How to choose AI legal software for a small law firm.

Start with your own documents, not the feature list.

The mistake most small firms make is comparing feature lists. Features are easy to demo and hard to trust. What actually decides whether the software works for you is what it learns from: your own past work, or a generic template bank.

A small firm's drafting voice lives in its own documents — its clauses, its tone, its way of structuring a letter or a motion. Software trained on the firm's own past cases drafts in that voice from day one. Software trained on generic legal text drafts in nobody's voice, and the lawyers spend the saved time editing the result back into theirs. So the first question on any demo is not "what can it do" but "whose documents does it learn from."

The second deal-breaker: what happens to your data.

A drafting assistant that learns from your past cases holds your clients' confidences. Three sentences settle it, and a vendor should be able to say all three plainly:

• Your firm's documents are used only to serve your firm. They are kept separate from other firms' data.
• They are never used to train shared models. Your precedents do not improve a competitor's drafts.
• They are protected with encryption in transit and at rest. This should be stated as fact, not buried in a whitepaper.

If a vendor hedges on any of these, keep looking.

Questions to ask on the demo.

Once the two deal-breakers are settled, use the demo to answer practical questions about your own workflow:

• Show me a document type we actually write. Not the demo document — one of yours. A demand letter, an engagement letter, a status letter. If the vendor will not try your documents, the demo proved nothing.
• Where does the lawyer's review happen? Watch the exact moment the draft stops and waits. The approval gate should be visible and explicit, not a promise you take on faith.
• What does the lawyer still do? Good software is honest about this: the lawyer reviews, edits and decides. Be wary of any vendor that presents AI as doing the thinking.
• How does it learn as the firm evolves? Your firm's best clauses change over the years. The assistant should learn from new approved documents, so it tracks your current practice rather than your practice from five years ago.

Run a short pilot on real matters.

A pilot should be short — weeks, not months — and run on document types the firm writes constantly. The success criteria are simple:

1. Do drafts arrive in the firm's own voice, from its own past work?
2. How much review and editing time do they actually save the lawyers?
3. Does every document pass through a visible approval gate before it is final?

If the answer to any of these is unclear after the first few matters, the tool is not the right fit — and you have learned that before paying for it.

Red flags.

• The vendor talks about "AI lawyers." Software is not a lawyer. It drafts; lawyers decide. Vendors that blur that line are selling something other than a tool for your practice.
• No visible approval gate. If drafts can reach a client or a court without a lawyer's sign-off, the product's design does not understand your profession.
• Vague answers about data. "Industry-standard security" is not an answer. You are looking for: used only to serve your firm, never to train shared models, encrypted in transit and at rest.
• The demo never touches your documents. A tool that cannot draft from your own precedents in a demo will not do it in production.

Frequently asked questions.

What should a small law firm check before buying AI legal software?

Check that the software learns from your own past work rather than generic templates, that every document waits for a lawyer's review and approval before it is final, and that your firm's documents stay separate from other firms' data and are never used to train shared models.

Can a small firm use AI legal software on real client matters?

AI drafting software can prepare first drafts for real matters when the firm keeps supervision the way it would with any staff work: the lawyer reviews, edits and approves every document before it is final. Explaw is built around that approval gate — the AI never finishes, signs, files or sends a document.

How long should a pilot with AI legal software take?

A pilot should be short — weeks, not months — and run on real document types the firm writes constantly. You should know by the end of the first few matters whether drafts arrive in your firm's voice and how much review time they actually save.