The Legal AI Buying Guide Most Firms Needed Before Casetext Disappeared

If you're comparing ai tools for law firms, the obvious questions are feature lists and demos. The more important questions are less flattering: What happens if the product gets absorbed into a larger platform? Can you see pricing before a sales call? Will it give you traceable citations? And are attorneys already feeding client data into a consumer chatbot with the default settings still on?

Casetext is the cautionary example. Many firms picked it because it offered useful research help without forcing a full Westlaw commitment. After Thomson Reuters acquired Casetext in 2023, the standalone product was folded into CoCounsel and the wider Westlaw ecosystem. For firms that chose Casetext to avoid a larger subscription stack, that was a forced workflow change, not a minor update.

This guide focuses on the parts most legal AI roundups skip: pricing opacity, acquisition risk, citation reliability, and confidentiality policy.

Casetext Wasn't Just Acquired — It Changed the Risk Model

Thomson Reuters acquired Casetext in 2023 and integrated its technology into CoCounsel. That is established fact. The practical effect for existing customers was also clear: firms lost a standalone option and had to either stay inside the Thomson Reuters stack or switch tools.

That matters because legal software buyers often evaluate features and ignore ownership risk. They shouldn't. If a product exists mainly as an acquisition candidate, your workflow can disappear into someone else's bundle with very little notice.

By contrast, Paxton AI publishes pricing publicly and offers a short trial, which suggests a self-serve growth strategy rather than pure enterprise gatekeeping. Vincent AI sits inside vLex, and vLex itself has since become more tightly tied to Clio after Clio announced its acquisition of vLex for $1 billion. That does not automatically make Vincent a bad choice. It does mean your research stack may become more dependent on another company's platform priorities.

The takeaway is simple: ownership structure is part of product evaluation.

The Real Pricing Problem: You Often Can't Budget Until After the Demo

Most legal AI buying guides pretend the market is transparent. It isn't.

Many of the best-known tools still require a sales conversation before they reveal meaningful pricing. That is manageable for an AmLaw firm with procurement staff. It is a waste of time for a 5-lawyer or 20-lawyer practice trying to compare options before calendar year budgeting.

Reported enterprise pricing for Harvey AI starts around $1,200 per user per month, with no public self-serve signup or public trial. That figure is widely reported in legal AI market coverage, but buyers should still treat final pricing as quote-dependent.

LegalOn is often described as accessible, but public sources do not provide a dependable entry-level number for law firm buyers. Based on how the company sells, it belongs in the enterprise-evaluation category, not the impulse-try category.

Paxton AI stands out because it publicly lists pricing and offers a 7-day trial. For a solo or small firm, that removes a lot of wasted evaluation time before the first substantive test.

Pricing Table: What You Can Verify Before Signing Anything

One pricing detail that matters: Relativity aiR is no longer usually treated as a separate line item. If your team already pays for Relativity, some AI functionality may already be included in the subscription.

Where Most Tools Stop: Research and Drafting, Then Nothing for Trial

A lot of legal AI products perform well in the pre-trial stages: research, summarization, first-pass drafting, document review. Fewer are built around what litigators actually need next.

NexLaw positions itself around that gap. The company's product materials emphasize litigation workflow coverage beyond research, including trial-prep functions and chronology work. For a personal injury practice dealing with medical records and timelines, that is more specific than the generic "AI for litigators" pitch you see elsewhere.

That doesn't prove NexLaw is the best option for every litigator. It does show a real product distinction. If your bottleneck is witness prep or chronology assembly, a contracts-first assistant such as Spellbook solves the wrong problem.

Hallucinations Aren't an Embarrassment Problem Anymore

Fake citations are now a professional-risk issue, not a social-media joke.

Federal court sanctions over fabricated citations are documented. Courts have already dealt with filings that included non-existent cases or misrepresented authorities generated with AI assistance. The legal significance is not that one tool hallucinated and another never will. The significance is that lawyers now need tools and workflows built around verification.

That changes how to compare products:

• Ask whether the system returns linked, checkable authorities.
• Ask whether the quote or summary can be traced back to source text.
• Ask whether attorneys can review the cited material without leaving the workflow.

Reported benchmark coverage cited in legal AI market analysis has placed Vincent AI at roughly 58% on Stanford's LegalBench-style legal evaluation references, often framing it as a strong accuracy-per-dollar option in its tier. That should be read carefully: not as proof that it is "accurate enough" in the abstract, but as a cost-to-performance signal.

