Comparisons
CoCounsel vs Lexis+ Protégé: The Research Incumbents' AI Compared
CoCounsel vs Lexis+ Protégé in 2026: KeyCite against Shepard's, a Claude Agent SDK rebuild against a multi-model harness, prices, jurisdictions, when to switch.
On this page
On 3 February 2026 one Anthropic blog post cut 18% from Thomson Reuters and 14% from RELX in a single session; within a year both had rebuilt their research assistants as agents. CoCounsel Legal is the pick if you already hold Westlaw and want the deeper drafting agents; Lexis+ with Protégé if you hold Lexis and want configurable workflows with customer-held keys. Switching for the AI alone rarely pays.
Why the parents matter more than the products
Both products are answers to the same threat. When Anthropic announced its first legal plugin for Claude Cowork, Thomson Reuters fell 18% intraday, RELX had its steepest fall since 1988 and Wolters Kluwer lost 13%. Investors were pricing the idea that a $20-a-month frontier model could do what the publishers charge hundreds of dollars a seat for. Within three weeks both Harvey and LexisNexis had announced Anthropic integrations, and the incumbents' counter-argument crystallised into a phrase Thomson Reuters' Joel Hron used: "trusted content plus governance".
The products that resulted look alike from a distance. Thomson Reuters made the "next generation of CoCounsel Legal" generally available in the US on 20 August 2026, built on Anthropic's Claude Agent SDK plus its own legal model. LexisNexis retired Lexis+ AI and launched Lexis+ with Protégé on 24 February 2026, then on 24 August 2026 added the Legal Intelligence Engine, which LexisNexis CTO Greg Dickason calls "an agentic harness" that picks models, sources and skills for each task.
Neither is a walled garden any more. CoCounsel is a partner connector inside Claude for Legal, LexisNexis is one of the 20-plus connectors, Thomson Reuters is among Astra for Law's 26 partner plugins, and RELX invested in Harvey in June 2025. The comparison is therefore not "which chat box" but "which library, citator and control layer do you want your agents grounded in". The CoCounsel Legal review and the Lexis+ Protégé review cover each product on its own terms; this page puts them side by side.
Layer by layer
| Layer | CoCounsel Legal (Thomson Reuters) | Lexis+ with Protégé (LexisNexis) |
|---|---|---|
| Corpus | Westlaw and Practical Law; "1 million professionals in 107 countries" use CoCounsel (TR, Aug 2026) | Lexis content; a knowledge graph LexisNexis puts at 200bn-plus documents (Feb 2026) |
| Citator | KeyCite, surfaced through Deep Research Verify | Shepard's, surfaced as Trust Markers |
| Model stack | Claude Agent SDK plus TR's own legal LLM | Anthropic, Google and OpenAI models plus LexisNexis' own; the Legal Intelligence Engine routes per task |
| Agents and drafting | Westlaw Brief Builder, Word drafting agent, Deep Research, Tabular Analysis (10,000 documents against 100 questions) | 300-plus configurable workflows, no-code workflow builder, Agentic Drafting (Henchman lineage), Protégé Work |
| Document store | Matter Workspaces; evidence integrations with Reveal and Everlaw | Vault, up to 100,000 documents per matter |
| Governance | Workspaces; Claude-based audit features not documented publicly | Workrooms with dual approvals, role-based permissions and audit trails; customer-held encryption keys |
| Price | Published: about $104–$639 per user per month; solo Westlaw Advantage plus CoCounsel Essentials $639.20 | No published list price found as of September 2026 |
| Availability | US GA 20 Aug 2026; Canada, UK and Australia "later in 2026"; earlier CoCounsel Legal reached the UK 26 Jan 2026 | US GA 24 Feb 2026 |
| Adoption (ILTA 2026) | 44% of firms | Lexis+ AI 26% of firms (the survey used the pre-Protégé name) |
| Vendor-reported satisfaction | NPS 77 | 81% of 543 surveyed lawyers "more comfortable" with AI grounded in legal sources |
The ILTA 2026 Technology Survey (14 September 2026; 500-plus firms) is the only independent adoption number on the table, and it shows CoCounsel at 44% against Lexis+ AI at 26%, level with Claude and a point ahead of Harvey. LexisNexis' own survey of 543 legal professionals (1 September 2026) is a preference statement, not a usage figure.
The citator is the product
Strip the agents away and each company is selling the same thing it sold in 1995: a citator wrapped around a corpus. KeyCite and Shepard's tell you whether an authority is still good law; the 2026 additions, Deep Research Verify and Shepard's Trust Markers, try to move that check inside the model's loop so the answer arrives with its treatment attached. Anthropic's open-source legal plugins take the same view from the other direction: any citation not checked against a connected research tool is tagged [verify].
