How small law firms are using AI in 2026

The short answer: small law firms are getting real value from AI on the business side of the practice, not on the law. The jobs that work are answering new client inquiries fast, chasing documents and signatures, drafting client status updates for an attorney to review, scheduling consultations and following up on billing. The jobs that do not work are giving legal advice, drafting filings, and anything that would push a client confidence into a tool the firm has not vetted. Those are not style preferences. Model Rule 5.5 and ABA Formal Opinion 512 draw the lines, and most vendor pages skip past them.

Start with the rule, because it decides everything else

Every state limits the practice of law to licensed attorneys, and the American Bar Association states it in Model Rule 5.5: a lawyer shall not practice law where doing so breaks the profession's regulation, or assist a nonlawyer in doing so. The practice of law includes giving legal advice, representing a client, and preparing documents that require legal judgment. For this purpose software is a nonlawyer. That single line is the boundary every AI decision in a firm has to respect.

So the useful question is not whether AI can answer a client's legal question. It usually can produce something that reads like an answer, and that is exactly the trap. The moment a tool tells a client how to handle a lease dispute or whether to accept a settlement, it has edged into the unauthorized practice of law, and the responsibility does not disappear because a machine wrote it. The firms doing this well look boring from the outside: they run unglamorous automation on work that carries no legal content at all.

What ABA Formal Opinion 512 actually says

On July 29, 2024 the ABA issued Formal Opinion 512, its first ethics guidance on generative AI. It did not create new rules. It applied the ones already in force: competence under Rule 1.1, confidentiality under Rule 1.6, communication under Rule 1.4, candor toward the tribunal under Rules 3.1 and 3.3, supervision under Rules 5.1 and 5.3, and reasonable fees. If your firm writes a one-page AI policy this year, that opinion is the thing it should cite.

The duty that governs day-to-day use is confidentiality. Rule 1.6 covers all information relating to a representation, whatever its source, and the opinion is explicit that a lawyer must evaluate whether a given tool protects that information before feeding client data in, and in some situations must get the client's informed consent first. In plain terms: confidences stay inside systems you control and have vetted, and they do not go into a general chatbot because it was convenient.

How small firms are using AI, specifically

Intake response is the job with the clearest return, because it is where small firms quietly lose work. Legal consumers call more than one firm, and the one that answers first and books the consultation usually gets the matter. A solo in court or a two-lawyer firm with everyone busy sends the inquiry to voicemail, and it goes to the next name on the list. An agent that replies in minutes, qualifies the inquiry against the firm's criteria, and offers a consultation slot turns a lost lead into a booked consult.

Document and signature chasing is the near-universal second job. At any moment a small firm has matters stalled on something the client has not sent: an unsigned engagement agreement, an unpaid retainer, a records release, an ID verification. Historically someone works that list when a gap appears in the day, which means it slips. The message carries no legal content, so the agent can chase it on a cadence, vary the third attempt, and flag the client who has gone silent for a partner to call.

Then there is client communication and money. The where-are-we-on-my-case email arrives constantly and rarely gets answered quickly. An agent drafts the update from the matter record and queues it for an attorney to approve, so clients feel informed without a partner writing the same note ten times. On billing, aged invoices and trust accounts that need replenishing get followed up on a schedule, in the firm's voice, instead of at quarter-end when the number has grown uncomfortable. The full job-by-job version of this lives on the AI for law firms page.

What small firms are not using AI for

Legal advice. When a client asks what they should do, that answer is the practice of law and it needs a licensed attorney's judgment and name on it. The behavior you want from an agent is to recognize a legal question, route it to the lawyer who owns the matter with the file history attached, and send a holding reply that confirms receipt without saying anything substantive.

Drafting filings and legal documents. Anything that requires legal judgment or gets submitted to a court is work product the attorney owns. And legal research sits in the same bucket: the moment analysis shades into advice, it belongs to a lawyer. If a firm wants a tool for that side of the work, it should be a purpose-built one that turns a question into legal research in plain English for an attorney to check, not a general work agent bolted onto client files.

Anything that exposes a confidence to an unvetted tool. This is the Rule 1.6 line, and it is the one most firms cross by accident when someone pastes a client email into a consumer chatbot to "just get a quick draft." The safe workflow is built so the tasks the agent handles do not require exposing a confidence in the first place.

Where the line sits, job by job

Is it ethical for lawyers to use AI?

Yes, and Formal Opinion 512 settles the question directly: lawyers may use generative AI provided they meet their existing duties and supervise the tool as nonlawyer assistance under Rule 5.3. That supervision framing is the right way to think about buying one. You would not let a new paralegal send a substantive client answer unread, so you do not let the agent either. The agent drafts anything that carries meaning; a person sends it. The agent handles end to end only the messages that carry no legal content: the document is missing, the consultation is confirmed, the invoice is past due.

Will AI replace paralegals or lawyers?

Lawyers, no, and the rules are a large part of why. Advising the client, exercising legal judgment and taking responsibility for the outcome are activities the profession attaches to a license. Paralegals, not wholesale either. AI is absorbing the administrative part of the role, intake, chasing, scheduling and record keeping, which frees a good paralegal for cite-checking, fact organization and client management, the legal-adjacent judgment that still needs a trained human. The people who feel real pressure are the ones whose service was purely administrative.

How to start without creating a problem

Pick one narrow, repetitive, non-legal job and get the format right before adding a second. Intake response or document chasing, nothing else, for two weeks. Brief the agent the way you would brief a new legal secretary: how fast inquiries get answered, which matters get a phone call instead of an email, how many chase attempts before escalation, how matter notes should be structured, and what never goes out without an attorney reading it.

Then write the boundary down in a one-page policy anyone in the firm can read: no legal advice from a tool, no filings, and no confidence leaves your vetted and consented systems. Have your own counsel look at it, because this article is general information rather than legal or ethics advice. From there the additions are obvious. The same agent that answers intake can run the inbox triage, keep your matter records current and assemble the same status report every week. The conventions you set in week one carry across all of it, which is the actual reason this compounds.