An AI receptionist for law firms is software that answers your phone, asks your intake questions, books consultations, and passes urgent calls to a person. It does not give legal advice, decide whether you have a conflict of interest, or replace the human judgment a frightened caller sometimes needs. Inside those limits it solves a real problem, because most law firms simply don’t pick up the phone.
Clio’s 2024 secret-shopper study of 500 firms found that only 40% answered a prospective client’s call, down from 56% in 2019. That is the gap AI reception is built to close. This guide covers what it handles well, where it fails, what the ethics rules say, what it costs, and how to test one without putting your firm at risk.
Not sure how many calls your firm is missing? Get a free intake and local search audit from Oxxleads. We look at how your firm handles calls and web inquiries and where you show up in local search. Book your free audit · Message us on WhatsApp
The short version: what it can and can’t do
| An AI receptionist can | An AI receptionist can’t (and shouldn’t try to) |
|---|---|
| Answer every call, at any hour, including several at once | Tell a caller whether they have a case |
| Ask the same intake questions in the same order every time | Accept or decline a matter, or clear a conflict |
| Book consultations straight onto your calendar | Discuss fees or likely outcomes beyond wording you have approved |
| Route urgent calls to the right person by rules you set | Reliably spot every real emergency on its own |
| Log notes, summaries, and transcripts to your CRM or case system | Guarantee confidentiality without the right vendor terms and setup |
| Answer approved FAQs (hours, locations, practice areas, what to bring) | Comfort an upset caller the way an experienced person can |
| Capture details the receptionist would otherwise miss after hours | Discuss an existing client’s case with whoever happens to call |
Why law firms are looking at AI reception in the first place
Phone coverage is the weak point of most small and mid-sized firms. Clio hired an independent research company in 2024 to contact 500 law firms by phone and email as if they were prospective clients. The results, as reported in the ABA’s Law Practice magazine and the Oklahoma Bar Journal, were:
- 40% of firms answered the phone (56% in 2019).
- Of the firms that missed the call, only about 20% called back.
- Taken together, 52% answered or returned the call, which means 48% were effectively unreachable by phone.
- Only 33% responded to an email inquiry.
Two cautions. This was a mystery-shop test, so it doesn’t measure your firm’s after-hours performance specifically. And it doesn’t prove that every missed call was a lost client. But the direction is hard to argue with. A person with a legal problem usually works down a list of firms, and the first one that answers has a head start on the consultation.
What happens on a call
An AI receptionist for law firms usually follows the same basic flow:
- It answers and discloses. “You’ve reached the AI assistant for [Firm]. This call may be recorded. How can I help?”
- It listens to why the person is calling, in their own words.
- It collects a short set of details: name, callback number, type of matter, location or jurisdiction, urgency, and the names of other parties if you want them for conflict checks.
- It checks for emergency triggers (more on those below) and transfers live if one fires.
- It books a consultation or logs a callback request and texts or emails a confirmation.
- It sends your team a summary and transcript so nobody has to re-ask the basics.
Notice what’s missing: at no point does it assess the case. That boundary is the whole design.
What it does well
After-hours and overflow coverage. Evenings, weekends, lunch breaks, and the moments when your receptionist is already on another line are the calls most likely to hit voicemail. This is where AI reception pays off first, and it is why starting with after-hours and overflow is the safest rollout.
Consistent intake. A human receptionist’s intake varies by mood, workload, and training. A configured script asks the same questions every time, so attorneys review complete records instead of “John called, wants a callback.” It also gives you real data on which practice areas the phone actually produces.
Instant booking. If the AI can see your calendar, a caller can book a consultation during the call instead of waiting for a callback the next afternoon. Fewer touchpoints means fewer places to lose them.
Multiple languages. Many vendors advertise 20 or more languages. That can matter a great deal for immigration, workers’ comp, and PI practices. Treat it as a claim to test with real accents, not a feature to take on faith.
Clean records. Transcripts and summaries create a searchable record of every inquiry, including the ones that never became clients.