That is real progress, and it is not immunity. Lacey v. State Farm (C.D. Cal., 6 May 2025) produced a $31,100 sanction against Ellis George and K&L Gates for nine bad citations out of 27, and the tools involved were CoCounsel, Westlaw Precision and Google Gemini. Fletcher v. Experian (5th Cir., 18 February 2026) fined counsel $2,500 for 16 fabricated quotations traced to vLex and CoCounsel. In an early-2026 analysis of the hallucination cases that named the tool, Westlaw AI accounted for 18% of them, second only to ChatGPT.
A verification routine that works on either
| Check | What the platform shows | What you still do |
|---|---|---|
| Existence | KeyCite link or Shepard's signal resolves to an opinion | Open the opinion; do not trust a resolved link that you have not clicked |
| Support | Deep Research Verify or a Trust Marker points to a passage | Read the passage; confirm it says what the answer claims, not something adjacent |
| Treatment | KeyCite flags or Shepard's signals | Check anything decided after the index date yourself |
| Quotation | Quoted text in the answer | Compare against the opinion verbatim; false quotations are the fastest-growing category in Charlotin's database (548 instances by 14 Sep 2026) |
| Jurisdiction | Answer names a jurisdiction | Re-run for each jurisdiction in play; every tool in Vals' 2025 study lost 11–14 points on multi-jurisdiction questions |
The citation-checking protocol on this site expands each row into a filing-ready checklist; California's 2026 guidance puts the rule in one line: no paper filed in court should contain a citation the responsible attorney has not personally read.
The hallucination record nobody has re-tested
The only independent test of either product line remains Stanford RegLab's "Hallucination-Free?" study, run on May 2024 versions: Lexis+ AI hallucinated on roughly 17% of 202 expert-scored queries and Westlaw AI-Assisted Research on roughly 33%; fully accurate and grounded answers came in at 65% and 41% respectively. Both vendors disputed the method. Since then the record is one of declining to be measured. In Vals' February 2025 report CoCounsel scored 89.6% on document question-answering and 77.2% on summarisation against a lawyer baseline of 50.3%, while Lexis+ AI withdrew from most tasks. In Vals' October 2025 legal research study, where ChatGPT Team scored 80% against a 69% lawyer baseline, Thomson Reuters, LexisNexis and vLex all declined to take part.
So, as of September 2026, no published independent number exists for the current CoCounsel Legal or Protégé. The benchmarks guide explains why vendor-run tests (Harvey's LAB, GC AI's In-House Legal Bench) cannot fill that gap. The practical consequence is that the hallucination rate you should plan around is the one you measure on your own questions, not the one in a brochure.
Drafting agents, vaults and workrooms
Where the two products have diverged is in what they built on top of the citator.
Thomson Reuters built for volume. Tabular Analysis runs 10,000 documents against 100 questions in one pass, Westlaw Brief Builder drafts from verified authorities, the Word drafting agent works inside the document, and the Reveal and Everlaw integrations pull evidence into the same workspace. It is the more obviously agentic of the two, which follows from being rebuilt on Anthropic's agent SDK.
LexisNexis built for control. The May 2026 release added Protégé Work, Agentic Drafting descended from the Henchman acquisition, a Vault holding up to 100,000 documents, and Workrooms with dual approvals, role-based permissions and audit trails, plus customer-held encryption keys. The Legal Intelligence Engine confirms its understanding and "presents an executable plan for user approval" before it runs. LexisNexis has also set a public target: automate 15–20% of what a lawyer does by 2028.
For a firm whose risk committee asks who approved what, Protégé answers the question in the product. For a firm whose bottleneck is a 10,000-document data room or a brief due Friday, CoCounsel's tooling is closer to the work. Neither gap is permanent; both companies ship monthly.
Jurisdictions and geography
CoCounsel Legal's next generation is US-only at general availability, with Canada, the UK and Australia promised for later in 2026; the earlier CoCounsel Legal reached the UK on 26 January 2026. Lexis+ with Protégé reached US general availability on 24 February 2026. For civil-law practitioners in Austria, Germany or CEE, both remain common-law research engines first; the grounding that makes them safer than a general chatbot on US questions does not transfer to a question on the ABGB, and the AI legal research page explains what to use instead.
If you are already on Westlaw
Stay, and pilot CoCounsel Legal properly. Your KeyCite habits transfer, the published price tiers let you budget without a sales call, and the Claude for Legal connector means your lawyers can reach Westlaw content from Claude Cowork if the firm adopts it. Watch the geography if you have offices outside the US and Canada. The one reason to look across the aisle is governance: if your engagement terms or a bank client's outside-counsel guidelines demand dual approval and customer-held keys, Protégé has them today and CoCounsel's public documentation does not.
If you are already on Lexis
Stay, and switch on Protégé's Workrooms before you switch on its drafting. Shepard's Trust Markers give you the same inside-the-loop verification story, the workflow builder lets a knowledge lawyer encode the firm's research method without code, and the Legal Intelligence Engine's approval step is the kind of human-in-the-loop control the SRA's 2026 warning notice and ABA Opinion 512 both assume. The reason to look across the aisle is scale of document work: if diligence and litigation review dominate, CoCounsel's Tabular Analysis and evidence integrations are the more developed toolset, and the Harvey vs CoCounsel page compares it against the platform alternative.
