Thomson Reuters builds CoCounsel, an AI assistant grounded in Westlaw case law and Practical Law guidance, for lawyers, tax and compliance teams who need cited, verifiable answers rather than plausible prose.
Used this tool? Rate it
Used this tool? Rate it
Thomson Reuters is a global content and technology company that sells authoritative information and professional software to lawyers, tax and accounting practitioners, compliance officers, trade teams and newsrooms. In an AI tools directory, however, the company is best understood through one product line rather than the whole conglomerate: CoCounsel, its AI assistant for professional work. The group organises itself around Legal, Tax and accounting, International trade and supply chain, Risk fraud and investigations, and the Reuters news business, and CoCounsel is the AI layer that now runs across most of those verticals.
What makes CoCounsel different from a general chatbot is not the model behind it but what it is allowed to reason from. The company describes CoCounsel Legal as reasoning from authoritative Westlaw primary law, Practical Law guidance, and your organization's own knowledge. That grounding is the entire product thesis: in legal and tax work, an answer that sounds right but cites a case that does not exist is worse than no answer at all.
Thomson Reuters markets CoCounsel as "Fiduciary-Grade AI™ built for high-stakes professional work", a phrase that signals the company's core bet. Thomson Reuters defines it as the standard for "how AI should work in high-stakes professions", aimed at professionals with duties of care and regulatory oversight. The positioning is deliberate: this is software sold to people who can be sanctioned by a court or a regulator for the work they submit.
The scale is substantial. Thomson Reuters has said CoCounsel reached one million customers across 107 countries and territories, running on a multi-model architecture that draws on frontier models from Anthropic, OpenAI and Google. That user figure is a vendor announcement rather than an audited disclosure, and should be read as such. (Based on public information)
CoCounsel did not originate inside Thomson Reuters. It launched in March 2023 as a product of Casetext, billed as the first AI legal assistant built on GPT-4, and Thomson Reuters acquired the company a few months later for $650 million in cash. Since then the brand has been extended well past its original litigation-support scope, absorbing tax, audit, accounting, risk and trade workflows, and becoming the umbrella name for the group's professional AI.
If you are evaluating AI for regulated professional work, the questions that matter are narrow and unusually consequential: what corpus does it search, can you verify every citation, does your confidential material leave your control, and how often is it wrong. This page answers those questions with sourced material where sources exist, and says so plainly where they do not.
There is no free tier and no public self-service signup for CoCounsel in the way consumer AI tools offer one. Thomson Reuters does not publish a price list for CoCounsel Legal; the product pages route buyers to a sales contact and a telephone number instead. Prospective buyers can request a demonstration, but the commercial terms are established through a sales conversation rather than a published rate card.
Paid access is structured as enterprise and professional subscriptions, typically sold in relation to the underlying research platforms rather than as an isolated AI add-on with a flat public price. Pricing in practice depends on firm size, contract term, which underlying subscriptions you hold and which modules you enable. Third-party sites publish estimated figures, but those estimates vary widely and contradict each other, so no specific number is reproduced here. The reliable statement is the one the vendor makes: pricing is quoted, not listed. Anyone budgeting for this should treat the total cost of the research platform plus the AI capability as the real figure, and should confirm renewal terms and seat minimums during negotiation.
The most important limitation is documented by independent researchers rather than the vendor. A preregistered Stanford RegLab study found that the legal research tools it tested, including Westlaw AI-Assisted Research and Ask Practical Law AI, hallucinated between 17% and 33% of the time. The same researchers concluded that providers' hallucination-free claims are overstated, while acknowledging that these tools do reduce hallucinations relative to general-purpose chatbots. Both halves of that finding matter: grounding demonstrably helps, and grounding is demonstrably not a guarantee.
Thomson Reuters is explicit that the technology "is designed to support human judgment, not replace it", and that professionals remain responsible for the work they sign off on. That is not boilerplate in this context — it is the correct allocation of risk given the measured error rates, and it means verification effort must be budgeted rather than assumed away.
Vendor-reported metrics should be read with the same care. Thomson Reuters cites its own customer survey figures, reporting that 76% agree the product improved research and drafting quality and 64% agree it helped reduce risk. These are self-reported satisfaction figures from the vendor's own research, not independently audited performance measurements, and they measure perception rather than accuracy.
Other considerations worth weighing before committing:
No. There is no free tier and no public self-service signup. Access is sold through enterprise and professional subscriptions arranged with the sales team, and the product pages direct interested buyers to contact the company rather than to a checkout page. A demonstration can be requested before purchase.
No, according to the company's own published statement. Thomson Reuters answers this directly: "Your user content and prompts are not used to train or improve CoCounsel and associated products or LLMs." The company further states that third-party partners such as OpenAI and Google are "contractually prohibited from using any customer data to train their models". Thomson Reuters also says it processes interactions with those third-party models in the United States and, where applicable, has controls to disable third-party abuse-monitoring so there is no human access to customer data.
Both. Thomson Reuters has described a multi-model architecture drawing on frontier models from Anthropic, OpenAI and Google alongside its own proprietary AI technology. The next generation of CoCounsel Legal is built specifically on Anthropic's Claude Agent SDK. So the underlying reasoning models are largely licensed from third parties, while the authoritative content, retrieval layer and professional workflows are Thomson Reuters' own.
Yes. This is the single most important caveat. Independent Stanford RegLab research measured hallucination rates between 17% and 33% across the leading legal research AI tools it tested, including Thomson Reuters products, and concluded that vendor claims of hallucination-free output were overstated. Grounding in an authoritative corpus reduces the problem relative to general chatbots but does not eliminate it, which is why every citation must be opened and read.
Thomson Reuters does not publish pricing for CoCounsel. Costs are quoted individually and depend on firm size, contract length, which underlying research subscriptions you hold and which modules you enable. Third-party estimates circulate online but vary widely and conflict with one another, so they are not reliable for budgeting. Request a formal quote, and evaluate the combined cost of the research platform and the AI capability together.
CoCounsel Legal is described as working across Westlaw, Practical Law, Microsoft 365, and document management systems. The primary interface is a web application accessed with an organisational account, and the tax, audit and accounting capabilities appear within their own product surfaces such as ONESOURCE and Checkpoint rather than in a single unified app.
Partially, and coverage is expanding rather than universal. Thomson Reuters announced CoCounsel Legal availability in the United Kingdom in January 2026, with further rollouts in markets including Canada and Australia. Because the tool's usefulness depends entirely on licensed content for your legal system, confirm coverage for your specific jurisdiction and practice area before purchasing.
Yes. The CoCounsel brand now spans tax, audit, accounting, risk and trade in addition to legal work. Capabilities include agentic support for indirect tax compliance through ONESOURCE, document analysis applied to audit sampling and workpaper generation, and integration with Checkpoint research content, with the knowledge base extended to authoritative accounting standards material.
The professional using it. Thomson Reuters states plainly that Fiduciary-Grade AI is designed to support human judgment, not replace it, and that professionals remain responsible for the advice they give and the work they sign off on. Courts and regulators have taken the same view, and no vendor assurance transfers professional responsibility away from the practitioner who files the document.
It depends heavily on practice volume. The economics favour organisations with enough document-heavy work to offset a substantial annual subscription covering both the research platform and the AI capability. For solo practitioners and small firms, the honest assessment is that the total commitment is significant relative to likely usage, and a narrower point solution may deliver better value than a full professional research stack. (Based on public information)
