The DUI arrest call comes at 2am. The custody emergency comes on a Sunday. Legal need does not keep office hours, and the firm that answers first usually signs the case. This is the honest guide to what an AI receptionist does on a legal intake call, the ethics questions you must ask before deploying one, what it should never do, and what it costs.
What it does: answers every call in seconds, identifies itself as AI, runs your intake script — new-client details, conflict-check data, case-type screening — and books the consultation into your calendar, at 2pm or 2am.
The ethics line: deployed with lawyer supervision (ABA Model Rule 5.3 territory), upfront AI disclosure, zero legal advice, and vetted data handling, an AI receptionist is a nonlawyer assistant like any other — just one that never sleeps. Confirm the specifics with your state bar.
What it costs: $29–$99/mo DIY platforms you build yourself, $250–$600/mo for a traditional message-taking answering service, or from $800/mo for a custom agent a team builds, integrates, and operates for you.
"It answers the phones" undersells what a well-built intake agent does. Here is the call, step by step.
New-client intake. The agent answers inside one ring, discloses that it is an AI assistant, and runs your intake script one question at a time: name and contact details, the type of matter in the caller's own words, key dates, whether they have spoken to other firms. The output is not a scribbled message — it is a structured intake record in your case management system before anyone on your team has coffee.
Conflict-check data collection. This is the piece generic answering services cannot do and most DIY bots skip. The agent captures the opposing party's name and the related parties your conflict-checking process requires, so your team can run the check before the consultation — not discover the conflict in the middle of it. It collects the data; it does not decide the conflict. That judgment stays with a lawyer, where it belongs.
Consultation booking. Once the matter type passes your screening rules, the agent offers real openings from your actual calendar, books the consultation, and sends a confirmation text with anything the caller should bring. If you charge for consultations, it says so, using your approved language.
The after-hours emergency. The 2am DUI arrest. The Sunday custody emergency. The workplace injury on a holiday. These are the highest-value calls a firm gets, and they go to whoever answers. The agent picks up in seconds, captures the arrest or incident details and the deadlines that matter, and executes your protocol: hot-transfer to the on-call attorney, or a committed callback window with a confirmation text to the caller. A voicemail box offers a beep. The family calling at 2am does not leave a message — they call the next firm on Google.
If you want the broader industry view — the missed-intake math, the integration list, the ROI model — our AI for law firms page has it. This post stays on the questions lawyers actually ask us: is this allowed, and is it safe?
It can be — but "can be" depends on how it is deployed and who vetted the vendor. Nothing in this post is legal advice; ethics rules and bar opinions vary by state, so treat this as the question list to bring to your vendor and confirm with your bar association before you sign anything.
ABA Model Rule 5.3 — the duty to supervise. Rule 5.3 makes lawyers responsible for the conduct of nonlawyer assistance they use, and bar guidance generally reads that as covering vendors and technology, not just employees. Practically, that means an AI receptionist is not a "set it and forget it" appliance. It needs the same things you would give a new human intake hire: a script you approved, boundaries you defined, spot-checks of its work, and a clear escalation path when it hits something outside its scope. Ask any vendor: who reviews the agent's calls, how often, and how fast do corrections ship? If the answer is "you, whenever you get to it," the supervision duty is yours alone.
Attorney-client privilege and confidentiality. An intake call is a prospective-client communication, and in most jurisdictions prospective clients get substantial confidentiality protection even if they never hire you. So ask where the call recordings and transcripts live, who at the vendor can access them, whether they train models on your calls, how long data is retained, and what happens to it if you leave. Some vendors will sign business-associate-style data agreements; others will shrug. The shrug is your answer.
Accidental attorney-client relationships. A caller can form a reasonable belief that a firm represents them from shockingly little — which is why the agent's script matters. It should never say anything like "we'll take care of your case" or imply representation has begun. It collects information and books a consultation. The engagement decision happens with a lawyer, after a conflict check, every time.
Advertising and solicitation rules. Your state bar's advertising rules apply to what the agent says just as they apply to your website: no misleading claims, no guarantees of outcomes, required disclosures where applicable. A vendor building agents for law firms should be able to talk about this fluently. If the sales call is all demos and no compliance vocabulary, keep shopping.
Three hard lines, and a competent vendor builds all three into the system as blocks, not suggestions.
Never give legal advice. "Do I have a case?" "What should I do?" "Is this worth suing over?" — answering any of these is the unauthorized practice of law for a nonlawyer system, full stop. The agent's scripted answer is some version of: that is exactly what the consultation is for, here is the next opening. It can answer firm-approved logistics — consultation fees, office hours, practice areas, what to bring — because that is information about your firm, not advice about their case.
Never promise outcomes. Not "our lawyers win cases like yours," not "you'll definitely get custody," not a settlement range. Beyond the advertising-rule problems, a promised outcome creates expectations that follow the firm into the engagement. The agent sets one expectation only: a real conversation with a real lawyer, at a specific time.
Never create the impression of representation. No "we'll get started on this right away." No engagement language. The script should make the status of the call unmistakable: intake for a potential consultation, conflict check pending, no attorney-client relationship yet. This protects the caller as much as the firm — they should know their matter is not being handled until a lawyer says it is.
Each option wins somewhere. The honest breakdown:
The pattern: a great human receptionist is the best intake experience there is — for 40 hours a week, at $3,500+ a month loaded, with sick days and turnover. A traditional answering service covers the other 128 hours by taking messages you return too late. The AI covers everything, runs your script perfectly every time, and hands anything emotional or complex to a human with the context already captured. Most firms we work with land on a hybrid: AI handles first response and after-hours, humans handle escalations and the consultations themselves. For the deeper three-way cost comparison, see AI receptionist vs answering service vs human.
