Legal research software.
Four categories of product get sold under one label. They solve different problems and are priced on different logic.
This page maps the landscape, prices what is publicly priced, and says which category fits which practice.
Last reviewed 2026-08-24
The four categories
The research incumbents
Westlaw and LexisNexis. Case law, statutes, regulations, an editorial citator, and a deep secondary source library. The citator and the treatises are the moat, not the search box.
Priced: Westlaw publishes self-serve rates for firms up to ten attorneys, roughly $107 to $400 per month depending on tier and coverage. LexisNexis solo entry reported near $114 per month. Above ten attorneys, both quote.
AI-native platforms
Harvey, Legora, Paxton, and the generative tiers the incumbents bolted on. Built around a model rather than an index, with the corpus attached underneath.
Priced: mostly not. One survey of twenty legal AI vendors in July 2026 found six publish a price and fourteen route to a sales call. Harvey does not publish; industry reporting puts it near $1,000 to $2,000 per seat per month for mid-market firms. Legora is reported near $3,000 per user per year with a ten seat minimum.
Free and bar-benefit tiers
CourtListener, Google Scholar, and the Fastcase or vLex membership benefit provided by more than 40 state bars. Real case law coverage at zero direct cost.
Priced: free. The tradeoff is the citator. Free tiers generally exclude the editorial treatment layer and the advanced AI features, which moves verification time from the invoice to the timesheet.
Practice management add-ons
Clio Duo, Vincent inside Clio Work, and the AI features attached to case management systems. These run the firm. They are not research platforms and do not claim to be.
Priced: Clio Duo reported at $49 to $59 per month on top of base Clio pricing, with bundles reported at $119 to $179. Vincent bundled into Clio Work reported near $199 per user per month.
Category four is not a competitor to category one. A practice management system runs the matter. A research platform answers the legal question. Firms that try to replace one with the other end up with neither, and that mistake shows up in renewals more often than any pricing error.
The silo problem underneath all four
Every platform in categories one, two, and four searches its own licensed content. Westlaw searches Westlaw. Lexis searches Lexis. Clio acquired vLex for one billion dollars in November 2025, and Clio users search vLex without getting Prosser, Wright and Miller, or Collier unless they subscribe separately.
The consequence is that coverage is a purchasing decision rather than a research decision. A firm that needs authority across publishers pays more than once, and a firm that cannot pay twice has a blind spot it cannot see from inside the tool.
Citation relationships in the public record are not owned by any publisher. Mapping them from public court records is what lets a system cross the silos rather than sit inside one. Full breakdown of the silo problem.
The pricing transparency gap
A July 2026 survey checked twenty legal AI vendor websites for a price a buyer could read without a sales call. Six published one. Fourteen did not. A separate June 2026 benchmark of the ten platforms in-house teams most often shortlist found only three publish a real per-seat rate.
Published rates in the category span from zero for self-hosted open source to roughly $2,999 per user per year. That is not a range. That is the absence of a market price.
The reason a vendor withholds the number is structural rather than sinister. A posted price lets a buyer compare and walk away. A quote establishes the buyer's firm size and budget before the seller names a figure. Both parties know this, which is why the demo call opens with questions about headcount.
The consequence for a small firm is that comparison shopping requires sitting through multiple sales cycles to learn four numbers. That cost is real and it is why solos frequently either overpay for an enterprise tool or avoid the category entirely.
The reliability variable
A Stanford RegLab and Yale study published in 2024 tested leading paid AI legal research tools for fabricated or unsupported output. The reported findings put the Westlaw AI-assisted research product near one inaccurate response in three, the weakest of the paid tools tested, with Lexis performing better and still producing hallucinated content at a measurable rate.
That result is the most important variable in the category and it appears on no pricing page. A generative system asked to produce a citation can produce a plausible one whether or not the case exists. Model quality reduces the rate. It does not change the shape of the failure.
The architectural alternative is retrieval-first: the system cannot return a citation absent from a verified index. That constraint sits below the model rather than beside it, which is why it holds regardless of how the model is prompted. This is the Mata problem and it is distinct from the privilege question in United States v. Heppner.