A switching test before you move
Moving research platforms is a multi-year library migration, not a tool swap. Run this before any decision.
- Write the problem in one sentence. "We need dual-approval workrooms for the bank panel" is a reason; "the AI seems better" is not.
- Price the whole move. Include the research subscription, retraining, the six months of dual running, and the precedent links that break.
- Run the same five questions on both, plus a $25 business chat account as a control: a point you argued last year with one adverse authority; a question settled in one circuit and open in another; a false-premise question; three verbatim quotations from an opinion you hold; and a drafting task restricted to authorities you supply.
- Open every citation in the underlying source and score each tool out of ten. Module 2 of the AI Lab for Lawyers course runs this same exercise on grounded and general tools, and the scores rarely match the sales deck.
- Check jurisdiction coverage and data terms in writing for every office that will use the tool.
- Decide, and if you stay, put the saved migration budget into the verification protocol and training, which is where sanctions are actually prevented.
The incumbents' bet is that grounding plus governance beats raw model quality for lawyers who file things. The evidence to date, including a database of 2,041 hallucination decisions worldwide and the sanctions cases that involve their own products, says the bet is sound only when the lawyer still opens the source.
Frequently asked questions
Is Lexis+ Protégé better than CoCounsel?
Neither is better across the board as of September 2026. CoCounsel Legal, rebuilt on Anthropic's Claude Agent SDK and generally available in the US since 20 August 2026, leads on agentic drafting and large-scale document analysis. Lexis+ with Protégé leads on configurable workflows, approval controls and customer-held encryption keys. The sensible choice is the platform whose library and citator your firm already relies on.
Does CoCounsel or Lexis Protégé hallucinate less?
Nobody outside the vendors has measured the current products. Stanford's RegLab found Lexis+ AI hallucinated about 17% of the time and Westlaw AI-Assisted Research about 33% in 2024, on earlier versions; both companies declined to join Vals' 2025 research benchmark. Sanctions cases such as Lacey v. State Farm show that a grounded tool still needs every citation opened and read.
How much does Lexis+ with Protégé cost compared with CoCounsel?
Thomson Reuters publishes a range of roughly $104 to $639 per user per month for CoCounsel bundles; a solo Westlaw Advantage plus CoCounsel Essentials package lists at $639.20. LexisNexis had not published a comparable Protégé price list in the sources reviewed for this page, so you negotiate. In both cases the AI is priced as part of the research subscription rather than as a separate product.
Can I use CoCounsel and Lexis+ Protégé outside the United States?
Partly. The next-generation CoCounsel Legal launched in the US on 20 August 2026, with Canada, the UK and Australia due later in 2026; an earlier CoCounsel Legal reached the UK in January 2026. Lexis+ with Protégé reached US general availability on 24 February 2026. For civil-law jurisdictions, treat both as common-law research tools and check coverage of your national sources before relying on either as a grounding layer.
Which AI models power CoCounsel and Lexis+ Protégé?
CoCounsel Legal is built on Anthropic's Claude Agent SDK combined with Thomson Reuters' own legal language model. Lexis+ with Protégé draws on models from Anthropic, Google and OpenAI plus LexisNexis' own, and its Legal Intelligence Engine, released on 24 August 2026, selects the model, sources and skills for each task automatically and presents a plan for the user to approve before it runs.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Tools
CoCounsel Legal (Thomson Reuters) Review: Westlaw-Grounded Agents
CoCounsel Legal reviewed as of September 2026. The Claude-built rebuild, Brief Builder, Tabular Analysis, Deep Research Verify, $104 to $639 pricing and grounding.
8 min read
Tools
Lexis+ with Protégé Review: Workflows, Vault and the Legal Intelligence Engine
Lexis+ with Protégé replaced Lexis+ AI in February 2026 with 300+ workflows, Shepard's-grounded verification and an agentic engine. What it does, costs and still gets wrong.
8 min read
Use Cases
AI Legal Research in 2026: What Works, What Hallucinates
AI legal research now beats a lawyer baseline on accuracy yet misapplies law in 24% of answers. Which tools are grounded, where errors hide, a routine that holds.
11 min read
Playbooks
How to Verify AI Legal Research: A Citation-Checking Protocol
Verify AI legal research with a five-check protocol: open every authority, match the quote, confirm holding and status, log it. Template, time budgets, tools.
11 min read
Comparisons
Harvey vs CoCounsel: Platform Play vs Westlaw-Grounded Research
Harvey vs CoCounsel for 2026: Westlaw and KeyCite grounding against firm-wide agents, published prices from $104 versus a 20-seat minimum, and a scenario table.
8 min read
Guides
Legal AI Benchmarks Explained: Vals, BigLaw Bench, LAB, HAQQ and LegalBench
Legal AI benchmarks explained: what Vals, BigLaw Bench, Harvey LAB, HAQQ and LegalBench measure, the September 2026 leaders, and how to run your own evaluation.
10 min read