Article
Jun 9, 2026
AI Receptionist for Law Firms: Scripts, Guardrails, and Real Costs
Only 40% of firms answer the phone. Here's what an AI receptionist can actually handle on intake, where it must stop, and what it really costs.
TL;DR: the honest version
Clio mystery-shopped 500 firms for its 2024 Legal Trends Report secret-shopper study (fielded by Lux, June-July 2024) and got a human voice 40% of the time, down from 56% in 2019. About 1 in 5 missed calls got a callback.
A well-scoped AI receptionist for law firms runs $50–$400/month at typical solo-to-mid volume. Human virtual receptionists list from about $245/month (Ruby) and $292.50/month (Smith.ai) at entry tiers.
The job is answer, screen, qualify, schedule, route. It is not giving legal advice, taking confidential matter details, or confirming representation.
Three guardrails matter more than any feature list: never advise, never confirm a conflict-free engagement, always hand off anything that smells like an emergency or an existing client in distress.
Pilot for two weeks against a written pass/fail sheet. If it can't beat your current answer rate with zero guardrail breaches, kill it.
Should your firm use an AI receptionist, and what does it cost?
Here's the rule we give firms: under ~10 inbound calls a week, skip it. Above that line, if you're missing more than two calls a week and have no dedicated intake specialist, pilot it. For most solo-to-mid firms the direct answer is yes, at $50–$400/month (full math in section 5). The rest of this piece is about whether your firm is the exception, and how to deploy without taking on ethics risk you can't defend at a bar hearing.
The intake problem is worse than the industry admits. Clio's 2024 Legal Trends Report secret-shopper study (500 firms, fielded by Lux, June-July 2024) found only 40% answered the phone, and roughly 20% of missed calls were returned at all. The 411 Locals study across 58 industries found about 62% of inbound calls to small businesses go unanswered. Law firms barely clear that dismal baseline — a 40% answer rate when the average small business manages 38%. Meanwhile HBR's classic study on lead response showed that answering within an hour makes qualification nearly 7x more likely than answering after.
The operator translation: every unanswered call is a retainer dialing the next firm. We covered the response-time math in this piece on lead-response benchmarks. The curve is the same in legal.
What an AI receptionist actually handles on a legal intake call
A properly built AI receptionist is a scripted intake worker with a voice, not a lawyer. It does five things, in this order:
Answer within two rings, in your firm's name, with a disclosed-as-virtual greeting.
Screen for caller type: existing client, new matter, opposing counsel, solicitor, emergency.
Qualify new matters against a short script you wrote: practice area, jurisdiction, basic facts the client volunteers, urgency.
Schedule a consult on the calendar that your intake attorney actually owns, or route to a human.
Route to voicemail, on-call attorney, or live transfer, based on rules you set in advance.
What it does not do: take detailed matter facts (that's a privilege problem we'll get to), give legal opinions, confirm whether the firm will take the case, or quote fees. Missed-call text is a different fix, and we wrote about it here for contractors. For a law firm, text-back is the bare minimum.
Intake routing. Every attorney-level worry routes through one gate: the Human Handoff Gate.
Notice where the orange node sits in that flow. The Human Handoff Gate is the trust node. Everything an attorney worries about (emergency calls, sensitive matters, existing clients in distress) collapses into that one decision. Build it carefully or don't build it at all.
The lines it must never cross
No vendor page writes this section. Three guardrails, in order of how badly you get hurt if you skip them.
Guardrail one: no legal advice, ever. The AI has a hard-coded refusal list. "Do I have a case?" gets "That's something the attorney will evaluate during your consult — I can get you on the calendar this week," rather than "Based on what you've described, it sounds like…" Every script we ship has a list of trigger phrases ("do I have," "am I entitled to," "what should I do") that route to a canned redirect, not a generated answer.