What it can’t do (the part most vendors skip)
1. It can’t give legal advice
Callers will ask: “Do I have a case?” “How much is it worth?” “Should I sign this?” A badly configured system may try to be helpful. That is how a receptionist tool ends up on the wrong side of the unauthorized-practice rules. ABA Formal Opinion 506 (2023) allows lawyers to delegate intake tasks to nonlawyers under supervision, with the condition that those tasks don’t become the practice of law. The same logic is the safest way to think about AI intake.
Build the refusal into the script, and write it in your own voice. For example: “I can’t say whether that’s a claim. Only an attorney can. I can take your details so someone can call you tomorrow morning. Is this the best number?”
2. It can’t clear conflicts or accept the matter
It can collect the names of opposing parties. It cannot decide whether you have a conflict or whether the firm will take the case. Run everything it captures through your conflicts process before an attorney discusses details with the caller.
Keep the intake short. ABA Model Rule 1.18, which covers duties to prospective clients, protects what callers tell you even if they never retain you, and its commentary tells lawyers to limit an initial consultation to the information reasonably needed to decide whether to take the matter. Information that is “significantly harmful” to the caller can disqualify your firm from representing the other side. A long, chatty AI intake that collects the whole story creates that risk. A six-question intake does not.
3. It can’t be trusted to recognize every emergency
Some callers are in crisis: someone reporting domestic violence, a family member who has just been arrested, a person who has been hurt and is still at the scene. Don’t rely on the AI to “sense” urgency. Hard-code the triggers (words, phrases, practice-specific situations), tell callers in immediate danger to hang up and call 911, and have the system transfer to a live person when a trigger fires. Then test it with actual call recordings or role-played calls, not just the happy path.
4. It can’t replace empathy on the hardest calls
Family law, criminal defense, and serious injury calls often come from people on the worst day of their year. Some will want a human, and some will hang up on a bot. Give every caller an obvious way to reach a person, and consider a hybrid setup where staff answer during business hours and the AI covers the rest.
5. It can’t promise confidentiality by itself
Confidentiality depends on your vendor’s terms and your configuration. ABA Formal Opinion 512 (July 29, 2024) says lawyers must consider confidentiality before putting client information into generative AI tools, and that informed consent is required for self-learning tools that take in information relating to a client’s matter. Boilerplate in an engagement letter doesn’t count as informed consent. The opinion’s confidentiality discussion also reaches prospective clients through Rule 1.18(b), which is exactly who calls your intake line.
In practice, that means you need written answers from any vendor about where audio and transcripts are stored, how long they are kept, who can access them, and whether your data is used to train models.
6. It can’t fix a weak follow-up process
If an AI books fifteen consultations and nobody confirms them, sends reminders, or calls back the caller who didn’t book, you’ve paid for faster failure. The tool feeds the pipeline, and your team still has to run it.
7. It will sometimes mishear
Names, addresses, accents, bad cell connections, and background noise cause errors. Have it read back phone numbers and spell names, and spot-check transcripts for the first month.
Compliance checklist (check your state bar’s rules)
This is general information, not legal advice. Rules vary by state, and some state bars have issued their own guidance on AI and chatbots. Ask your bar or ethics counsel for your jurisdiction.
| Issue | What to do | Why it matters |
|---|---|---|
| AI and recording disclosure | Say at the start of every call that it is an AI assistant and that the call may be recorded | Roughly a dozen states require all-party consent to record (California, Florida, Illinois, Massachusetts, Pennsylvania, and Washington are commonly listed). Callers cross state lines, so disclosing on every call is the simple safe choice. AI-disclosure rules are also spreading |
| No attorney-client relationship yet | Have the AI say it isn’t a lawyer and that no representation exists until the firm confirms one | Avoids caller confusion and implied-representation arguments. An Oregon State Bar opinion on website chatbots recommends this kind of disclaimer |
| Limit what’s collected | Names, contact details, matter type, jurisdiction, deadlines, urgency | Rule 1.18 and ABA Formal Opinion 510 (2024) |
| Conflicts | Send every captured name through your conflicts system before the attorney engages | Rule 1.18 |
| Vendor vetting | Get data retention, security, subprocessors, and training-data terms in writing | ABA Formal Opinion 512 |
| Supervision | Assign a lawyer or manager to review transcripts and scripts regularly | Model Rule 5.3 on supervising nonlawyer assistance, and Opinion 512’s supervision section |
| Advertising limits | Approve all scripted statements. No “we win” claims or guarantees | Model Rules 7.1 and 7.3 |
| Existing clients | Route them to staff. Don’t let the AI discuss case details with an unverified caller | Rule 1.6 |
Want a second set of eyes on your intake process? Our free intake and local search audit shows where calls and inquiries drop out before they reach an attorney. Book it here.
