Westlaw Precision AI: An Independent Review for Practicing Lawyers
Westlaw has been the default legal research shelf for a generation of American lawyers. The addition of generative AI on top of that shelf — first as AI-Assisted Research, later folded into the Precision tier and branded as CoCounsel-powered features — is one of the more consequential product shifts of the last two years. This review evaluates Westlaw Precision AI on its merits for a practicing lawyer: what it actually does, where your prompts and matter facts travel, how it lines up with recent ethics guidance, and who should pay for it.
This is an evergreen review in Basil's independent legal-tools library. Basil is an on-device meeting notetaker built by a practicing lawyer; Westlaw Precision AI is a research tool, not a competitor. If it matters to you, Basil pairs naturally alongside Westlaw for capturing the client interviews and internal case-strategy calls that feed the research questions you eventually put to Precision. You can read more at Basil for Law.
What Westlaw Precision AI Actually Does
Westlaw Precision AI is the generative-AI overlay on Thomson Reuters' Westlaw Precision research platform. It bundles several distinct capabilities that Thomson Reuters describes on its Westlaw Precision product page and its CoCounsel pages:
- AI-Assisted Research. You ask a natural-language legal question and receive a synthesized answer with pinpoint citations to Westlaw content — cases, statutes, secondary sources, KeyCite treatment.
- Quick Check with AI. Upload a brief and receive suggested authorities to strengthen or challenge it, plus flags on the reliability of cited cases.
- Ask Practical Law AI. A generative interface over the Practical Law how-to library and forms, aimed at transactional and workflow questions.
- Precision search enhancements. Editor-tagged issues, fact patterns, motion outcomes, and holding narrowing filters — the non-generative half of Precision that many lawyers actually find most useful day-to-day.
- CoCounsel skills. Document review, deposition prep, contract comparison, and drafting workflows, delivered through the CoCounsel interface and increasingly stitched into Westlaw itself.
The design choice worth noting: answers are grounded in Westlaw's editorial corpus and returned with citations. This is retrieval-augmented generation, not a general-purpose chatbot guessing at law. When it hallucinates — and any generative system can — the failure mode tends to be a real case cited for a proposition it does not quite support, rather than a fabricated case name. That is still a problem you have to check for, but it is a materially different problem than what unaccompanied ChatGPT-style tools produce.
The Confidentiality Lens
For a practicing lawyer, the first question about any AI tool is where your inputs go, who can see them, what happens to them, and what a subpoena on the vendor could reach. Here is what Thomson Reuters publishes.
Processing location and model use. Thomson Reuters' CoCounsel and Westlaw AI trust page and its AI Principles state that customer inputs and outputs are not used to train foundation models, and that third-party model providers (including OpenAI and Anthropic, which power parts of the stack) are contractually restricted from training on customer content. Processing happens in Thomson Reuters' cloud environment, not on your device. This is a hosted-SaaS model with vendor commitments about training, not a local one.
Retention. Retention specifics depend on your contract and the module. Thomson Reuters' Westlaw terms and product-specific documentation govern; if retention duration matters to your matter (it usually does), get it in writing from your account rep and confirm the settings in the admin console.
DPA and privacy posture. Thomson Reuters offers data processing agreements and publishes SOC 2 attestations for its legal products. Its privacy statement is the starting point; enterprise customers should negotiate the DPA rather than rely on the default.
Subpoena exposure. Because prompts, uploaded briefs, and generated outputs sit on Thomson Reuters infrastructure, a third-party subpoena served on the vendor could in principle reach them subject to the retention window and the vendor's response practices. This is not unique to Westlaw — it is true of every cloud-hosted legal AI tool — but it is worth stating plainly. The mitigation is disciplined: do not paste privileged client narrative into the research prompt when a more abstract legal question will do; use client-identifying facts only when the module actually needs them (Quick Check on a real brief does; a case-law question rarely does).
