The short version. Lexis+ with Protégé is the grounded research platform with the better published error record, though the study behind that is now dated. CoCounsel is the answer if your firm already lives in Westlaw. vLex Vincent is the only serious option a solo can price without a sales call. Clio Duo is the one aimed at where a small firm actually loses money, which is billing rather than research. ChatGPT or Claude is excellent for everything that is not a citation. And every documented sanction in this area came from treating it as though it were.
A note on prices, and it is unusual enough to state up front: we could not verify a single price in this category at source. Clio’s and vLex’s pricing pages both returned 403 to us on 15 August 2026, and Lexis and Harvey quote by sales call. Every figure below is third-party reported and labelled as such. Our house rule is that we do not print numbers we have not read on the vendor’s own page, so treat everything here as an order of magnitude for budgeting, not a quote.
How we ranked, when the risk is discipline
This is the first list we have published where the risk of a tool is professional discipline rather than a wasted subscription. So the criterion that dominates everything else is grounding: does the output tie back to a retrievable source you can open, or is it generated text that resembles law?
We weighted:
- Does every proposition come with a source you can click? Not a citation-shaped string, a link to a document in a database.
- Does it fit a firm that is not an AmLaw 100? Most of this category is sold to large firms by sales teams. Solos and small firms are the majority of the profession and an afterthought in the market.
- Is the price knowable? A product you cannot evaluate without a sales cycle has a real cost even before you buy it.
- Does it help with the part of practice that is not research? Most firms lose more money to unbilled time than to slow research.
Three of the five below are research platforms, which breaks a rule we normally hold to. On our other lists we refuse to spend multiple slots on one decision. Here that is the wrong instinct: these are mutually exclusive purchases, the choice between them is the actual question lawyers have, and the right answer depends entirely on which database your firm already pays for. So we have treated them as three answers to one question and said plainly which buyer each is for.
1. Lexis+ with Protégé: best grounded research for most firms
Pricing by sales call; LexisNexis does not publish figures. Reported deployments sit in the hundreds of dollars per user per month, on top of an existing Lexis subscription. Not verified at source.
LexisNexis replaced Lexis+ AI with Lexis+ with Protégé in February 2026, moving from a research assistant to a workflow platform with agentic drafting and multi-model routing underneath. The reason it ranks first is the reason anything ranks in this category: its output is anchored to documents in the Lexis database rather than generated from general training data, and it is the product with the better independent measurement behind it.
That measurement deserves care, because it is the most-cited and most-abused statistic in legal AI. Stanford’s RegLab and HAI tested these tools across 202 expert-scored legal queries. Published as a preprint in May 2024 and peer-reviewed in the Journal of Empirical Legal Studies in 2025. Lexis+ AI hallucinated on roughly 17% of queries; Westlaw’s AI-Assisted Research on roughly 33%.
Two things follow, and most coverage takes only the first. Yes, Lexis measured better. But the number that should change your behaviour is 17%, not the gap: the best-performing grounded legal research tool in an independent study was wrong about one query in six. Also note the study predates both products’ current versions, Lexis+ AI no longer exists as tested. Treat it as proof that grounding reduces error rather than eliminating it, not as a current scoreboard.
Who it’s wrong for: firms committed to Westlaw, where you would be paying twice for overlapping databases. And solos, for whom the sales process alone is a deterrent and the price is not designed for you.
2. CoCounsel: best if your firm already lives in Westlaw
Reported published tiers from around $75/month for basic access to around $500/month for full case-law search, with the top tier requiring a separate Westlaw subscription. All-in costs are commonly reported at $300–$600 per user per month. Not verified at source.
The most recognised name in legal AI and, for a large share of firms, the correct choice for an unglamorous reason: it is built on the database you already pay for, and your team already knows the citator. Switching research platforms is not a software decision, it is a retraining and re-licensing decision, and the AI layer is rarely worth it.
The drafting and review workflows, deposition preparation, document review, contract analysis, summarising a production, are where it earns its keep. That is a broader surface than research alone, and it is where the hours actually are in litigation practice.
Its AI-Assisted Research measured worse in the Stanford study than the Lexis product, and Thomson Reuters has iterated since. We would not switch databases on the strength of a 2024 study, and we would not skip verification on the strength of a vendor’s response to it either.
Who it’s wrong for: anyone not already on Westlaw. The total cost only makes sense as an add-on to a subscription you have. And any firm expecting the tiers to be self-serve; the useful configuration involves a conversation.
3. vLex Vincent: best for solos and small firms
Reported at around $79 per user per month, and reported to be the only major platform offering a free tier. Not verified at source. VLex’s pricing page returned a 403 to us on 15 August 2026.
The market’s structural failure is that the majority of practising lawyers work in firms of one to five people, and almost every serious legal AI product is priced and sold as though they do not exist. vLex is the exception that is both grounded and approachable, which is why it ranks here despite less independent testing than the two above.
For a solo, the calculation is not “which platform is best” but “which grounded platform can I afford at all”. The alternative most solos actually choose is a general chatbot with no database behind it, and that is the choice that produces sanctions. A cheaper grounded tool beats an excellent one you do not buy.
Who it’s wrong for: practices whose work depends on the depth of Westlaw’s or Lexis’s secondary sources and citators. Coverage is the trade, and in some practice areas it is not one you can make.
4. Clio Duo: best because the money is in billing, not research
Pricing per user per month across several tiers, with the AI features on the higher ones. We could not verify current figures. Clio’s pricing page returned a 403 to us on 15 August 2026.
Every other entry helps you produce work. This one helps you get paid for it, and for most small firms that is the larger number by a wide margin. Time that goes unrecorded because a call ran into the next call is revenue that simply never existed, and no research tool recovers it.