Guardrail two: limited intake on first contact. Confidentiality attaches to prospective-client communications in most jurisdictions (ABA Model Rule 1.18), which is exactly why you don't want a stranger pouring matter details into a voice agent that may be probed for them later. The AI captures name, callback number, practice area, jurisdiction, urgency, conflict-check basics (other party's name, if volunteered). It does not solicit narrative facts. "Tell me what happened" is not in the script. "An attorney will get the details during your consult" is.
Guardrail three: the AI does not run the conflict check. It captures the names. The conflict check is a human action against your conflicts database before the consult is confirmed. If your PMS supports automated conflict screening, the AI can trigger the run, but the clearance is a human decision. Same logic as not letting an agent send wires without an approver above a dollar threshold: the machine raises the flag, a human clears it.
If a vendor tells you their AI is "fully compliant" with bar ethics rules, that's a sales tell. There is no certification. There are guardrails you can describe and audit. Anyone who oversells the first is hiding the absence of the second.
Intake scripts by practice area
The single biggest mistake firms make is deploying one script for every practice. PI intake is not estate intake. Here are the openings and required fields we use as starting points.
Personal injury. Opening: "Thanks for calling [Firm]. I'm a virtual assistant — I can get you scheduled with an attorney today. First, are you safe and is anyone still injured?" Required capture: incident date, jurisdiction, whether a police report exists, whether the caller has spoken to insurance, other party's name (for conflicts). Hard escalation: ongoing medical emergency → 911 reminder + human handoff.
Family law. Opening: "Thanks for calling [Firm]. I'm a virtual assistant. Before we go further — are you in a safe place to talk?" Required capture: matter type (divorce, custody, support), jurisdiction, opposing party's name, whether any orders are already in place, urgency window. Hard escalation: any mention of immediate danger → human handoff + safety resource line.
Estate planning. Opening: "Thanks for calling [Firm]. I'm a virtual assistant — happy to get you on the calendar with one of our estate attorneys." Required capture: estate-planning vs probate vs trust admin, state of residence, rough timeline. This is the lowest-urgency practice and the easiest to fully automate to a scheduled consult.
Criminal defense. Opening: "Thanks for calling [Firm]. I'm a virtual assistant. Is the person needing the attorney currently in custody?" Required capture: in custody yes/no, jurisdiction, charge type if known, court date if known, caller's relationship to defendant. Hard escalation: in-custody calls → on-call attorney, not voicemail, regardless of hour.
Each script does three jobs: a safety check in the first ten seconds, a short structured capture, and a named escalation rule. Those three jobs are the contract between you and the system.
What an AI receptionist for law firms actually costs
Vendors quote per-minute, per-call, and per-month interchangeably, which is why the ranges floating around look confusing. The honest math:
AI voice agent infrastructure runs $0.05–$1.00 per minute retail, with managed platforms in the $0.25–$0.50/minute range per Aircall's breakdown.
A solo-to-mid firm taking 50–200 calls a month at 2–4 minutes each lands between $50 and $400 at managed rates; most firms we scope land near $150.
Human virtual receptionist services price by minute or call bucket. Ruby lists entry plans from about $245/month for 50 receptionist minutes; Smith.ai starts from about $292.50/month for 30 calls. Both climb quickly with real volume, so re-check the pricing pages before you sign.
An in-house intake hire runs roughly $3,000+/month loaded at a junior rate (our estimate; varies by market), and they don't work nights or weekends.
Where the wedge is: ethics risk controls are configurable and auditable on AI, hoped-for on humans.
The wedge is the orange row. On an AI system, ethics risk controls are settings you configure and audit. With a human virtual receptionist, you're trusting a training manual and hoping the training stuck. Both can be done well. Only one gives you a deterministic audit trail of what was said and what wasn't. For a fuller per-tier teardown see our AI receptionist cost piece.
If you want the pass/fail sheet we use for these pilots, ask us for it. It's a one-pager.
Integration with your practice management stack
None of this matters if intake data dies in a voice transcript. The integrations worth checking:
Clio Grow / Clio Manage. New-lead push with structured fields, calendar booking against attorney availability, conflict-check trigger. Most AI receptionist vendors ship a native Clio connector; confirm it writes leads into Grow rather than dumping raw contacts into Manage.