AI vs. answering service vs. in-house: how the options compare
| Option | Typical entry pricing (2026) | Strengths | Weak spots |
|---|---|---|---|
| In-house receptionist | Salary plus benefits and overhead (varies by market) | Judgment, empathy, knows your clients | One line, set hours, sick days and turnover |
| Live human answering service | Ruby plans start around $235–$250/month for 50 minutes. Smith.ai’s human plans start around $292–$300/month for 30 calls | Human warmth, after-hours coverage | Costs climb with call volume. Staff know your firm less well |
| AI-only receptionist | Smith.ai’s AI product starts around $95/month. Other AI-first tools advertise roughly $30–$200/month | Lowest cost, 24/7, handles several calls at once | No real judgment, edge cases, needs careful setup |
| Hybrid (AI first, human handoff) | AI base price plus a handoff fee (Smith.ai lists $3 per handed-off call) | AI handles the routine, people handle the hard calls | More moving parts to configure |
Prices come from vendor and third-party comparison pages in 2026 and change often. Confirm on each vendor’s own site before you decide.
What it costs vs. what it’s worth
Don’t compare the monthly fee to a receptionist’s salary alone. Compare it to the cases you lose.
The formula: Missed new-client calls per month × share that never call back × consult rate × signing rate × average fee
An illustrative example (not a benchmark. Plug in your own numbers):
| Assumption | Value |
|---|---|
| New-client calls per month | 60 |
| Share missed (voicemail, after-hours, busy) | 40% → 24 calls |
| Share of missed callers who never try again | 50% → 12 calls |
| Calls that become consultations | 30% |
| Consultations that sign | 25% |
| Average fee per case | $6,000 |
12 calls × 30% × 25% × $6,000 = about $5,400 per month in fees you wouldn’t otherwise have captured.
Change any of those inputs and the answer moves a lot, which is the point. Before you buy anything, pull 30 days of call logs from your phone provider and count the real numbers: calls received, calls missed, and calls after hours. Then you’re deciding on your data, not a vendor’s case study.
How to set one up so it helps instead of hurting
- Start with after-hours and overflow. Leave the daytime front desk alone until you trust the system.
- Write the intake script yourself. Keep it to six to eight questions per practice area. Don’t let a vendor’s generic template decide what your firm asks.
- Hard-code emergency triggers and test each one.
- Set a human response standard. Every AI-captured lead gets a personal call or text within a fixed window, such as one hour during business hours.
- Always offer a human. A caller should be able to ask for a person at any point and get one, or get a firm callback time.
- Test with 20 realistic calls before going live: heavy accents, angry callers, rambling callers, a child calling, a wrong number, someone asking for legal advice, someone describing an emergency.
- Review transcripts weekly for the first month, then monthly.
Questions to ask any vendor before you sign
- Where are call audio and transcripts stored, for how long, and who can access them?
- Is our data used to train or improve your models? Can you confirm that in writing?
- Which third parties (speech-to-text, language model, telephony) touch our calls?
- Can we set the exact opening disclosure and recording notice?
- What happens on a transfer if no one answers? Is it a warm transfer?
- What does the AI do when it doesn’t know the answer? Can we restrict it to approved answers only?
- How does it connect to our case management or CRM, natively or through a third-party tool?
- Can we export and delete recordings and transcripts on request?
- How are overages billed (per call, per minute), and is there a month-to-month option?
