AI for Small Law Firms: What to Buy First and What to Skip

For a small law firm, the right first AI purchase is almost always case law research, not a firm-wide platform. Research is where the unbillable hours actually go, it is the one category with published per-seat pricing (roughly 29 to 30 dollars a month at the self-serve end), and it is the only category you can evaluate properly in an afternoon. The enterprise platforms that dominate the coverage are not sold to you: reported seat minimums around twenty, with twelve month terms, decide that before price does.

Most writing about AI for law firms is written with a two hundred lawyer firm in mind. That firm has a legal ops function, a procurement process and a budget line for software. A six lawyer firm has a managing partner who also handles the payroll, and the advice does not transfer. This is the version for the second firm.

What is the best AI for a small law firm?

Start with case law research. Not because research is the most exciting thing AI does, but because it is the task where a small firm's economics are worst. Research time is frequently written down or written off entirely, especially on fixed-fee and contingency work, so an hour saved there converts straight to margin rather than to a smaller invoice. Drafting and review matter too, but they usually sit on top of billable work that a client is already paying for.

The second reason is that research is the one category with honest published pricing. Cases publishes 29 dollars per user per month and midpage publishes 30. Paxton AI publishes 499. Westlaw Advantage publishes 256.75 per user per month for a single circuit and 399.75 for all states plus federal, for firms up to ten attorneys. You can compare those figures without talking to anyone. In most other legal AI categories you cannot, and a category where nobody publishes a price is a category where a small buyer has no leverage.

Can a small firm use Harvey AI?

Not in practice, as of August 2026. Harvey is an enterprise legal AI platform sold to large firms and corporate legal departments, and it publishes no price at all. Its pricing URL returns a 404, there is no free tier and no self-serve signup, and the only route in is a demo request. Reports consistently describe a seat floor of around twenty seats with a twelve month term, which ends the conversation for a firm with four lawyers regardless of what the per-seat number turns out to be.

That is not a criticism of the product. Harvey's published security posture is genuinely strong, and its Vault and Knowledge modules work over a firm's own document set, which is a hard problem and a real reason large firms pay for it. It is simply built and priced for a different buyer. We go through the evidence in detail on our Harvey AI page, including what is actually published versus what is only repeated.

The general rule this illustrates is worth holding onto. If a vendor has no pricing page, no trial and a sales-led motion, you can usually tell within thirty seconds whether you are the intended customer, and you can stop evaluating.

What AI tasks actually pay off for a small firm?

In rough order of return for a firm under about fifteen lawyers:

• Case law research and precedent finding. The highest-value first purchase. Ask a question, get on-point authority with citations you can open. Cheap, published pricing, immediately testable.
• Citation and good-law checking. Finding a case and knowing it is still good law are different jobs. KeyCite and Shepard's remain the two established citators and neither is cheap, but this step is not optional if you are filing.
• First-draft briefs and memos. Real time savings, with the important caveat that every citation in the draft has to be independently confirmed before it goes anywhere near a court.
• Contract review against a playbook. Worth it if transactional work is a meaningful share of your practice. Not worth it if you are a litigation shop.
• Deposition and transcript summaries. Genuinely useful in litigation, and easy to spot-check because the summary carries page and line references back to the record.
• Intake and admin. Real savings, but this is practice management software rather than legal AI, and it is a separate buying decision.

Notice what did not make the list: document review and discovery ranking. Those platforms are priced for matters with document sets large enough to justify them, and most small firm matters are not. Renting that capability per matter when a big case actually lands is nearly always cheaper than carrying it.

Buy for the task, not for the category

The most expensive mistake a small firm makes with AI is buying one general assistant and expecting it to cover everything. The vendors do not actually compete with each other evenly. A contract review tool and a case law research tool have almost nothing in common except the marketing word, and a platform that does both usually does one of them well and the other adequately.

This applies well beyond legal AI. Firms with an insurance defense or construction practice, for example, often carry a separate administrative burden around tracking certificates of insurance and their compliance dates, and that is its own software category rather than something a legal research platform will ever touch. Trying to solve unrelated problems with one purchase is how firms end up paying for a platform and still doing the work by hand.