The running record of what happens in court when the constraint is absent is maintained at heppner-problem.com. As of mid 2026 the underlying public database tracked more than 1,500 court decisions worldwide involving AI generated material, with sanctions escalating from four figure fines to license suspensions, including the first indefinite suspension in Nebraska in April 2026.
Choosing by practice shape
| If your practice looks like this | Start here | Why |
|---|---|---|
| Litigation team, KeyCite or Shepard's daily, appellate work | Research incumbent | The editorial citator and treatise library are the product. No alternative reproduces four decades of classification. |
| Solo or two attorneys, single jurisdiction, occasional research | Bar benefit plus a published-price tool | Check the bar tier first. Pay only for the citator and workflow the free tier does not cover. |
| Small firm, Microsoft 365 already deployed, document-heavy | Retrieval-first tool inside M365 | No new vendor approval, no third party holding client data, and the research sits where the document already is. |
| In-house team, contract-heavy, low case law volume | Word-native contract assistant | Case law depth is not the constraint. Drafting throughput is. |
| Am Law 100 with a research department | Incumbent plus AI-native platform | Budget supports both, and the research department absorbs the tool switching cost that kills this configuration at smaller firms. |
| Firm already running Clio or MyCase | Add research, do not replace management | Practice management and legal research are separate jobs. Replacing one with the other produces gaps in both. |
Where Marcella sits
Marcella is a retrieval-first legal intelligence platform that runs inside the firm's existing Microsoft 365 tenant. It is category one architecture at category three transparency, and it is deliberately not a practice management system.
Citations are retrieved from an indexed corpus of more than 10 million judicial opinions built on CourtListener, the Harvard Caselaw Access Project, 50-state statutory codes, the full eCFR, CRS reports, SEC EDGAR, IRS publications, and bar ethics opinions from 51 jurisdictions. Not in the index, not in the output. Every output carries a six-item Daubert attestation.
The citation treatment engine maps how courts have cited and applied authority across jurisdictions from public court records, covering 26 canonical treatise families and 14 ALI Restatements. That is what crosses the publisher silos.
Client data never leaves the firm's Microsoft tenant. No external AI vendor sits in the data path, there is no training on queries or documents, and PII is stripped by Presidio anonymization. On the BOYA tiers the deployment runs in the firm's own Azure tenant.
Every rate is published: $99 per seat per month for 1 to 25 seats, $278 per month flat for 3 seats on BOYA, $1,498 per month flat for 20 seats on BOYA 20, custom above that. This is launch pricing and rates increase after the launch period. Full detail on the pricing page.
What Marcella does not offer: a licensed treatise library, litigation analytics on judges and motions, or four decades of editorial citator classification. Those are real gaps and they are why some firms should keep an incumbent subscription.
Questions
What is the best legal research software for a solo attorney?
Start with the free tier your state bar already provides. More than 40 bars include Fastcase or vLex with membership. Add a paid tool only for what the free tier does not cover, which is usually the editorial citator and the document workflow. Published entry rates in the paid category start near $99 per seat per month.
How much does legal research software cost?
Published rates span from zero for free and bar-benefit tiers to roughly $2,999 per user per year at the top of the published range, with quote-only enterprise platforms reported considerably higher. Most vendors do not publish. One July 2026 survey of twenty vendors found six that post a readable price.
Is AI legal research reliable?
It depends entirely on architecture. A Stanford RegLab and Yale study reported that leading paid generative research tools produced fabricated or unsupported output at measurable rates, with one product near one in three responses. Retrieval-first systems constrain output to a verified index, which removes the failure mode rather than reducing its probability.
Can free legal research replace Westlaw?
For finding cases, largely yes. CourtListener and Google Scholar carry broad coverage of federal and state appellate opinions. For verifying them, no free option includes a full editorial citator, so the verification work moves to the attorney. The complete free method is on the good law verification page.
Do I need legal research software if I use Clio or MyCase?
Yes. Practice management systems run the matter: intake, calendaring, billing, documents. They do not answer the legal question or verify the citation. The two categories complement each other and substituting one for the other leaves gaps in both.
Marcella runs inside the Microsoft 365 tenant your firm already has. No new vendor approval required.
Join the Waitlist