Vincent AI vs Westlaw: how they compare in 2026

The AI is the product. Vincent is sold as a legal AI assistant, running more than 20 workflows across research, litigation and transactions on a pipeline that combines generative and rules based AI with vector search. The library it draws on is sold on its own as Clio Library, so what Vincent adds is the model work.

Westlaw works as a research service without its generative features. The base Westlaw Edge plan sells case law, statutes, regulations, KeyCite, the Key Number System, headnotes, Litigation Analytics and Quick Check with no generative AI, and Thomson Reuters still licenses it on its own. The models drive the two AI tiers. AI-Assisted Research answers plain language questions with links to authority, and Westlaw Advantage adds Deep Research, which plans and runs multistep research and returns a report, along with Litigation Document Analyzer, Complaint Analyzer, Claims Explorer, AI Jurisdictional Surveys and AI summaries of dockets. Thomson Reuters describes Deep Research as custom trained AI working with Westlaw's own research tools rather than a general chatbot placed over the content.

Independent measurement, a named citator and grounding a user can open are all in place. Vincent AI took part in the February 2025 Vals Legal AI Report, entering six tasks, and scored between 53.6 and 72.7 percent. It beat the lawyer baseline on document question answering at 72.7 percent, transcript analysis at 64.8 percent and document summarization at 58.9 percent. The evaluator praised it for declining to answer where it lacked enough data rather than giving a made up answer, and noted those refusals lowered its scores. Every research output cites primary sources with direct links and carries a Cert check on whether the cited law remains valid. Clio cites randomized controlled trials showing Vincent 3.67 times more reliable than leading language models, with no method published. vLex describes the Vals results as meeting or beating human benchmarks in four of five tasks, while the evaluator reports six tasks entered and three above baseline. vLex agreed to join the later Vals legal research study and withdrew before results were published, saying the study was not designed for enterprise AI tools.

Deep Research runs an agentic process over Westlaw and Practical Law content with Westlaw's own research tools. It quotes primary law in the language of the source and links each authority to the full document. Its Verify tab lists cited and further relevant law to check the analysis, and AI Jurisdictional Surveys show KeyCite status beside each source. Litigation Document Analyzer reviews a whole brief for inaccurate citations, misstatements of law and hallucinated cases. Thomson Reuters says Deep Research can occasionally produce inaccuracies. In June 2024 it said internal testing showed AI-Assisted Research accurate about 90 percent of the time, without publishing the test set. The preregistered Stanford RegLab and HAI study (Magesh and others, Journal of Empirical Legal Studies, 2025) ran in May 2024. It found AI-Assisted Research accurate on 42 percent of 202 questions and hallucinating on about a third, nearly twice as often as the other legal tools tested. Thomson Reuters disputed the result, and Deep Research came after the test.

Vincent works as a first draft assistant under lawyer review, with real review surfaces and no stated stopping point. Section 12(d) of vLex's terms asks that output be evaluated for accuracy, including by expert human review. Users see the sources an answer rests on and can exclude sources. Every research output carries linked citations and a Cert check. Inside Clio Operate, announced in August 2026 with general availability in November 2026, Vincent routes findings to the right lawyer and updates the matter record with lawyer confirmation. Every interaction there sits in the audit trail. Vincent Projects adds permissions and activity tracking on shared work. Nothing states what a workflow does unaided, where it stops, or what happens after an output is wrong.

Each AI mode has a stated job and a stated stopping point. AI-Assisted Research answers a question the lawyer asks. Deep Research plans a multistep investigation and runs it without further prompts, shows its approach in step by step research notes, and returns a report with arguments on both sides and links to each authority. Its Enhance tab asks questions to tailor the report, and its Verify tab lists cited and further relevant law. The Deep Research help article sets a categorical limit. Its summary and list of authority should never be used to advise a client, write a brief or motion for a court, or otherwise be relied on without further research. The General Terms add that a licensed professional or similar expert must review AI output before anyone relies on it, uses it or passes it on. Nothing says what Thomson Reuters does once a wrong answer is reported.

Named customers speak for the product, short of measured outcomes. Sarah Hirebet, Director of Knowledge Management at Stradley Ronon, is quoted saying the firm's lawyers love Vincent AI. The Waterloo Region Law Association's library says it uses Vincent to find similar material for members, and that research that took hours now takes minutes. BakerHostetler, Fisher Phillips, Cuatrecasas, Ferrari and Ryanair appear among the logos. Clio says 8 of the 10 top law firms in the world use Vincent. It cites independent studies showing productivity gains of at least 38 percent, with no study named or linked. No named customer is tied to a dated figure with a method.

