Key takeaways
- A live service is stronger on emotional and unscripted calls; an AI receptionist is stronger on consistency, bursts of calls and a full record of every call.
- The hidden cost of message-only answering is the callback that repeats the intake, usually made by someone paid more than the agent who took the message[2][3].
- Test languages on a recorded call with either option; more than one in five US residents speak a language other than English at home[4].
- Supervision duties under Rule 5.3 apply to both, so plan call sampling and test calls whichever you choose[5].
The same call, answered two ways
Take an ordinary weeknight call to a family law firm. A woman says her husband has moved out and taken the children to his mother's house, and she wants to know what she can do before Friday. With a live answering service, an agent at a shared call center picks up in your firm's name, reads your script off a screen and types a message. With an AI receptionist, software picks up, asks the questions you approved and sends a summary with the recording. Both count as coverage. The difference is what is waiting for your intake person at 8:30 the next morning.
Live answering is a mature industry. In 2023 the US had 1,035 telephone answering service establishments employing 41,582 people[1], and customer service representatives in the wider business support services group, which includes answering services and call centers, earned a mean of $18.83 an hour in May 2025[2]. The agent who takes your call is usually taking calls for many businesses in the same shift. That shapes everything else in this comparison: how deep the intake goes, how consistent it is and what the firm has to supervise.
Side by side
| Factor | Live answering service | AI receptionist |
|---|---|---|
| Who answers | A trained agent from a shared pool, working from your script | Software following your approved greeting and questions |
| Intake depth | Often message-only; full intake is usually a higher tier | As deep as the questions you approve, on every call |
| Consistency | Varies by agent, shift and how busy the queue is | Same questions in the same order; a setup mistake repeats until fixed |
| Several calls at once | Depends on staffing at that hour; peaks can mean hold time | Answers simultaneous calls without a queue |
| Emotional or odd calls | A person can improvise and reassure | Handles what it was set up for; needs a clear route to a person |
| Languages | Bilingual agents, if the service employs them on that shift | Varies by product; test rather than assume |
| What the firm receives | A typed message, sometimes a recording | Summary, recording and transcript of each call |
| Changing the script | A change request, then retraining across the agent pool | Edit, test and switch over the same day |
Neither column wins every row. A person is better with the caller who is crying or confused, and with the call nobody planned for. Software is better at doing the same thing correctly at 3 a.m. and at 3 p.m. on a Monday when six calls land together. The rest of this comparison is about which rows carry the most weight in your practice. If you are still choosing between human options, legal answering service against virtual receptionist separates those first.
Intake depth and the callback that repeats it
The most expensive weakness of message-only answering never shows up on the invoice. The message reads “Maria, custody question, please call back,” so your intake person rings Maria and asks everything again: the names for the conflict check, the county, the children's ages, whether there is already an order. That assumes Maria picks up. Eight in ten Americans say they generally don't answer when an unknown number calls[6], so one thin message can turn into two or three attempts and a voicemail.
Full-intake answering exists, and some services do it well; message-only against full-intake services explains the tiers and what each costs you in follow-up. An AI receptionist that is set up properly takes the full intake on every call by default, because asking eight questions instead of two costs it nothing extra. The difference lands in your staff's week.
The person taking a message is paid about what a legal receptionist earns. The person redoing the intake is usually a paralegal at around $32 an hour, and in a small firm it is often the lawyer, at a mean of nearly $92[2][3][7][8]. The model below puts a monthly figure on that callback time. Set the inputs from one week of your own messages.
| Input | Example |
|---|---|
| Thin messages per day | 8 |
| Minutes to call back and redo intake | 12 |
| Share that would arrive complete instead | 60% |
| Hourly wage of the person calling back | $32.45 |
| Calls a month | 174 |
| Staff hours on these calls | 35 |
| Hours handed off | 21 |
| Monthly value of that staff time | $676 |
How it is worked out: Calls a day × 21.7 working days × minutes per call ÷ 60 × share handled = hours; hours × hourly cost.
Assumptions
- Hourly cost is the May 2025 mean wage for paralegals and legal assistants in legal services, before benefits.[3]
- Most people don't answer unknown numbers, so one message often needs more than one callback attempt.[6]
- Message counts, minutes and the complete share are placeholders. Count a week of your own messages to set them.
