AI Receptionist for Law Firms: What to Automate First

In summer 2024 Clio paid a research agency to call 500 US law firms posing as prospective clients. 40 percent answered the phone. Counting callbacks, 52 percent made contact at all, which leaves 48 percent essentially unreachable. Five years earlier the same study found 56 percent picked up. Firms went backwards.

So when a managing partner asks me whether an AI receptionist for law firms is worth the setup, the pickup rate is the easy half of the answer. The intake script is the half that decides whether the thing helps you or quietly costs you a case.

The short answer

An AI receptionist for law firms answers every inbound call, captures the caller's name, contact details, matter category and jurisdiction, checks that name against your existing matters, and either books a consultation or routes something urgent to a human. What it should not do is take a detailed account of the facts before anyone has cleared the conflict.

Here is what most people get wrong

Almost every voice agent build starts from the same instinct: qualify harder, ask more, hand the attorney a full brief so the consultation opens warm.

For a plumbing company that instinct is correct. For a law firm it's a liability.

ABA Model Rule 1.18 gives a prospective client a duty of confidentiality even when you never sign them. Formal Opinion 492, issued in June 2020, states the consequence plainly: if you obtain information that could be significantly harmful to that person, you can be barred from representing the other side, and the bar extends to your whole firm unless you screen it properly. It also tells lawyers to caution prospective clients against volunteering detail in a first contact, because the more you collect the likelier some of it crosses that line.

So picture a keen voice agent at 11pm asking an inbound caller to describe what happened and what they want out of it. All of it gets transcribed and filed. Three weeks later the opposing party rings the same firm, and you find you've built a conflicts machine and pointed it at yourself.

Nobody sells it that way. Every legal intake demo I've sat through treats depth of qualification as the headline feature. For most practice areas it's the bug.

What the first call should actually collect

Keep it thin on purpose.

Name, phone, email. Matter category at the level of personal injury, family, employment or immigration. Jurisdiction, because a caller two states over is somebody else's client. How they found you. And whether a date is driving the urgency, like a hearing or a custody handover at the weekend.

That's enough to run a conflicts check, judge whether the matter fits the practice, and put a consultation on a calendar. It isn't enough to disqualify anybody.

I write the boundary straight into the agent instructions. When the caller starts narrating facts, the agent acknowledges it and steers back: "I want the attorney to hear that from you directly, so let me get you booked in first." Callers accept it. They already expect the lawyer, not the front desk, to take the story.

Run the conflicts check before you offer times, not after

Most firms run conflicts when the file opens. That's too late if your receptionist already booked the consultation and the caller already talked.

The order that works:

Capture the thin intake.
Check the caller's name, plus the adverse party's name if it came up unprompted, against existing and closed matters.
Offer appointment times only once that comes back clean.
If anything flags, the agent says the firm will confirm availability and follow up, and a human makes the call.

Step two is the one everyone skips, because it's the only one needing a live lookup into your practice management system. Clio, Smokeball and MyCase all expose enough through their APIs to check a name in the time it takes the agent to say "let me just confirm something." Build it. An intake bot without a conflicts lookup is a faster way to make a mistake.

There's a second rule worth reading before you sign any vendor. Model Rule 5.3 and its third comment cover nonlawyer help from outside the firm, and they put the supervision burden on you, including the terms that protect client information. A voice platform taking prospective client calls sits inside that. Ask where recordings live, how long transcripts are kept, who at the vendor can read them, and whether you can delete a caller's record on request. Vague answers there mean the answer on the contract is no. We ran the same diligence for clinics in our HIPAA and AI receptionists piece, and the pattern is identical even though the statute isn't.

Where the calls you're missing actually are

After hours is the obvious one. Somebody arrested on a Friday night, or a spouse served with papers at 7pm. The caller is at peak motivation and your voicemail is what answers. Clio found more than half of firms failed to return a voicemail inside 72 hours, and by then that caller has retained someone else.

The less obvious one is Monday between nine and eleven. Everybody rings at once, your receptionist takes them one at a time, and the rest hit a hold queue they abandon. That's a concurrency problem rather than a staffing problem, and we broke the math down in why one receptionist cannot take six calls at once. Why the speed-to-lead stat everyone quotes is nearly two decades old sits in our speed to lead post.

What it won't do for you

No legal advice, and it should refuse when asked. No fee quoting beyond a published consultation rate. A caller who is distressed, or ringing from a police station on a bad line, needs a human fast rather than a script grinding through step four. And it can't judge whether a matter is worth taking, which is a partner's job and stays one.

If a vendor tells you their agent handles all of that, they're selling you the demo instead of the deployment.

Where to start

Pick one number and one practice area. Point your after-hours line at the agent, leave business hours alone, and give it a fortnight. You'll end up with a list of the calls that used to hit voicemail and what they were worth, which is the only number that settles this argument internally.

Our pricing runs on booked consultations rather than minutes or seats, and the free two-week proof of concept exists so you can run that test before anything gets charged. Send us the line and we'll wire the conflicts lookup into the intake flow. You can read the transcripts yourself.

FAQ

Is an AI receptionist ethical for a law firm to use?

Nothing in the ABA Model Rules bans it. Rule 5.3 makes you responsible for supervising nonlawyer help from outside the firm and for the arrangements protecting client information, so the obligation is diligence on the vendor and the script rather than avoiding the tool.

Can an AI receptionist run a conflicts check?

Yes, if it can reach your practice management system. Clio, Smokeball and MyCase all support a name lookup by API. Without that integration the agent books consultations blind, which is the main risk to avoid.

How much detail should an AI receptionist take from a caller?

As little as possible. Contact details, matter category, jurisdiction and urgency. Formal Opinion 492 warns that collecting significantly harmful information from a prospective client can disqualify the whole firm from the other side.

What happens when a caller refuses to talk to an AI?

The agent should offer a human immediately and either transfer or take a callback. Arguing with a caller who asks costs you the matter.

Does an AI receptionist replace a legal intake specialist?

No. It covers the hours and the overflow your specialist can't, then hands over a clean record. Firms that get value from it keep the human for consultations and give the agent everything outside office hours.

How long does it take to set up an AI receptionist for a law firm?

A single line with a thin intake script and a calendar integration takes a few days. The conflicts lookup into your practice management system is what sets the real timeline.

Will callers know they're talking to an AI?

Most will work it out, and the agent should say so if asked. Several states now require disclosure for automated callers, so build it in rather than hoping nobody notices.

Sources

Clio, 2024 Legal Trends Report secret shopper study, by the agency Lux, 20 June to 5 July 2024: highlights
Clio, 2019 Legal Trends Report study of 1,000 firms: press release
ABA Formal Opinion 492 on duties to prospective clients under Rule 1.18: ABA Journal summary

AI Receptionist for Law Firms: The Intake Call That Decides Who Gets the Case