Key takeaways
- Judge products on the hard legal calls: the opposing party, a caller who already has a lawyer, an arrest at night, a fee question.
- Conflict names should come before the story; ABA Formal Opinion 506 lets trained nonlawyers collect them while a lawyer makes the final call[2].
- Disclose the automation in the greeting and add a recording notice; about 11 states primarily require every party's consent to record[3].
- Score vendors on a weighted scorecard, then run fictional test callers through each one before any real call is forwarded.
Buy for the hard calls
Most AI receptionist demos are built around a salon or a dental office: a friendly voice, an open slot, a confirmation. A law firm's line is harder. The caller may be the opposing party. The person asking about a consultation may already have a lawyer. A mother may ring at 11 p.m. because her son was arrested, and what she tells whoever answers is information the firm must protect even if it never takes the case[4]. Buy for those calls and the easy ones take care of themselves.
The pressure to buy something is real. In Clio's 2024 secret-shopper study of 500 US firms, only 40% answered the phone, down from 56% in 2019[1]. Small offices feel the gap most and have moved slowest: fewer than 20% of US firms with four or fewer employees reported using AI in spring 2026, against 37% of firms with at least 250 employees[5]. This guide is for the firm making that first purchase: what the product has to do, what it must never do, how to score vendors side by side and how to test before a real caller hears it.
Six jobs it has to do on a legal call
Strip away the voice quality and the dashboard and a legal receptionist has six jobs. Score every product on these before you look at anything else.
| Job | What good looks like | How to test it |
|---|---|---|
| Answer on your schedule | Picks up every forwarded call in the hours you choose: all day, after hours or overflow | Forward a test line for one evening and call it from three phones at once |
| Follow your intake | Asks the questions your firm approved for each practice area, in your order, and stops there | Give it a family law caller and a DUI caller and compare both against your written script |
| Capture conflict names first | Takes the caller's name, the other parties and any company involved before the story | Have a tester open with a long story and see whether the names still come first |
| Route urgent calls | Follows a written escalation tree with backups and a fallback when nobody answers | Trigger the urgent rule while the on-call phone is switched off |
| Book or request the consultation | Books into one connected calendar, or takes a request your staff confirm | Ask for a slot that is already taken and one outside consultation hours |
| Hand off a usable record | Summary, recording and transcript for each call, with urgent calls marked | Read five summaries against their recordings and count missing names and dates |
The conflict row is where general-purpose products most often fall short, because their default script asks how it can help and lets the caller talk. ABA Formal Opinion 506 accepts that trained nonlawyers can collect the names of the parties and the incident dates and run a conflict check, as long as a lawyer reviews any name that comes up and makes the final determination[2]. A receptionist that hears the whole story before it hears the other side's name has already taken in information the firm may have to wall off. The right order is laid out in what to collect first for a conflict check.
Urgent routing needs the same precision. “Put emergencies through to the attorney” is not a rule. Here is an example of one: the caller says a client or family member is in custody now, or has a hearing within two days; ring the on-call attorney, then the backup, then the managing partner; if nobody answers, save a detailed urgent message and tell the caller plainly what happens next. Products differ widely on how much of that they can follow. Escalation rules for after-hours calls has templates you can adapt.
Six things it must never do
This list matters more than the first one. A missed call costs a lead; a wrong answer on a legal call can cost far more, and it is on a recording.
- Give legal advice. “You probably have a case” or “that sounds like a misdemeanor” is advice, whoever or whatever says it.
- Negotiate fees or tell callers what services they need. Opinion 506 reserves both, along with questions about the engagement agreement, for the lawyer[2].
- Promise an attorney. “Someone will call you tonight” is a commitment. “I have marked this urgent and the on-call attorney is being called” is a fact.
- Pass itself off as a person. Asked directly, it should say what it is and offer a callback from a member of staff.
- Keep interviewing someone who already has a lawyer for the matter. A screening question belongs near the top of the call[6].
- Take a statement from the opposing party. Once the other side's name matches a client, the call becomes a short message, not an intake.
Ask each vendor to show, in a recorded test call, how its product handles three of these. Marketing pages describe the easy path. A test call shows the edges.
Disclosure, recording and supervision
Callers want to know what they are talking to. Half of US adults say they are more concerned than excited about the growing use of AI in daily life, and only 10% are more excited than concerned[7]. That is not a reason to avoid automation. It is a reason to say what it is in the first sentence and to make a person easy to reach.
Source: Pew Research Center, 2025[7]
Show the data
| Response | Share of US adults |
|---|---|
| More concerned than excited | 50% |
| Equally concerned and excited | 38% |
| More excited than concerned | 10% |
The rules point the same way. Florida's bar says AI chatbots that communicate with clients or third parties must disclose that they are an AI program and not a lawyer or firm employee[6]. Utah requires a business using generative AI with consumers to disclose it when the person clearly asks[8]. The FCC has proposed, but not adopted, a rule that callers using AI-generated voices say so at the start of each call[9]. It concerns calls a business places rather than calls it receives, but it shows where expectations are heading. Greeting wording that works is in telling callers they are speaking with AI.