General-purpose ChatGPT should not be your filing engine. It can help with internal structuring, issue spotting, or rewriting. It should not be treated as a citation authority for anything that reaches a court without line-by-line verification.

The Confidentiality Problem Starts With Consumer Defaults

One of the easiest mistakes to make is also one of the least discussed: attorneys using consumer AI accounts for client-related work without checking the data policy first.

Consumer ChatGPT accounts have historically allowed conversations to be used for model improvement by default, unless the user changes the relevant setting. Policies can change, so firms should verify the current OpenAI terms and workspace settings directly before use. The risk is straightforward: if a lawyer pastes client facts, draft communications, or contract language into a personal account without proper controls, confidentiality issues follow.

Claude Team and enterprise workplace products such as Microsoft 365 Copilot are typically sold with stronger commercial privacy commitments than consumer chatbots. That does not make them automatically safe for every matter. It does mean the contract and admin controls usually matter as much as the model itself.

Before anyone compares output quality, the firm should answer three boring questions:

1. Is user data retained?
2. Is it used for training?
3. Can an admin control that at the workspace level?

If the answer to any of those is unclear, the tool is not ready for client work.

Adoption Is Up. Governance Is Behind.

According to the 2026 FTI Consulting and Relativity General Counsel Report, in-house generative AI adoption reached 87%, up sharply from the prior year. That is a real signal that legal teams are already using these systems.

It is not proof that legal teams are using them well.

BCG and similar consulting research have repeatedly identified output trust and review procedures as major barriers to broader deployment. The likely explanation is simple: buying access is fast, but creating review rules, training staff, and writing matter-specific policies is slow.

So the number worth caring about is not adoption. It is supervised adoption. A firm with five approved use cases and a documented review process is in better shape than a firm where everyone is experimenting in private.

A Caution Flag Most Roundups Skip

Robin AI appears in many legal AI lists because it is well-known and easy to mention. Some independent market roundups, including reporting from thelegalprompts.com, have attached a buyer-caution label to it.

Publicly available explanations for that caution are limited, so this should not be presented as a settled verdict against the company. What it does show is that firms should not confuse visibility with certainty. If a tool appears everywhere, that may reflect strong marketing, not strong fit for your workflow.

When a roundup includes a product but offers no details on pricing, implementation, citation controls, or data handling, treat the recommendation as incomplete.

How to Pick the Right Stack for Your Firm

The wrong way to buy legal AI is to ask which platform is best overall.

The better question is: where does your firm lose the most expensive hours?

A transactional team spending its week in Microsoft Word has different needs from a PI firm building timelines, reviewing medical records, and preparing examinations. That sounds obvious, but many firms still buy based on brand familiarity instead of workflow fit.

A practical filter looks like this:

The other useful rule: do not start with a broad assistant if the real pain is narrow and repetitive. A contract-heavy practice does not need a giant research suite first. It needs better redlining speed and fewer manual clause comparisons.

FAQ

Is Harvey AI realistic for a small firm?

Usually no. Reported pricing starts around $1,200 per user per month, and the product is sold through enterprise-style procurement. For most small firms, that alone takes it out of contention unless a client or practice area generates unusually high value from the workflow.

What replaced Casetext?

Casetext was absorbed into Thomson Reuters' CoCounsel and Westlaw ecosystem after the acquisition. Firms that depended on Casetext as a standalone option generally had to migrate into that ecosystem or move to alternatives such as Paxton AI or Vincent AI.

Which option is easiest to price without a sales call?

Paxton AI is one of the clearest examples because it publishes pricing and offers a 7-day trial. Vincent AI is also easier to budget than many enterprise-only competitors because public pricing is available for at least some packages.

Can attorneys use ChatGPT for client work?

Only with serious caution and only after checking the current data controls. A personal consumer account is the riskiest setup because settings and retention rules may not match the firm's confidentiality obligations. For filed work, a legal-specific system with traceable authorities is the safer default.

Do any major legal AI products offer a real free plan?

Most do not. Trials are more common than permanent free tiers. That distinction matters because a 7-day trial helps evaluation, but it does not reduce long-term budget risk.

What To Check Before You Approve Any Tool

Before the next demo, ask vendors for four specific answers in writing:

• What is the exact starting price per user or per matter?
• Are outputs linked to source authorities?
• Is customer data excluded from model training by contract?
• What happens to the product if ownership changes or the feature is absorbed into another tier?

That is the checklist that most law firms skip when comparing ai tools for law firms. It is also the checklist most likely to save you from paying enterprise prices for a product that either cannot be verified, cannot be budgeted, or cannot survive a platform reshuffle.