Thomson Reuters says more than 50,000 law firms and legal departments use Westlaw, including 80 percent of the Am Law 100, along with more than 6,000 state government agencies and courts in 94 percent of states. The Department of Health and Human Services has granted Westlaw an Authority to Operate. A Thomson Reuters case study names Justly Prudent, a civil rights litigation firm, and credits CoCounsel and Westlaw AI with five times return on investment and a doubling of litigation capacity, without a method a reader can check. At the Deep Research launch in August 2025, Morgan Lewis's chief AI and knowledge officer was quoted on its reasoning, without figures. Usage of the AI tiers is not reported apart from Westlaw as a whole.

Vincent's commitments on confidentiality and segregation are real, and its own training position is not stated. Customers choose how long conversation logs and files are kept, and the data is permanently deleted after that. Each client's data is logically isolated, and in Clio Operate Vincent works within the firm's permissions and ethical walls. Zero retention agreements bind the AI providers, which Clio says means documents and queries are never stored or used for model training. vLex's terms bind both parties to confidentiality, though they define the customer's confidential information as its drafts and reports. The same terms give vLex a perpetual, irrevocable, sublicensable license to user content to provide the service, improve its products and create aggregated, deidentified information. Nothing addresses privilege or work product directly.

Thomson Reuters' Legal AI Terms of 23 January 2026 govern AI features in its legal products. They bar Thomson Reuters and its third party providers from developing, training or fine tuning any generative or foundation model with user prompts, user content or outputs unless agreed separately in writing. The same terms let Thomson Reuters use prompts and outputs to evaluate safety and effectiveness, with automated safeguards meant to keep sensitive and identifying information out of the samples reviewed. The General Terms let a party disclose the other's confidential information under law, subpoena or court order only with prior notice where legally permitted. The Data Security Addendum keeps each customer's data logically segregated from other customers' data. Attorney client privilege and work product are not addressed by name, and nothing describes separation by matter within a firm.

vLex states a real position on advice and review in its terms, short of supervision duties and jurisdiction limits. Section 12(c) says the services, and anything from its staff including reference attorneys, are not legal advice and create no attorney client relationship. Section 12(d) asks that machine generated output be evaluated for accuracy, including by expert human review. Clio says every research output gives lawyers a clear path back to the source before an answer reaches a client, filing or courtroom. The audience is professional throughout, from small firms to courts and government. No jurisdiction limits are stated across more than 100 countries, and competence and supervision duties are not addressed by name.

The General Terms say a licensed professional or similar expert must review AI output before anyone relies on it, uses it or passes it on. The customer alone is responsible for verifying its accuracy, and Thomson Reuters is not liable for decisions based on it. The Deep Research help article says its output should never be used to advise a client or write a brief or motion without further research. It presents the tool as a way to speed research rather than replace it. Westlaw is sold to law firms, legal departments, courts, government agencies and law schools, and the online plans are for law firms. Nothing addresses a lawyer's competence duties or names a jurisdiction's ethics rules.

Clio publishes evaluation and governance headings without a mechanism a buyer could audit. Vincent took part in the independent Vals Legal AI Report of February 2025, and the results were published, including tasks where it scored below the lawyer baseline. Its pages describe a hybrid pipeline of generative and rules based AI and cite randomized controlled trials, with no method given. The trust center for Vincent lists documents on the AI feature, employee AI use and AI training data and bias, available by request. No accountable owner, release testing gate or finding on uneven output across matter types, parties or populations is published.

Westlaw Edge, Precision and Advantage sit inside Thomson Reuters' ISO/IEC 42001:2023 certificate for its AI management system, issued by MSECB, held since March 2025 and valid to March 2028, and the certificate is published on the trust center. The Data and AI Governance and Security Program whitepaper of March 2025 makes the Responsible AI and Data Trust team responsible for AI governance and sends sensitive data uses to a Model Ethics Committee. Every business model must be registered in a central compliance hub and go through a data impact assessment that sets ethics controls such as human oversight and fairness, a model risk assessment, documented design and testing, monitoring and decommissioning. Attorney editors test and validate Westlaw's AI. No test results, bias findings or differences in performance across practice areas or jurisdictions are published.

Vincent's security material covers most of the ground, short of incident timing and deletion at the end of a contract. Customers choose how long conversation logs and files are kept, with one year as the default, and the data is permanently deleted at the end of that period. Role based access limits internal access to authorized staff under documented circumstances. Data is encrypted in transit and at rest with FIPS 140-2 cryptography, and each enterprise customer has its own master key held in hardware security modules. The trust center for Vincent names five subprocessors with their locations. AWS hosts the service, AWS Bedrock, OpenAI and Google Vertex AI supply the AI, and Turbopuffer runs the vector database. Zero retention agreements bind the AI providers. The privacy policy says affected users are told of a breach, with no time frame given. vLex's terms grant a perpetual, irrevocable license to user content and say nothing on deletion after the subscription ends.

