AI Receptionists for Law Firms: Intake Setup and ABA Opinion 512

The useful work is narrow and repetitive: answering on the first ring at 11 p.m., asking the same intake questions every time, and getting the caller onto an attorney's calendar. Those are the tasks a tired person is most likely to rush at the end of a long shift.

The line it must not cross is anything that sounds like legal judgment. 'Do I have a case?', 'What is it worth?' and 'Should I talk to the adjuster?' all get the same answer: an attorney will review your information and call you back by a stated time.

This article is general information, not legal advice. Your state's rules of professional conduct and ethics opinions control, and they vary.

Competence (Rule 1.1). Opinion 512 says lawyers need not become AI experts but must reasonably understand what a tool can and can't do. For intake, that means knowing how the system decides to escalate and how often it mishears names, dates and places.

Confidentiality (Rule 1.6, with Rule 1.18(b) extending similar protection to prospective clients). Before information goes into a generative AI tool, the lawyer must weigh the risk of disclosure inside and outside the firm. For self-learning tools, the opinion concludes that client informed consent is required before inputting information relating to a representation, and that boilerplate consent in an engagement letter isn't enough. That makes 'does the vendor train models on our callers' data?' the first diligence question, not the last.

Supervision (Rules 5.1 and 5.3). Managing lawyers must set clear policies on permitted AI use and train staff. For outside providers, the opinion points back to earlier cloud and outsourcing guidance: check the vendor's security, retention and liability terms, and make sure you'd be told about a breach.

The opinion's footnotes also cite Florida Bar Ethics Opinion 24-1, which discusses chatbots under Florida's rules against misleading or unduly manipulative advertising. Florida firms should read the two together.

Model Rule 1.18 treats anyone who consults a lawyer about possibly forming a client-lawyer relationship as a prospective client, and bars the lawyer from using or revealing what they learned even if no representation follows. The rule's comment advises limiting the initial consultation to what reasonably appears necessary to decide whether to take the matter. An AI flow should follow the same discipline.

The comment also says a consultation is more likely when a lawyer invites information without clear warnings that limit the lawyer's obligations. A short, plain statement at the start of the call does real work.

Notice what the flow below leaves out: detailed injury descriptions, who was at fault, treatment specifics and what the other driver said. Those belong in the consultation, after conflicts are cleared. Collecting them earlier adds risk under Rule 1.18 without improving the callback decision.

Use this as a starting checklist. Adjust the wording to your state's rules and your firm's voice, and keep the approved version under change control so nobody edits the prompt casually.

Federal wiretap law, 18 U.S.C. 2511(2)(d), makes recording lawful when the person recording is a party to the call or one party has consented, unless the purpose is criminal or tortious.

Some states are stricter. California Penal Code 632 prohibits recording a confidential communication without the consent of all parties. Injury callers routinely describe medical details they expect to stay private, so treat intake calls as confidential and tell every caller about recording up front.

Outbound calls are a separate question. In a February 2024 declaratory ruling, the FCC confirmed that the Telephone Consumer Protection Act's restrictions on 'artificial or prerecorded voice' calls cover current AI technologies that generate human voices. Those calls need the called party's prior express consent unless an exemption applies. If your system places callbacks or reminder calls, ask for that consent during the inbound call and log the answer.

Opinion 512 borrows its vendor checklist from earlier guidance on cloud computing and outsourcing. Turn it into written questions and keep the answers on file. That record is part of showing reasonable supervision if anything goes wrong.

Then test the answers. Ask for a sandbox number, call it with your own test script, request the recording and transcript, delete them, and ask for confirmation that they are gone from backups too. How a vendor handles that request tells you more than its sales deck.

Many PI firms already run intake through a legal CRM such as Clio Grow or Lawmatics, or a general platform like GoHighLevel. The receptionist should create or update a lead there, attach the transcript and recording, set the matter type and flags, and assign the callback to a named person. Calendar booking needs its own limits: only slots attorneys have opened for consultations, buffer time after each call and no double-booking.

Follow-up is where most intake systems leak. Our guide to automating lead follow-up covers the text and email sequences that run after the call, and offline conversion tracking shows how to send signed-case outcomes back to your ad platforms, so you can see which campaigns produce clients and not just calls.

Fields worth mapping from day one:

Set a baseline first. Pull a month of call logs before launch so answer rate by hour and time to callback have something to compare against; without it, every improvement is a guess.

Review a sample of transcripts every week for the first month. You are looking for two failure types: calls it should have escalated and didn't, and moments where it drifted toward advice. Fix the prompt or the flow, then re-test before the change goes live.

For a sense of what custom conversational AI costs to build and run, see our breakdown of AI chatbot costs for small businesses. If you want help designing the flow or building on your own stack, that is the work our AI solutions team does.