis shows strength in flexibility and model diversity, with better measured accuracy and broader AI capabilities.
The Performance Reality Check
The competition isn’t just about features—it’s about accuracy. A 2024 Stanford University study revealed concerning hallucination rates across legal AI platforms: LexisNexis’s system produced incorrect information 17% of the time, while Westlaw’s AI research tools hallucinated more than 33% of the time.
- In Redo of Its Study, Stanford Finds Westlaw’s AI Hallucinates At Double the Rate of LexisNexis, June 4, 2024
- Hallucination- Free? Assessing the Reliability of Leading AI Legal Research Tools, March 14, 2025
Both companies have responded aggressively to these findings. LexisNexis has implemented what they call a “four-pronged approach” to accuracy, including automated metrics to detect hallucinations, human oversight, and transparent citations. Westlaw’s Deep Research system is explicitly designed to “eliminate errors and hallucinations” through direct citations from their authoritative dataset and sophisticated verification tools.
The improvement trajectory is encouraging. Both platforms now provide comprehensive citation trails, allow lawyers to quickly verify AI-generated insights, and integrate verification directly into the research workflow.
- Citation Context Now Available on Protégé, April 3, 2025
- Deep Research in Westlaw and CoCounsel: Building Agents That Research Like Lawyers, August 27, 2025
What This Means for Legal Practice
The implications extend far beyond making research faster. These systems are changing the fundamental economics of legal work.
Large law firms report that associates can now handle significantly larger caseloads while delivering more thorough analysis. Corporate legal departments are finding they can tackle complex regulatory questions without immediately turning to outside counsel. Solo practitioners gain access to research capabilities that previously required teams of junior attorneys.
But perhaps the most significant change is psychological. As one Morgan Lewis executive noted about Westlaw’s Deep Research: “This level of transparency is essential to maintaining the oversight and trust lawyers need to confidently adopt AI in practice.”
The systems aren’t replacing legal judgment—they’re amplifying it. They handle the mechanical aspects of research while freeing lawyers to focus on strategy, client counseling, and advocacy.
The Partnership Web
The technical infrastructure behind these advances reveals the complex web of AI partnerships reshaping the legal industry. Westlaw uses a multi-model approach, working closely with Anthropic (Claude 4 for orchestration), OpenAI, and Google, selecting different models based on task effectiveness.
LexisNexis has built a multi-cloud infrastructure spanning partnerships with Mistral, AWS (hosting Anthropic models), and Microsoft (hosting OpenAI models). Their model-agnostic approach lets them deploy the best AI for each specific use case.
The legal industry, often seen as a conservative laggard in technology adoption, has suddenly become one of the most sophisticated implementers of cutting-edge AI.
Looking Ahead: The Next Phase
Both companies are already testing what comes next. Westlaw is exploring multi-agent collaboration with lawyers “in the loop” and integration of firm-specific knowledge bases. LexisNexis is expanding their agentic capabilities and developing what they envision as “a portfolio of AI agents” specialized for different legal tasks.
The strategic alliance between LexisNexis and Harvey (the legal AI startup) suggests that the boundaries between legal research platforms and specialized AI tools are blurring. Meanwhile, Westlaw’s positioning of their August 2025 release as the “final” version of Westlaw suggests they’re moving to continuous AI-driven evolution rather than traditional software release cycles.
For legal professionals, this is unprecedented innovation. The pace of improvement in legal AI has accelerated dramatically in 2025, with both platforms delivering capabilities that seemed futuristic just months ago.
The Bottom Line
The success of both Westlaw’s specialized approach and LexisNexis’s flexible strategy suggests that both are pushing the boundaries of what’s possible, creating tools that make legal work and knowledge management in general more efficient, more thorough, and more accessible.
For lawyers, the message is clear: the future of legal practice isn’t about competing with AI—it’s about learning to work with AI partners that can handle the mechanical aspects of research while amplifying human judgment and expertise.
Igor's Musings About Technology 