What this is not. Westlaw Precision AI is not on-device processing and does not claim to be. That is a legitimate architecture, especially for a tool whose value depends on access to a massive proprietary corpus, but it means the confidentiality story rests on contract and vendor governance rather than physical isolation. Compare that with tools like Basil that keep audio and transcripts on the lawyer's own machine; the two approaches solve different problems.
Ethics-Opinion Fit (ABA 512)
ABA Formal Opinion 512 (July 2024) sets out the framework U.S. lawyers should apply to any generative AI tool: understand the technology, protect client confidentiality under Model Rule 1.6, supervise the output, communicate with clients about use where material, and consider reasonable fees. Westlaw Precision AI maps to that framework better than most tools on the market.
- Competence and understanding. Thomson Reuters publishes reasonably detailed documentation on how retrieval and generation work in the product, which supports the Rule 1.1 competence obligation.
- Confidentiality (Rule 1.6). The no-training commitments and DPA availability address the core confidentiality concerns, though they do not eliminate cloud-processing risk. A lawyer still has to make an informed judgment about what facts go into a prompt.
- Supervision and verification. The citation-grounded output design makes verification tractable. You can click through to the underlying case and read it. That does not remove the obligation; it just makes discharging it realistic.
- Client communication and fees. Opinion 512 flags that AI-driven efficiencies may affect what is a reasonable fee. If you are billing hourly for research that Precision AI compresses from three hours to forty minutes, think through your engagement letter.
Recent decisions reinforce the point that AI-tool choice is now a diligence question. In US v. Heppner (S.D.N.Y. Feb. 2026), Judge Rakoff held that a litigant's chats with a public AI platform were not privileged. Westlaw Precision AI is not a public consumer platform, and its use in a lawyer's research workflow is a very different posture — but the case is a reminder that the tool's terms and architecture matter, not just its marketing.
Strengths That Are Genuinely Earned
Several things about Westlaw Precision AI are simply good, and it would be dishonest to review the product without saying so.
- The corpus. Westlaw's editorial infrastructure — headnotes, KeyCite, Practical Law — is decades of investment that generative retrieval sits on top of. The answers are only as good as the source, and the source is very good.
- Citation grounding. Every AI answer I have seen from the tool comes with linkable citations. That is the correct design choice for legal research and it materially reduces (though does not eliminate) hallucination risk.
- Quick Check. Running a draft brief through Quick Check before it goes out is a genuinely useful backstop for missed authority and unfavorable treatment. It is the feature I hear the most unprompted praise for from litigators.
- Integration. If your firm already lives in Westlaw and has Practical Law seats, the AI features surface where you already work rather than as a separate tab you forget to open.
Limits and Honest Weaknesses
- Price opacity. Precision AI pricing is quote-based and varies by firm size, module bundle, and negotiation. That is fine for enterprise procurement and frustrating for solos.
- Cloud dependency. As noted above, the confidentiality story is contractual, not architectural. Firms with the most sensitive matter types (national-security work, sealed matters, certain regulated industries) should think carefully about what they upload.
- Verification still required. Grounded citations reduce hallucination; they do not eliminate mischaracterization. Every AI-generated summary needs to be read against the underlying opinion. This is a feature of the state of the art, not a Westlaw problem specifically.
- Uneven feature maturity. Some CoCounsel skills are excellent; others are clearly earlier in their curve. Ask for a demo on your actual workflow before you buy on the strength of the demo they want to show you.
Pricing and Who It's For
Thomson Reuters does not publish Precision AI pricing on its site; the current details live behind the Precision product page contact form. Expect a subscription layered on top of your existing Westlaw plan, with meaningful firm-size and module variation. Reputable industry reporting from LawNext and the LawSites blog tracks the evolution of the offering if you want context before your renewal call.
Who should buy it:
- Litigation and appellate practices where Quick Check and AI-Assisted Research plausibly compress hours of brief-cite-checking and issue-scanning per matter.