MyCase / Smokeball. Contact creation, matter stub, task assignment to intake attorney. Native connectors are rarer here, so expect a Zapier or webhook middle layer and treat it as one more failure point to monitor.
Lawmatics. Lead nurture sequence kickoff after the AI captures the consult booking. Lawmatics exposes an open API; the practical question is whether your vendor triggers the sequence natively or needs a custom webhook.
The pattern in every case: the AI writes a structured record rather than a transcript blob, tags it as unverified intake, and assigns a human to clear conflicts before the consult is confirmed. The structured record is the deliverable. If all you get back is a voicemail with a callback number, you've paid for something you already had.
When NOT to use an AI receptionist
Three cases where we tell firms to skip it:
White-glove clientele. If your average matter is $50k+ and clients expect a named human by the second ring, an AI greeting is a brand mismatch. Hire a person.
Under ~10 inbound calls a week. Same skip line as section 1. The setup cost (a few hours encoding scripts) outweighs the answer-rate lift at that volume. Use missed-call text-back and move on.
No written intake process. If you can't hand a new paralegal a one-page script and have them run intake competently, you can't hand an AI one either. Fix the process first.
A two-week pilot plan
Two weeks is enough to know. Forward your main line to the AI after-hours and measure four things against a written pass/fail sheet:
Answer rate. AI must beat your current human answer rate by at least 20 percentage points during covered hours.
Qualified-intake rate. Of calls answered, what share produced a structured record with all required fields? Target: 70%+ for new matters.
Transfer accuracy. Of calls routed to a human, how often was the routing correct? Target: 95%+.
Guardrail breaches. Any instance of the AI giving advice, soliciting narrative facts, or confirming engagement. Target: zero. Non-zero kills the pilot.
We walk through how to budget worked-example pilots in this piece on AI agent cost. Same logic applies here: scope small, measure honestly, kill what doesn't clear the bar.
FAQ
How much does an AI receptionist for law firms cost?
For a typical solo-to-mid firm taking 50–200 calls per month, expect $50–$400 per month all-in, based on managed per-minute pricing of $0.25–$0.50 — most firms we scope land near $150. Human services list from about $245/month (Ruby) or $292.50/month (Smith.ai) and climb with usage. In-house intake staff runs roughly $3,000+/month loaded, by our estimate.
Can an AI receptionist give legal advice?
No, and it should be configured to refuse. A properly built system has a hard-coded refusal list: phrases like "do I have a case" or "am I entitled to" route to a canned redirect that books a consult. If a vendor demos an AI that opines on case strength, that's the system you don't deploy.
Is AI phone intake confidential enough for a law firm?
It can be, if scoped narrowly. The AI should capture name, callback, practice area, jurisdiction, urgency, and conflict-check basics, never narrative facts. Confidentiality can attach to prospective-client communications under ABA Model Rule 1.18, so the script defers matter details to the attorney consult. Vendor data handling and retention terms matter as much as the script.
Can an AI receptionist run a conflict check?
It can capture the names needed and trigger the search in your PMS. It should not clear the conflict. Clearance is a human decision against your conflicts database before the consult is confirmed. Treat the AI like a junior intake clerk: it raises the flag, an attorney signs off.
Will potential clients hang up on an AI receptionist?
Some will, especially in family law and criminal defense where callers are in distress. The mitigation is a fast, named human-handoff path within the first 15 seconds for anyone who asks. Measure your own hang-up rate during the two-week pilot rather than trusting anyone's benchmark.
Run the pilot
Forward your line to the AI after hours this week. Run the two-week pilot against the pass/fail sheet. If it can't beat your answer rate with zero guardrail breaches, turn it off. You're out about $50. If it can, you've just bought back every weekend call you were losing. Scoping the scripts, the guardrails, and the pass/fail sheet is the part we build, so talk to us when you're ready.