The Data Security Addendum, last modified 26 October 2026, encrypts customer data with 256 bit AES in transit and at rest. It limits access to authorized personnel with multifactor authentication, keeps each customer's data logically segregated, and commits to notice of a security breach within 72 hours. It offers a summary of penetration test results on written request and destroys customer data at the end of the service on request. The General Terms delete production data within 90 days after termination on written request, and the Data Processing Addendum dated 15 October 2026 deletes personal data on request. Thomson Reuters names the outside AI providers behind its generative features as Amazon Web Services, Anthropic, Google, Microsoft and OpenAI. No retention period is stated for prompts, uploaded documents or research history during a subscription, and the Westlaw data retention overview for Litigation Document Analyzer sits behind a Whistic account.

Liability is handled through a standard limitation clause that disclaims the exposure the product creates. vLex's terms, updated March 2024 and naming Vincent AI among the services they cover, cap vLex's total liability at $100, with carve outs for fraud and willful misconduct. Consequential loss is excluded. Section 12(d) calls machine generated output probabilistic and asks that it be evaluated for accuracy, including by expert human review. The customer indemnifies vLex for breaches of the terms and for third party claims over documents it uploads. No indemnity runs from vLex to the customer, and no warranty on output is given. Section 15(E) says these terms prevail over a subscription agreement where the two conflict.

Thomson Reuters' General Terms for the United States (version 6.0, 1 September 2026) cap each party's liability in a calendar year at the fees paid for the affected service over the prior twelve months. That cap doubles for breaches of confidentiality, data protection or security, and indirect and consequential loss is excluded. Thomson Reuters defends and indemnifies the customer against third party claims that use of the service infringes intellectual property, and that indemnity sits outside the cap. The service warranty promises conformity with the documentation, with repair, modification or replacement as the remedy. For AI output, the terms disclaim liability for its accuracy, legality or reliability and for decisions based on it.

Real integrations are named, including the legal ones, with some depth on the deepest. vLex names Microsoft Word, Outlook and iManage connections. Inside Clio Operate, Vincent reads the structured matter record, including facts, history, documents and workflows, and can draw on a firm's connected document systems. It returns outputs to fields, chronologies and Wiki pages in Operate, and updates the record with lawyer confirmation. Vincent Studio lets enterprise customers build firm workflows without code, and Clio's custom team builds further integrations on request. No setup guide says what each connector moves, in which direction, or what an administrator configures.

Westlaw's product and plan pages name no document management, practice management or Microsoft Word integration. The AI tools work on what a lawyer types or uploads: Litigation Document Analyzer, Complaint Analyzer and Quick Check take an uploaded brief or complaint. Deep Research also runs inside CoCounsel Legal, the separate Thomson Reuters assistant, which connects to iManage, NetDocuments and SharePoint. Thomson Reuters lists terms for a document management system integration among its legal products without saying what it moves or how a firm sets it up.

Vincent runs as a multitenant service, with each client's data logically isolated at the infrastructure level, and residency is offered. Customers can host data and keys in the United States, the European Union or Australia. Clio's trust center lists AWS hosting in all three regions. It lists AWS Bedrock and OpenAI as AI providers in the United States and the European Union, and Google Vertex AI in all three. Each enterprise customer gets its own encryption master key. Nothing states which AI region serves a customer whose data sits in Australia, or what else changes between tiers.

Westlaw is a hosted service, and the Data Security Addendum keeps each customer's production data logically segregated from other customers' data. The Department of Health and Human Services has granted Westlaw an Authority to Operate, and Thomson Reuters says Westlaw is in the final stage of FedRAMP authorization. Separate security summaries cover the American, UK, Canadian, Australian and New Zealand editions. No hosting provider, region or residency commitment is published for the American service, and nothing separates where AI processing runs from where data is stored.

Certification is current, named and reachable through a trust center. Clio's trust center for Vincent and vLex lists SOC 2 Type 2, ISO/IEC 27001:2022 with its statement of applicability, HIPAA, GDPR and a VPAT. It offers the SOC 2 report, a penetration test summary and security and privacy white papers through an access request. vLex's support article dates SOC 2 to January 2025 and ISO 27001 to June 2025. Security systems undergo regular independent assessments and real time monitoring. The auditor and the audit period are not published.