Clio is the practice-management default for small firms, and Duo layers AI onto the parts of it that are genuinely tedious: summarising a matter before a client call, drafting routine correspondence with the matter context already loaded, surfacing what has not been billed. The context is what makes it different from a general assistant. It already knows the matter, the parties and the history.
If your firm has no practice management system at all, that is the purchase to make before any AI on this page. The AI is a reason to pick Clio over an alternative; it is not the reason to buy practice management.
Who it’s wrong for: firms standardised on another practice management platform. The AI layer is not a reason to migrate a matter database. And large firms, whose stack is decided elsewhere.
5. ChatGPT or Claude: best for everything that is not a citation
Both sell an individual tier around $20 a month. We describe rather than quote: OpenAI’s pricing page returned a 403 to us on 15 August 2026.
Enormously useful and, used correctly, the best value on this page. Restructuring an argument, drafting a client update in plain English, summarising a long document you have supplied, preparing questions, turning notes into a first draft, explaining an unfamiliar area well enough to know what to research properly.
Everything on that list has one property in common: the source material is something you provided, or the output is not a statement of law. That is the line, and it is the entire content of this section.
Crossing it is what produces sanctions, and this is thoroughly documented. In Mata v. Avianca, a New York federal court fined attorneys $5,000 over a brief citing cases ChatGPT had fabricated. Reported federal sanctions since have ranged from around $2,500 to $31,100, with one Florida attorney reported sanctioned $86,000 across eight cases. The tool does not know it is inventing a citation, because a fabricated case and a real one are the same kind of object to it.
The rule that prevents all of it: never ask a general model what the law is, and never let a citation reach a filing without opening the case. Not checking the citation format. Opening the document and reading the holding. A model asked to verify its own citation will frequently confirm it.
Who it’s wrong for: any lawyer who will not adopt that rule, for whom this is the most efficient route to a disciplinary complaint the profession has produced. Also check your jurisdiction’s position on client confidentiality and third-party AI processing before pasting anything privileged. Several bars have issued guidance, and it is not uniform.
Firms this list isn’t for
AmLaw 100 firms. Your stack involves Harvey or a bespoke deployment, procurement, and a security review. Nothing here is priced or scoped for you.
In-house legal teams. Your problem is contract volume and intake triage, not case-law research, and the tooling that matters is CLM.
Anyone in a jurisdiction outside the platforms’ coverage. Grounding is only as good as the database, and these are US-centric. A grounded tool with poor coverage of your jurisdiction is a confident tool with a thin library.
Anyone hoping to skip the verification step. There is no tool on this page, or on the market, that removes the obligation to read the case you cite. Every product here reduces the work of finding it. None of them takes on the professional responsibility for it, and none of them can.
Excluded, and on what grounds
Harvey. Genuinely capable, entirely enterprise, sold by sales call to large firms. It would be dishonest to rank a product whose price we cannot approximate for a readership of solos and small firms.
Casetext’s standalone products. Folded into the Thomson Reuters stack; if you are evaluating them you are evaluating CoCounsel.
Dedicated “AI citation checkers”. A useful category in principle. In practice they verify that a citation exists and resolves, which is the easy half. Whether the case says what your brief claims it says is the half that gets people sanctioned, and it requires reading.
General document automation. Real value for transactional practices, and a different list. Most of it is not AI in any meaningful sense.
Questions lawyers ask
Can I be sanctioned for a citation an AI invented?
Yes, and lawyers have been. Mata v. Avianca produced a $5,000 fine over fabricated cases from ChatGPT, and reported sanctions since have ranged from around $2,500 to $31,100, with one attorney reported sanctioned $86,000 across eight cases. The professional obligation to verify what you file is unchanged by how the draft was produced.
Do the grounded legal research tools hallucinate too?
Yes, less. Stanford’s study of 202 expert-scored queries. Preprint May 2024, peer-reviewed in the Journal of Empirical Legal Studies in 2025. Found roughly 17% hallucination for Lexis+ AI and roughly 33% for Westlaw’s AI-Assisted Research. Both products have changed since and Lexis+ AI no longer exists as tested, so read it as evidence that grounding reduces error rather than eliminating it. Verification remains mandatory at every tier.
Why won’t anyone in legal AI publish a price?
Because the category is sold to firms rather than to individuals, and enterprise sales motions price by negotiation. It is a real cost to you: you cannot compare options without entering several sales cycles. It is also why vLex’s approachable pricing matters more than its feature parity.
Is it safe to paste client documents into ChatGPT or Claude?
Check your jurisdiction and your engagement letters first. Several bar associations have issued guidance on confidentiality and third-party AI processing, and it is not uniform. Consumer tiers, business tiers and API access differ in how data is handled, and the distinction matters more here than in any other profession we cover.
What should a solo buy first?
Practice management, then a grounded research tool you can actually afford, then an assistant. In that order. The billing gap is bigger than the research gap for almost every solo practice, and a research platform you skip because it needed a sales call leaves you doing legal research in a general chatbot, which is the worst outcome on this page.
Will AI replace lawyers?
It is compressing research, first drafts and document review. Real work that real people were paid for. What it does not do is take professional responsibility, which is the thing a client is actually buying and the thing that cannot be delegated to a system that will not be sanctioned. The exposure is to work sold as volume, not to judgment.
Last reviewed: 15 August 2026. Nothing here is legal advice or a recommendation about professional obligations; your bar’s guidance governs. Unusually for us, no price on this page was verified at the vendor’s own site. Legal software vendors either quote by sales call or blocked our requests, and every figure is labelled third-party reported. We re-review quarterly and will replace reported figures with verified ones as vendors publish them.
Leave a Reply