Recording is the other notice. Federal law sets one-party consent as the minimum, but about 11 states primarily require every party's consent[3], and a firm in one state takes calls from all of them. The safe default is a short recording notice in the greeting. Recording consent rules by state covers the details.
Supervision is the part buyers skip. California's State Bar lists autonomously facilitating client intake as an example of agentic AI and says a lawyer's diligence and supervision must correspond to the system's level of access and autonomy[10]. In practice someone at the firm reads summaries against recordings every week for the first month, and the product should make that easy: searchable transcripts, live listening and a way to step into a call. Checking AI call summaries for accuracy has a sampling routine.
Message-taker or full intake
Products sold as AI receptionists range from a voicemail replacement that writes down a name and number to something that runs the firm's own intake. The gap shows up on the hard calls, and the comparison below is the quickest way to find out which kind you are being shown.
| Message-taking receptionist | SI Employee | |
|---|---|---|
| First questions | Name, number and reason for calling | Your approved questions for each practice area |
| Conflict data | The caller's name only | Every party's name, asked before the story |
| Urgent calls | A message, or one transfer | An escalation tree of up to three people plus a fallback |
| Consultations | A callback request | Booked on one tested calendar, or a request staff confirm |
| Record of the call | A short message | Summary, recording and transcript, urgent calls marked |
| Live control | None | Listen in, whisper guidance or take over |
Capabilities vary by product. Treat the left column as the floor to test any vendor against.
The right-hand column is how an SI Employee from AssistCall is set up for a law firm. The firm forwards its number for the hours it picks, approves a greeting that carries an automated-assistant disclosure and a recording notice, signs off the questions one practice area at a time and sets the escalation tree. When a call goes through to an attorney, the attorney hears a short briefing before the caller is connected. Some things are not live yet and should not be sold to you as if they were: syncing with case management systems such as Clio or MyCase, texting callers and chat on your own website are all planned. See what it does for the current list.
Score vendors the same way
Put every product through the same scorecard and weight the rows that carry legal risk above the ones that look good in a demo. A product that fails a high-weight row is out, whatever it scores on voice quality.
| Criterion | Weight | What to ask the vendor for |
|---|---|---|
| Intake follows your script | High | Your questions loaded for one practice area and tested on a recorded call |
| Conflict names before facts | High | A test caller who launches straight into the story: does it redirect? |
| Escalation and no-answer path | High | A recording of what the caller hears when nobody picks up |
| Disclosure and recording notice | High | The exact greeting text, and who approves any change to it |
| Records and retention | Medium | Where recordings and transcripts are kept, who can open them and for how long |
| Booking | Medium | Which calendar it writes to and what happens when no slot fits |
| Supervision tools | Medium | Live listening, takeover and search across transcripts |
| Voice and pacing | Low | Two recorded calls with real background noise and a nervous caller |
Test with fictional callers first
A live trial on real prospects is the wrong place to learn how a product treats someone who already has a lawyer. Write test callers, run each one twice and score the recordings. A fuller rollout plan is in piloting AI phone answering at a law firm.
Write eight callers
A new matter in each practice area you take, an existing client, the opposing party, someone who already has a lawyer, a fee question, an arrest after hours and a caller who asks whether it is a real person.
Run each one twice
Use different testers and different phrasing. Consistency matters as much as the first result.
Score from the evidence
Mark the scorecard from the recording and the summary, not from memory of the call.
Break the transfer path
Switch off the on-call phone and confirm the caller is told plainly and an urgent message is saved.
Read the summaries cold
Give five summaries to an attorney who did not hear the calls. Could they act on each one without listening to it?
- SI Employee
- Caller
- System
- Your team
01SI Employee
Greeting and disclosure
Firm name, automated-assistant notice and recording notice.
02SI Employee
Conflict names first
Caller, other parties and any business involved, before the story.
03SI Employee
Already represented?
If yes, contact details only and a message for the firm.
04Caller
Matter and key dates
Practice area, incident or hearing dates, county and how they found the firm.
05System
Urgent rule checked
A custody or near-hearing trigger goes down the on-call tree.
06SI Employee
Consultation set
Booked on the tested calendar, or a request staff confirm by morning.
07Your team
Summary to intake lead
Summary, recording and transcript arrive for review.
Example sequence. The order and wording come from the firm's approved intake for that practice area.