Example only. The figure is staff time, not lost cases; lost cases from slow callbacks would come on top.
Consistency, peaks and judgment
Consistency is where software is strongest and where its failures do the most damage. An agent pool drifts: one agent asks for the other party's name, the next forgets, a third skips the hearing date because the queue is long. Software does not drift, but if the conflict question sits in the wrong place it sits there on every call until someone moves it. That is why the first weeks of any AI setup need recordings checked against the script.
Peaks matter in practices where calls bunch up: a TV spot airs, a storm hits, the county jail releases its overnight list. A center that is thinly staffed at that hour puts callers on hold, and a caller on hold to a law firm often hangs up and dials the next one. Judgment runs the other way. A caller who says something unexpected, such as a threat, a medical emergency or a confession, needs a person or a precise rule, and a human agent is more likely to handle the unscripted moment well. On the AI side the fix is a short list of triggers that send the call to a person, and a tested fallback when nobody picks up.
- System
- SI Employee
- Caller
- Your team
01System
Firm forwards its line
After hours, and in the day when nobody picks up within four rings.
02SI Employee
Greeting and disclosure
Firm name, automated-assistant notice and recording notice.
03SI Employee
Approved intake
Conflict names, matter type, key dates and a callback window.
04Caller
Caller needs a person
Asks for one, or says something on the firm's trigger list.
05System
Escalation tree
Up to three people in a set order, then a tested fallback.
06SI Employee
Nobody answers
The caller is told plainly and an urgent message is saved.
07Your team
Morning review
The intake lead reads the summary and listens where needed.
Example setup. The triggers, the order of people and the hours are the firm's choice.
That flow is how an SI Employee is set up for a law firm. The firm forwards its own number for the hours it chooses, approves every question, and sets an escalation tree of up to three destinations with a fallback. During a call, an attorney or staff member can listen in, whisper guidance the caller does not hear, or take the call over. The criminal defense example models a three-attorney firm sorting its night calls this way, with only custody-related calls reaching the on-call attorney.
- New inquiry3 events
- Intake captured5 events
- Urgent routing2 events
- Consultation2 events
- Firm review2 events
Activity
- Call
After-hours call answered
Example caller · greeting includes AI and recording notice
- Call
Practice area: estate planning
In scope for the firm · approved will and trust script
- Call
Contact and callback captured
Name, mobile, email · prefers afternoons
- Call
Names for conflict review
Spouse and adult children listed for the firm's own check
- Call
Consultation booked
Tue 2:30 pm · video · tested calendar
- Call
Summary sent to intake team
Summary, recording and transcript in the AssistCall account
- Call
Lunchtime overflow call
Family law · custody modification question
- Call
Outside the pilot practice area
Approved referral wording · noted for firm review
- Call
Caller has a hearing tomorrow
Matches the firm's urgent rule: court date within 48 hours
- Call
Trying on-call attorney
Destination 1 of 3 · no answer after 25 seconds
- Call
Connected to backup attorney
Destination 2 of 3 · transfer completed
- Call
Message for existing client's attorney
Document question · no case details discussed
- Call
Office hours and parking question
Answered from the approved profile · no message needed
- Call
Consultation request taken
Probate · team to confirm a time with the caller
- 47Inquiries answered this weekExample
- 19Answered after hoursExample
- 5Urgent calls routedExample
- 16Consultations booked or requestedExample
- Under 1 minMedian time to staff alertExample
Example only. Fictional business, people and numbers; not a recording or a live account.
Languages
22%
of US residents aged 5 and older speak a language other than English at home; about 61% of them speak Spanish[4]
For many practices that makes language a deciding row rather than a nice extra[4]. Reporting on ABA Formal Opinion 500, the ABA Journal noted that lawyers must obtain a qualified, impartial interpreter or translator where one is needed[9]. That duty sits with the lawyer, but the first call sets the caller's expectations of what the firm can do.
Do not take a sales sheet's word for either option. A service may list Spanish but have one bilingual agent per shift. An AI product may hold a fluent conversation and still return a summary that garbles names or lands in the wrong language. Ask for a recorded test call in each language your callers use, with names and addresses spelled aloud, and check exactly what reaches the attorney. Intake for callers with limited English proficiency covers the intake side.