The Westlaw security profile on Whistic, linked from the Thomson Reuters trust center, lists a Westlaw SOC 2 Type 2 report dated October 2025 and a bridge letter of November 2026, alongside a SIG Lite questionnaire, security summaries by country and an AI security FAQ. Downloading them needs a Whistic account under a Thomson Reuters confidentiality agreement, and the profile does not say the release is immediate. The ISO/IEC 42001:2023 certificate, which names Westlaw Edge, Precision and Advantage, is published openly. The Department of Health and Human Services has granted Westlaw an Authority to Operate, and FedRAMP authorization is in its final stage. The Data Security Addendum lets a customer review an executive summary of ISO 27001 or SOC 2 results on request. The SOC auditor and the report's scope are not stated publicly.

Clio names the AI providers behind Vincent and where they run, without naming the models. The trust center for Vincent lists AWS Bedrock and OpenAI as AI providers in the United States and the European Union. It lists Google Vertex AI in the United States, the European Union and Australia, and Turbopuffer as the vector database. Vincent's pages describe a hybrid pipeline of generative and rules based AI with vector search over the library. Zero retention agreements bind the model providers. No model or version is named. The trust center offers a subscription to updates, and no commitment to notify customers of a provider change is published.

Thomson Reuters names the outside AI providers whose terms apply to its generative AI products: Amazon Web Services, Anthropic, Google, Microsoft and OpenAI. Deep Research is described as custom trained AI combined with Westlaw's research tools. Nothing maps a provider or model to AI-Assisted Research, Deep Research or the litigation tools, says where inference runs, or commits to notice before the providers change. Thomson Reuters announced its own legal model, Thomson, in August 2026 for CoCounsel Legal's tabular analysis, and nothing says it runs in Westlaw.

Clio shows how Vincent is packaged and publishes no price. Its pricing page places Vincent in Clio for Enterprise, the offer for Am Law firms and corporate legal departments, alongside Clio Operate. It says administrators control who can use which Vincent resources. Vincent Studio is for Enterprise customers. Vincent is bought through a demo, and vLex offers a free trial with no length or terms stated. No price, unit of charge or plan figure is published. For smaller firms Clio sells Clio Work, built on the same library and the Cert citator, as an add on or inside the Signature Work and Elite Work packages, also without a published price.

Thomson Reuters prices Westlaw online for new customers at law firms of up to 10 attorneys, by number of attorneys, jurisdiction and term. Westlaw Edge with AI-Assisted Research starts at $155.35 a month for single circuit coverage and $266.50 for all states and federal coverage, both on a three year term. Westlaw Advantage starts at $256.75 a month for single circuit coverage and $399.75 for all states and federal. Two year terms take 12 percent off and three year terms 18 percent, and plans include new features released during the initial term. Westlaw Edge without AI is sold online with no starting price stated. Firms with more than 10 attorneys and existing customers are quoted by sales, and documents outside a plan are charged at rates in Westlaw's subscriber pricing guide.

Coverage is quantified, and its limits are named in part. The library holds more than one billion documents from more than 100 countries, with country collections named from Brazil to the United Kingdom. More than 850 million court records sit behind the litigation profiles. Vincent runs in English, Spanish and Brazilian Portuguese. Workflows cover research, litigation, transactions and litigation intelligence. vLex serves large and small law firms, legal departments, bar associations, law schools, law librarians and government departments. Clio sells Vincent through its enterprise offer to Am Law firms and corporate legal departments, and its offer for smaller firms is Clio Work. The Vals evaluator noted its study covered a small part of the product, and vLex said document summaries do not draw on its legal database. Outside Canada, the United Kingdom and Ireland, most cases show only a citation count rather than treatment.

Westlaw serves law firms of every size, corporate legal departments, courts, prosecutors, public defenders, city attorneys, law enforcement, other government agencies and law schools. Thomson Reuters publishes practice area pages for business and commercial law, civil litigation, criminal defense, family law, labor and employment, mergers and acquisitions and personal injury. Content covers state and federal case law, statutes, regulations, court rules, administrative materials, dockets, briefs and secondary sources such as Wright & Miller and Corpus Juris Secundum. The plans set the limits. Westlaw Edge carries no generative AI, AI-Assisted Research and AI Jurisdictional Surveys come with Edge with AI-Assisted Research, and Deep Research, Litigation Document Analyzer, Complaint Analyzer and Claims Explorer come only with Advantage. A single circuit plan covers state and federal primary law for one circuit. Online buying stops at 10 attorneys, and separate editions serve the UK, Canada, Australia and New Zealand.