Three tiers, same as every other industry we serve. (For the market-wide teardown beyond legal, read what an AI voice agent costs in 2026.)
DIY voice AI platforms — $29–$99/mo plus usage. Synthflow, Vapi, Retell and similar tools sell excellent infrastructure and nothing else: you write the intake script, design the conflict-data flow, wire the calendar, and maintain it forever. For a solo practitioner who enjoys the tooling, this beats voicemail by a mile. The catch for law firms specifically: the ethics boundaries above become entirely your engineering problem, and the supervision duty sits with you alone.
Traditional legal answering services — $250–$600/mo, or $1–$2 per minute. Human operators, legal-adjacent scripts, 24/7 coverage. The structural limit has not changed in thirty years: they take messages. They do not run your real intake, they do not capture conflict data reliably, and they book nothing. You pay per minute for the privilege of calling the prospect back after they have already talked to a competitor.
Custom built-and-operated — from $800/mo. This is our model, so read it knowing we are biased. A team studies your intake process, builds the agent around your practice areas and screening rules — with the no-advice, no-promises, no-implied-representation boundaries engineered in — integrates it with Clio, Lawmatics, or your calendar, and operates it: weekly call reviews, script updates when your firm changes, a human accountable for the supervision layer Rule 5.3 expects. Our plans run $800/mo for one agent on one channel, $1,500/mo for two agents, $2,500/mo for three, with custom Agent Workforce scopes for multi-office complexity.
The ROI frame is simple: what is one signed case worth? At a $3,000–$15,000 average case value — higher for PI and family law — recovering one or two after-hours intakes a month covers any tier on this list. We have seen the lead-generation side of this up close in our DUI and family-law case studies: the clicks are among the most expensive in advertising, which makes every unanswered call a very expensive miss. Run your own intake volume through the missed-call calculator and the math usually argues for itself.
30 minutes. We look at your call volume, your practice areas, and your current intake process — then tell you straight which tier fits, even when the answer is a $99 tool and not us.
Three situations where we tell firms to keep their money — and we do, on actual sales calls.
Your intake volume is genuinely low. A solo practice taking a handful of calls a day, with a paralegal who answers by the second ring, does not have a response-time problem. The missed-call math will tell you — if the annual number comes in small, a disciplined call-forwarding setup to your cell covers the gap for free.
Nobody will own the supervision. Rule 5.3 does not care that the nonlawyer is software. If no lawyer at the firm will review calls, approve script changes, and own the escalation path — and you are not paying a vendor to operate that layer with you — do not deploy one. An unsupervised intake agent is how a firm ends up explaining an AI's promise of representation to a bar grievance committee.
Your practice areas demand human-first contact. Some intake conversations — violent crime victims, certain family-law crises, anything where the caller's state is the entire call — deserve a trained human from the first second. AI can still cover after-hours triage and logistics, but if nearly every call you get is that call, the fit is wrong and a good vendor will say so.
And one meta-point: if a vendor pitches you an AI receptionist for your firm without raising Rule 5.3, privilege, and UPL before you do — that tells you everything about how the build will go.
It can be, if it is deployed with the same supervision you would give any nonlawyer assistant. ABA Model Rule 5.3 makes the supervising lawyer responsible for nonlawyer conduct, which bar guidance generally reads as covering vendors and technology. That means scripted boundaries, disclosure that the caller is speaking with AI, no legal advice, and a vendor you have vetted for confidentiality. Confirm the specifics with your state bar — rules and opinions vary by jurisdiction.
No — and a well-built one is hard-blocked from trying. Answering legal questions, evaluating the merits of a case, or predicting outcomes is the unauthorized practice of law for a nonlawyer system. The agent should collect facts, answer firm-approved logistics questions (consultation fees, office hours, practice areas), and book the consultation. Anything that smells like advice gets a scripted hand-off to a lawyer.
Three tiers. DIY voice AI platforms run $29–$99/mo plus usage, and you build and maintain the agent yourself. Traditional legal answering services run $250–$600/mo or $1–$2 per minute to take messages. A custom built-and-operated AI intake agent starts at $800/mo, integrated with Clio, Lawmatics, or your case management system and reviewed weekly by humans.
They should. A well-built agent discloses that it is an AI assistant in the first sentence and offers a path to a human at any time. Disclosure is not just good ethics — it is practical: callers who feel deceived hang up and call the next firm. Most callers stay on the line when the disclosure is upfront and the agent is fast and useful.
That is the highest-value use case. The agent answers in seconds, captures the arrest details, the location, the timeline that matters for deadlines, and the caller's contact information, then hot-transfers to your on-call attorney or triggers the callback protocol you set. A voicemail box does none of that — and the family calling at 2am will dial the next firm on the list.
You should — and get it in writing. Ask any vendor where recordings and transcripts are stored, who can access them, how long they are retained, and whether you can export or delete everything on exit. Intake calls contain some of the most sensitive information a caller will ever share, so treat vendor data handling as a confidentiality decision, not an IT detail.
We build AI voice agents for law firms with the ethics boundaries engineered in — disclosure, no advice, no promises, conflict-data capture, human escalation — and we operate the supervision layer with you every week. Book a 30-minute call and we will tell you honestly whether it fits your practice.