- Can you put us in touch with another law firm in our practice area?
A 30-day pilot, with numbers you can check
| Metric | How to measure it |
|---|---|
| Baseline missed and after-hours calls | Phone provider logs from the 30 days before launch |
| Calls answered | Call log, AI plus human |
| Calls abandoned in the first 30 seconds | Call log |
| Qualified consultations booked | Calendar or CRM |
| Consultations that show up | Calendar or CRM |
| Time from AI intake to a human follow-up | Timestamps in your CRM |
| Transcripts with errors or off-script answers | Weekly sample review |
| Caller complaints or “I want a person” requests | Transcript tags |
If consultations go up and the transcript review is clean, widen the rollout. If the error count is high or callers keep asking for a person, fix the script or change vendors before you expand.
Frequently asked questions
Can an AI receptionist give legal advice?
No, and it shouldn’t be allowed to. It can take down a caller’s situation and route them to an attorney, but any answer that interprets the law or judges a case belongs to a lawyer. Build that limit into the script and test it.
Is it ethical for a law firm to use an AI receptionist?
Generally, firms can use these tools, but the ethics duties still apply: competence, confidentiality, supervision of nonlawyer help, and duties to prospective clients. ABA Formal Opinions 506, 510, and 512 are the best starting points. Your state bar’s guidance controls, so check it.
Do I have to tell callers they’re talking to an AI?
Treat disclosure as mandatory. Some states have bot-disclosure rules, and a plain statement at the start of the call protects the firm from complaints that a caller was misled. Combine it with the recording notice.
Do I need consent to record calls?
It depends on the state. Roughly 38 states plus DC are one-party consent, and about 12 are all-party. Because callers can be anywhere, many firms announce recording on every call.
How much does an AI receptionist for a law firm cost?
Published entry prices range from roughly $30 to $200 a month for AI-first tools, with hybrid and human-staffed services starting in the low hundreds per month and rising with volume. See the comparison table above and confirm current pricing with each vendor.
Can it replace my receptionist?
For after-hours and overflow, often yes. As a full replacement, usually not. Most firms do better with a hybrid setup that keeps a person for complex, emotional, and existing-client calls.
How long does setup take?
Vendors often advertise same-day setup, and the technical part can be that quick. Writing a good intake script, setting emergency rules, and testing realistically takes closer to one to two weeks.
Which practice areas get the most out of it?
Any practice that depends on inbound calls benefits, including personal injury, family law, immigration, criminal defense, employment, and estate planning. High-stakes areas such as criminal and family law need the most careful emergency and escalation rules.
The bottom line
An AI receptionist for law firms is a good answer to one narrow problem: calls that go unanswered. Used for after-hours coverage and overflow, with a short script, strict limits, and a human following up, it can recover consultations you’re losing right now. Used as a stand-in for a lawyer or a fully unsupervised front desk, it creates ethics and reputation risks that no monthly fee covers.
The best first step costs nothing: count your missed calls.
Want help finding where your firm loses inquiries? Claim your free intake and local search audit and we’ll show you where calls and web leads fall through, and how your firm appears when local clients search. Book your free audit · Message us on WhatsApp
This article is general information, not legal advice. Ethics rules and recording laws differ by state, so confirm requirements with your state bar or ethics counsel before deploying any AI intake tool.
Sources
- Clio 2024 Legal Trends Report secret-shopper findings, as reported in Law Practice magazine (ABA) and the Oklahoma Bar Journal, September 2025
- ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024), summarized by The Bar Examiner and the Florida Bar News
- ABA Formal Opinion 506 on nonlawyer intake and ABA Formal Opinion 510 (2024) on Rule 1.18, discussed in Washington State Bar News and Minnesota Lawyer
- Model Rule 1.18 and remote intake: Attorney at Work
- Oregon State Bar guidance on chatbots and AI agents for intake (summary): Ezel
- Call-recording consent: Dialzara and NextPhone
- Pricing references: Dialzara comparison of Ruby and Smith.ai, Smith.ai review (Fast.io), NextPhone Smith.ai alternatives
