Legal document review · Production

Deep Research in Westlaw and CoCounsel: Building Agents That Research Like Lawyers

The problem

Westlaw's existing AI-Assisted Research using LLMs produced helpful narrative summaries from cases, statutes, and regulations, but these were only a starting point; attorneys still had to manually dive deeper to gather authority, surface edge cases, and resolve nuanced legal questions that require iterative multi-step investigation.

First attempt

Traditional RAG-based systems could produce a surface-level answer but could not handle multi-step legal reasoning — they answered the question without assessing the strength of the inference, jurisdictional variation, or competing arguments.

Workflow diagram · grounded in source
1
Initial legal question submitted
trigger
“Plan multi-step strategies from the initial question”
2
Multi-step research plan generation
ai_action
“The system generates multi-step research plans and explores alternative theories when its first approach fails”
3
Parallel research execution
ai_action
“In Deep Research, the system can search, open documents, analyze content, and react in real time”
4
KeyCite citation validation
validation
“It uses KeyCite warnings to flag questioned or overruled cases”
5
Citation breadcrumb following
ai_action
“When a case references others, the system follows those breadcrumbs and adjusts its strategy, just as a skilled researcher would”
6
Memory and findings integration
ai_action
“The system maintains memory across the session, tracks explored arguments, relates findings, and integrates parallel research threads into a coherent report”
7
Verified report with inline citations
output
“The main report includes inline citations, KeyCite flags, and excerpts to support fast verification”
8
Lawyer verification
human_review
“In Deep Research, the system generates; the lawyer verifies. We've built tools to accelerate this loop. All citations in Deep Research come from authoritative Westlaw sources. The Sources tab presents citations visually, with direct link…”
Reported outcome

Deep Research is now live in Westlaw Advantage and CoCounsel, providing agentic legal research that plans, executes in parallel, validates citations, and produces verified reports — with comprehensive analyses completing in about 10 minutes.

Reported metrics
Time to complete comprehensive analysesabout 10 minutes
Reported stack
Claude 4KeyCiteWestlawCoCounsel
Source
https://medium.com/tr-labs-ml-engineering-blog/deep-research-in-westlaw-and-cocounsel-building-agents-that-research-like-lawyers-508ad5c70e45
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Deep Research is now live in Westlaw Advantage and CoCounsel, providing agentic legal research that plans, executes in parallel, validates citations, and produces verified reports — with comprehensive analyses complet…

What tools did this team use?

Claude 4, KeyCite, Westlaw, CoCounsel.

What results were reported?

Time to complete comprehensive analyses: about 10 minutes (source-reported, not independently verified).

What failed first in this deployment?

Traditional RAG-based systems could produce a surface-level answer but could not handle multi-step legal reasoning — they answered the question without assessing the strength of the inference, jurisdictional variation…

How is this legal document review AI workflow structured?

Initial legal question submitted → Multi-step research plan generation → Parallel research execution → KeyCite citation validation → Citation breadcrumb following → Memory and findings integration → Verified report with inline citations → Lawyer verification.

Related legal document review cases