What after-hours coverage could be worth
Price comparisons make more sense once you know what the uncovered hours are worth. About 15% of US wage and salary workers usually work evenings, nights or other non-daytime schedules[11], and plenty of others will only make a private call about a legal problem once they are home. The model below values after-hours calls from new prospects. Change the inputs to your own numbers.
| Input | Example |
|---|---|
| After-hours calls per week | 12 |
| Share from new prospects | 50% |
| Share that would sign | 25% |
| Average fee per new matter | $3,000 |
| After-hours calls a month | 52 |
| From new customers | 26 |
| Bookings if answered | 6.5 |
| Monthly after-hours value | $19,485 |
How it is worked out: After-hours calls a week × 4.33 weeks × new-customer share × booking rate × average value.
Assumptions
- About 15% of wage and salary workers usually work a non-daytime schedule, which pushes some calls past office hours.[11]
- Clio's 2024 secret shoppers got a person on the phone at only 40% of firms, so many firms miss these calls today.[1]
- The sign rate and fee are placeholders. Replace them with your own intake and billing figures.
Example only. Treat the result as a ceiling on what better coverage could be worth, not a forecast.
Set any product's annual cost against that ceiling, then add the staff time it saves and the supervision time it needs. The fuller decision method is in whether an AI receptionist is worth it. For a worked scenario, the solo injury lawyer example models an attorney with one paralegal covering evening and weekend ad calls.
Contract and data questions
Intake calls carry confidential information from the first minute. California's guidance warns lawyers not to put confidential client information into a generative AI tool that may present material risks to confidentiality or security without the client's informed consent[10]. Get written answers to these before signing anything.
- Where are recordings and transcripts stored, and can we set how long they are kept?
- Who at the vendor can listen to our calls, and under what circumstances?
- Is caller information used to improve the vendor's systems, and can we switch that off?
- Can we export every call record, summary and contact if we leave?
- Does our number stay ours? With call forwarding it does; with a vendor-issued number, read the porting terms.
- How are calls or minutes counted, and what happens in a month when volume doubles?
- Who approves changes to the greeting and the questions, and is there a record of each change?
Two comparisons are worth reading before the shortlist is final: an AI receptionist against a human one, and how a basic AI receptionist differs from a fuller intake setup.
Frequently asked questions
Can an AI receptionist run a conflict check?
It can collect what the check needs: the caller's name, the other parties, any business involved and the key dates. ABA Formal Opinion 506 accepts trained nonlawyers gathering names and running checks, but a lawyer reviews any name that comes up and makes the final determination[2].
Do we have to tell callers they are talking to AI?
Rules differ by state. Florida's bar requires AI chatbots that talk with clients or third parties to say they are an AI program and not a lawyer or employee[6], and Utah requires disclosure when a consumer clearly asks[8]. Putting a short disclosure in the greeting covers both and costs nothing.
What if a caller pushes for legal advice?
A well-configured product says that only an attorney can answer that, offers a consultation and moves on. Test it: have a tester ask the same question three different ways and check every recording.
Can it book consultations straight into our calendar?
Some products can. Ask which calendar it writes to, how it handles a slot that is already taken and what happens when nothing fits. Where no calendar is connected and tested, the safer pattern is a consultation request your staff confirm.
Does it replace our receptionist?
Usually not on day one. Most small firms start with after-hours and overflow coverage, keep their own person for daytime calls and existing clients, and widen the hours once a month of recordings shows the intake is right.
References
Illinois Supreme Court Commission on Professionalism (2Civility) (2024). Phone Calls Ignored, Emails Unanswered: Fixing the First Impression Problem for Law Firms (opens in a new tab) ↑
American Bar Association Standing Committee on Ethics and Professional Responsibility (2023). ABA Formal Opinion 506: Responsibilities Regarding Nonlawyer Assistants (opens in a new tab) ↑
Reporters Committee for Freedom of the Press (2026). Introduction to the Reporter's Recording Guide (opens in a new tab) ↑
American Bar Association (2025). Rule 1.18: Duties to Prospective Client, ABA Model Rules of Professional Conduct (archived copy) (opens in a new tab) ↑
U.S. Census Bureau (2026). Large Firms With at Least 20 Employees Biggest AI Users (opens in a new tab) ↑
The Florida Bar (2024). Florida Bar Ethics Opinion 24-1 (opens in a new tab) ↑
Pew Research Center (2025). How Americans View AI and Its Impact on Human Abilities, Society (opens in a new tab) ↑
Utah State Legislature (2025). S.B. 226 Artificial Intelligence Consumer Protection Amendments (2025 General Session, enrolled) (opens in a new tab) ↑
Federal Register (Federal Communications Commission) (2024). Implications of Artificial Intelligence Technologies on Protecting Consumers From Unwanted Robocalls and Robotexts (proposed rule) (opens in a new tab) ↑
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Job Flexibilities and Work Schedules (2024-25 data) (opens in a new tab) ↑