- Firms already deep in Westlaw where the marginal cost of the AI layer is smaller than the cost of retraining on a new research platform.
- Transactional groups with heavy Practical Law usage who will benefit disproportionately from Ask Practical Law AI.
Who should hesitate:
- Solos on tight budgets for whom Lexis+ AI, Fastcase/vLex Vincent, or even careful non-AI research may pencil out better.
- Practices dominated by sealed or highly sensitive matter types where cloud processing of prompts is a real constraint.
Verdict
Westlaw Precision AI is a serious tool, sold by a serious vendor, with a confidentiality posture that is realistic rather than performative. It does not eliminate the lawyer's verification obligation and does not promise to. What it does is compress the mechanical labor of legal research while keeping the citation trail visible, which is what a well-designed research AI should do.
| Pros | Cons |
|---|---|
| Citation-grounded answers over a first-rate proprietary corpus | Cloud-hosted; confidentiality rests on contract, not architecture |
| Quick Check is a genuinely useful backstop on outgoing briefs | Pricing is quote-based and opaque for smaller firms |
| Vendor commitments against model training on customer content | Verification of every generated summary still required |
| Deep integration with Practical Law and existing Westlaw workflows | CoCounsel skill maturity varies; demo on your actual work |
| Reasonable fit with ABA Formal Opinion 512 framework | Not appropriate as-is for the most sensitive matter categories without careful review |
If you already have Westlaw and do meaningful research volume, Precision AI is worth the seat. Pair it with disciplined intake — including on-device meeting capture for the client conversations that generate the research questions in the first place; see Basil for Law — and you have a defensible modern stack.
This review is for information only and is not legal advice.
Frequently asked questions
Does Westlaw use my prompts or uploaded documents to train its AI models?
Thomson Reuters states on its CoCounsel and Westlaw AI trust materials that customer inputs and outputs are not used to train foundation models, and that third-party model providers are contractually restricted from training on customer content. Confirm the specifics in your contract and DPA, because retention and processing details can vary by module and by negotiated terms.
Is Westlaw Precision AI processed on my device or in the cloud?
It is a cloud-hosted service running on Thomson Reuters infrastructure. Your prompts, uploaded briefs, and generated answers are processed in the vendor's environment. That is a legitimate architecture for a tool built on a massive proprietary corpus, but it means confidentiality relies on contract and vendor governance rather than physical isolation on your machine.
How does Westlaw Precision AI fit with ABA Formal Opinion 512?
It aligns reasonably well. The citation-grounded output design supports the supervision and verification duties, the no-training commitments address core Rule 1.6 confidentiality concerns, and Thomson Reuters publishes enough documentation to support the competence obligation. Lawyers still have to make informed judgments about what facts they put into prompts and about how AI-driven efficiency affects reasonable fees.
Can I rely on Westlaw Precision AI's answers without checking the underlying cases?
No. Citation grounding materially reduces hallucination risk compared to general-purpose chatbots, but generative systems can still mischaracterize a real case's holding. Every AI-generated summary should be read against the underlying opinion before it lands in a brief or client memo.
How much does Westlaw Precision AI cost?
Thomson Reuters does not publish Precision AI pricing publicly. It is quote-based and varies with firm size, module bundle, existing Westlaw commitments, and negotiation. Contact Thomson Reuters through the Westlaw Precision product page for current pricing and be prepared to negotiate on term length and included modules.
Could a subpoena served on Thomson Reuters reach my research prompts?
In principle, yes — anything held on vendor infrastructure within its retention window is potentially reachable by third-party process directed at the vendor. This is true of every cloud-hosted legal AI tool. The practical mitigations are limiting client-identifying facts in prompts to what the module actually needs, negotiating retention terms, and reserving the most sensitive matter narratives for tools with stricter architectural isolation.
Meeting notes with no server to subpoena
Basil transcribes and summarizes entirely on-device — privilege-safe by architecture. See Basil for Law →
This review is for information only and is not legal advice.