The practical version: write down the three tasks eating the most unbilled hours this quarter. Buy the best tool for the top one. Re-check in ninety days. That sequence beats any platform evaluation, and it costs almost nothing to run.

What does AI for a small law firm cost?

At the research end, 29 to 30 dollars per user per month buys a published, cancellable, self-serve tool. That is the floor, and for a lot of small firms it is also the ceiling they need. A full research platform with agentic AI runs from about 257 to 400 per user per month on Westlaw Advantage's published rate for firms up to ten attorneys. Paxton AI publishes 499.

Above that, the market goes quote-only. LexisNexis, Bloomberg Law, CoCounsel Legal, vLex and Harvey all publish nothing, which means every number you see attached to them online is a third-party estimate. Those estimates routinely disagree with each other by a factor of ten, so treat them as noise rather than as budget input. Our legal research pricing comparison keeps the published and reported figures in separate columns for exactly this reason.

One thing worth negotiating even at small firm scale: ask for the renewal rate and the term length before you sign, not the first-year discount. Legal research contracts commonly run two to three years, and a published monthly rate you can cancel is worth real money against a cheaper long term.

What are the ethics rules for a small firm using AI?

The same rules as everyone else, and they are not onerous. ABA Formal Opinion 512, issued 29 July 2024, maps the existing Model Rules onto generative AI rather than creating a new regime: competence, confidentiality, client communication, candor to the tribunal, supervision and reasonable fees. Nothing in it prohibits the technology.

Two of the six bite harder at a small firm than at a large one. Confidentiality, Rule 1.6, requires you to know how a tool handles your input, and the opinion says you should generally get informed client consent before putting client confidences into a generative AI tool, with boilerplate engagement-letter language expressly not good enough. Large firms have an information security function to work this out. A small firm has to ask the vendor two questions in writing: do you train on my data, and how long do you retain it.

Fees, Rule 1.5, is the other one. If a research task that used to take three billable hours now takes twenty minutes, the reasonable fee is twenty minutes, and you cannot bill the client for the time you spent learning the software. Firms that treat AI as a margin lever rather than a capacity lever are the ones most likely to end up in a fee dispute. Our full walkthrough is in can lawyers use AI for legal research.

The one habit that prevents almost every reported failure

Never put a citation in a filing you have not personally opened. That is the whole thing.

The public database of decisions in which a court found reliance on hallucinated material held 1,868 cases as of its 8 August 2026 update, and they are, almost without exception, that single failure repeated. In the six federal decisions collected by Norton Rose Fulbright from February to April 2026, not one attorney was sanctioned for using AI. Every one was sanctioned for filing authority nobody read. In the worst of them, Whiting v. City of Athens, the Sixth Circuit ordered counsel to pay the appellees' appellate fees, double costs, and 15,000 dollars each in punitive sanctions.

Small firms are more exposed to this than large ones for a structural reason: there is often no second reader. A big firm's brief passes an associate, a senior associate and a partner. A solo's brief passes the solo. That makes the verification step a workflow you have to build deliberately rather than one you can rely on the org chart to supply.

Where to start this week

Pick the research tool, run it against three questions you already know the answer to, and check whether the cases it returns say what it claims they say. That test takes twenty minutes and tells you more than any feature list, because it distinguishes a tool that retrieves from real case law from one that generates plausible text. Do the same thing with any vendor you are considering.

If the shortlist stage is where you are stuck, our guide to legal AI for lawyers sorts the market by task with published pricing, and the LexisNexis alternatives and Westlaw alternatives pages cover what you keep and what you give up if you are already under contract somewhere.

Cases is built for that first task and nothing else. Ask a research question in plain English, get on-point US federal and state precedents with the holding and a citation you can open and read. Published at 29 dollars per month, no quote and no seat minimum. We do not run a citator and we say so. Informational research, not legal advice, always verify the citation.