Overview
Casetext was the maker of CoCounsel, the first AI legal assistant built on GPT-4, and was acquired by Thomson Reuters for $650 million in 2023. The product now lives on as CoCounsel Legal within Thomson Reuters, integrating with Westlaw, Practical Law, and Microsoft 365. It helps legal professionals complete research, drafting, and document analysis with speed and precision via agentic AI workflows.
Our take
Casetext, now CoCounsel Legal, provides AI legal research and drafting with quote-based pricing. It is grounded in a real case-law database, which is the property that matters in this category — a legal AI answering from the model alone is how lawyers end up sanctioned for citing cases that do not exist.
Worth it if
Law firms wanting research and drafting assistance grounded in a verifiable legal database.
Look elsewhere if
Anyone who would skip verifying citations — the duty to check does not transfer to software.
✨ Features
Deep Research
Builds a multistep research plan using AI grounded in trusted Westlaw and Practical Law content.
Document Analysis
Reviews thousands of documents at once to surface key facts, issues, and clauses.
End-to-End Drafting
Drafts and edits legal documents within Microsoft Word with citation validation.
Agentic Workflows
Runs expert-created prompts and automated processes for litigation and advisory tasks.
⚖️ Pros & Cons
Pros
Cons
💰 Pricing
CoCounsel Legal
Custom
- AI legal assistant
- Research, drafting & document analysis
- Quote based on firm size
With Westlaw / Practical Law
Custom
- Integrated trusted legal content
- KeyCite citation validation
- Contact sales for bundle pricing
Enterprise
Custom
- Org-wide deployment
- Microsoft 365 & DMS integration
- Tailored agentic workflows
Categories
❓ Frequently Asked Questions
Does database grounding eliminate hallucinated citations?
It reduces them substantially, because answers cite documents that exist and can be opened. It does not remove the obligation to confirm each citation says what the brief claims — misattribution is more common than fabrication in retrieval-based systems and considerably harder to spot.
What does a legal AI actually do well?
First-pass research, summarising long records, comparing clauses across agreements, and drafting standard documents from templates. What it does not do is exercise judgement about strategy or advise a client. In most jurisdictions that distinction is also a licensing one, not merely a practical one.