Supervision duties stay with the firm
Whichever you choose, the firm stays responsible. ABA Model Rule 5.3 requires lawyers with supervisory authority to make reasonable efforts to ensure nonlawyer assistance is compatible with their professional obligations[5]. The comment quoted in Formal Opinion 498 extends that to services outside the firm, weighing the provider's experience and reputation and the terms that protect client information[10].
Florida's guidance adds a point many firms miss. Nonlawyers who conduct an initial intake interview should identify their nonlawyer status, keep to factual questions, give no legal advice and send legal questions to a lawyer[11]. That applies to an agent answering in your firm's name as much as to software. What differs is the object of supervision: with a service you supervise a vendor through reports and call samples; with an AI receptionist you supervise a configuration through recordings. Supervising staff and vendors under Rule 5.3 covers the duty itself.
- Every month, listen to ten calls end to end, picked at random rather than from the vendor's highlights.
- Check each message or summary for the conflict names and the one date that matters most in that practice area.
- Once a month, call the line after hours as a fictional caller, using the routine in testing your answering service with test calls.
- Keep the script versioned: date every change and record who approved it.
- Review who at the vendor can hear your calls and how long recordings are kept.
When each one wins
A live answering service tends to win when your callers are likely to hang up on automation, when many calls are emotional or unpredictable and warmth is part of your intake, or when you need a person to follow an established relay protocol for on-call staff. Attitudes matter here: half of US adults say they are more concerned than excited about the growing use of AI in daily life[12].
An AI receptionist tends to win when you want the full intake on every call at any hour, when calls arrive in bursts a shared center struggles to absorb, when you change scripts often and need the change live the same day, and when you want a recording and transcript of every call to coach and audit against.
Plenty of firms use both, AI first with people behind it or the reverse. Answering service alternatives lays out the layered options, and the AI receptionist buyer's guide has a vendor scorecard and a test plan. The decision framework in whether an AI receptionist is worth it adds supervision time to the cost side. AssistCall's family law intake example works the same choice through for a firm much like the one in the opening call.
Frequently asked questions
Is an AI receptionist cheaper than an answering service?
Compare the whole cost, not the invoice. A cheaper message-only service can cost more once you count the staff time spent calling back to redo intake, which the model above estimates from your own message volume and wage rates[3].
Can callers tell they are talking to software?
Usually, and they should be told. Put an automated-assistant disclosure in the greeting and make a person easy to reach. Florida's guidance already expects intake by nonlawyers to identify itself as such[11].
Do answering-service agents have to say they are not lawyers?
Under the Florida guidance on nonlawyer intake interviews, yes: they should identify their nonlawyer status, ask factual questions only and refer legal questions to a lawyer[11]. Check your own state's rules and put the line in the script.
Can we use both?
Yes, and many firms do. A common split is AI answering after hours and on overflow, with the firm's own staff or a live service handling daytime calls and callers who ask for a person.
Which handles Spanish-speaking callers better?
It depends on the provider, not the category. Ask for recorded test calls in each language your callers use and check that names, dates and addresses reach the attorney correctly.
References
U.S. Census Bureau (2023). County Business Patterns 2023: U.S. detail file (cbp23us), NAICS 561421 Telephone Answering Services (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Occupational Employment and Wage Statistics, May 2025: Business Support Services (NAICS 561400), Customer Service Representatives, mean hourly wage (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Occupational Employment and Wage Statistics, May 2025: Legal Services (NAICS 541100), Paralegals and Legal Assistants, mean hourly wage (opens in a new tab) ↑
U.S. Census Bureau (2025). New Data on Detailed Languages Spoken at Home and the Ability to Speak English (opens in a new tab) ↑
American Bar Association (2026). Rule 5.3: Responsibilities Regarding Nonlawyer Assistance, ABA Model Rules of Professional Conduct (archived copy) (opens in a new tab) ↑
Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Occupational Employment and Wage Statistics, May 2025: Legal Services (NAICS 541100), Lawyers, mean hourly wage (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Occupational Employment and Wage Statistics, May 2025: Legal Services (NAICS 541100), Receptionists and Information Clerks, mean hourly wage (opens in a new tab) ↑
ABA Journal (American Bar Association) (2021). New ABA ethics opinion clarifies obligations for language access in the lawyer-client relationship (opens in a new tab) ↑
American Bar Association Standing Committee on Ethics and Professional Responsibility (2021). Formal Opinion 498: Virtual Practice (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) ↑