Picture the call every law firm knows it is missing. It comes in at 9:40 on a Friday night: someone has just been arrested, served with papers, or told a hearing is set for Monday. They search, they find your firm, they call — and they reach voicemail. Most legal callers in that moment do not leave a message and wait. They dial the next firm on the list.
Vendor websites quote dramatic statistics about how many calls law firms miss and how much revenue walks away. Treat those numbers as marketing, not measurement: they come from the companies selling the fix. What you can measure is your own call log — most firms that pull four weeks of phone records find a real gap in the evenings and at weekends.
An AI receptionist can close that gap — but in a regulated profession, how you deploy it matters as much as whether you deploy it. This is the implementer’s guide the sales pages skip: intake script, conflict-check data rules, disclosure, recording consent, and a two-week pilot.
The Short Answer: What an AI Receptionist for Law Firms Does
An AI receptionist for a law firm answers calls around the clock, runs a scripted intake, collects the names needed for a conflict check, books consultations into your calendar and practice-management system, and routes true emergencies to a human. It never gives legal advice, quotes fees on a matter, or decides conflicts.
Everything else here is about doing those jobs safely — the script, the data rules and the escalation paths you design before launch are what separate a helpful system from a risky one.
Four Jobs It Should Do — and the Lines It Must Never Cross
Give the system a narrow job description: answer and screen every call, run the structured intake, book consultations, and log everything for follow-up. A well-configured AI voice agent handles those jobs consistently at 2 p.m. or 2 a.m. Everything outside the job description gets escalated, not improvised.
| Safe to delegate to the AI | Never delegate — human only |
|---|---|
| Answering calls and identifying the caller’s need | Giving legal advice or assessing a case’s merits |
| Collecting scripted intake details | Deciding whether a conflict of interest exists |
| Checking names against the conflict database and flagging matches | Quoting fees, predicting outcomes or promising representation |
| Booking, rescheduling and confirming consultations | Handling a caller in crisis — transfer immediately |
| Answering approved FAQs: hours, location, parking, what to bring | Discussing live matters with opposing parties, insurers or media |
| Sending reminders and call summaries | Accepting service of documents without human confirmation |
If a vendor demo shows the AI doing anything in the right-hand column, walk away — do not be impressed.
Designing the Intake Script, Question by Question
The intake script is the product. Write it as you would brief a careful human receptionist, then test it relentlessly:
- Open with disclosure and consent. The first sentence identifies the assistant as automated and states the call may be recorded. No exceptions.
- Identify the caller and the matter type. Name, callback number, whether texting is acceptable, and the practice area — everything later branches by matter type.
- Collect the conflict-check names early. Full legal names of the caller and every other party, spelled out — before any narrative is collected (the next section explains why).
- Ask the minimum an attorney needs for a first decision. Key dates (incident, service of papers, court date or deadline), whether another firm already represents the caller, and whether anyone is in danger.
- Screen for urgency and escalate. A court date within days, an arrest or a safety risk triggers an immediate transfer attempt to your on-call human, with a fallback if no one picks up.
- Close with a booking or a clear next step. Offer times, confirm, explain what happens next and what to bring, and send confirmation by text or email.
Then adapt the middle questions per practice area. Personal injury needs the incident date and location, any treatment started, and the insurers involved. Family matters need whether children are involved and whether court orders exist. Criminal matters need the charge, the court date and custody status. Immigration needs the caller’s status category and any deadlines. Keep each branch short — intake should produce a record an attorney can act on in two minutes.
One discipline applies to every branch: collect the minimum necessary. Never ask a first-time caller for a full date of birth, ID numbers or financial details — if a field is not needed for the conflict check or first review, it does not belong in the script.
Conflict Checks: Data the AI May Collect — and Decisions It Must Never Make
A conflict check is a search, not a judgment — and the AI’s role ends at the search. The safe pattern:
- Collect exact data. The caller’s full legal name, known aliases, and full names of all adverse and related parties — spellings confirmed letter by letter.
- Search the same database your firm uses. Names are checked against your practice-management system’s conflict records (such as Clio or MyCase), never against a separate list the AI keeps itself.
- Flag, never clear. Any exact or partial match is flagged for attorney review. The AI never tells a caller they are “cleared” or that the firm can represent them.
- Stop collecting on a possible match. The script goes neutral: it stops gathering details and explains the firm must review first — collecting a narrative after a conflict surfaces can hand your firm information it should never have received.
- Protect the data like client data. Intake records hold sensitive information about people who may never become clients, so access controls, retention limits and secure storage apply from day one, as with any data security review.
The output is a flagged record and a neutral summary. The decision — conflict or no conflict — stays with a qualified lawyer, every time.
Disclosure and Recording Consent: Rules That Vary by Jurisdiction
Two duties meet at the start of every call, and the same opening lines handle both.
AI disclosure. Callers should know promptly that they are speaking with an automated assistant, not a person. Some jurisdictions require automated systems to identify themselves to the public, and regulators everywhere treat misleading a caller as a conduct problem. In the UK, where firms are SRA-regulated and callers say “solicitor” rather than “attorney,” the principle is identical: disclose, in plain language, in the first seconds. Australian professional-conduct framing lands in the same place: the obligations sit with the solicitor.
Recording consent. Most US states allow recording with one party’s consent, but a significant minority require everyone’s — and you cannot know which state an after-hours caller is dialling from. In the UK, recording engages UK GDPR: the call is personal data, needing a lawful basis and transparency. Australian recording rules also vary by state. The practical answer everywhere: announce the recording and the automated assistant at the top of every call — one compliant opening script beats four jurisdiction-specific ones.
This is implementation guidance, not legal advice. Before launch, have your ethics counsel read the final script — the cheapest risk reduction in the project.
Connecting It to Clio, MyCase and the Rest of Your Stack
If the system only emails a summary after each call, someone still has to re-type everything. A proper setup pushes the data where it belongs. Through AI integrations with your practice-management system, each call creates or updates the contact record, attaches the summary and recording link, runs the conflict search against live data, and places the consultation on the right attorney’s calendar.
Vendor pages routinely promise Clio and MyCase integration. In the pilot, verify its depth: does it create a matter or only a contact? Does it de-duplicate existing callers? Does the conflict search cover archived matters? The answers separate a real integration from a call logger.
What It Costs: Pricing Models, Honestly Explained
Ignore headline prices and look at the model — the model decides what a busy month costs you:
- Flat monthly plans bundle an allowance of calls or minutes. Predictable — but check overage costs and whether after-hours calls share the allowance.
- Per-minute pricing suits low, spiky volumes and punishes long calls. Legal intake calls are often long — model your real average duration.
- Per-call pricing forecasts simply from your call log. Confirm what counts as a call: is a 20-second wrong number billed like a 12-minute intake?
Published plans currently run from under $100 a month for basic AI-only tiers to several hundred for high-volume or hybrid services. Prices change frequently — verify against live vendor pricing before budgeting, and add setup plus the staff time spent on the pilot below.
Size the opportunity with your own data: pull four weeks of phone records and count unanswered after-hours calls and weekday calls that rang out. That count — not a vendor’s missed-call statistic — tells you whether the project pays.
The Two-Week Pilot: Test Before It Speaks for Your Firm
Never switch the system on for all calls on day one. Run a controlled pilot:
- Week one — after-hours only. Forward calls outside office hours, your lowest-risk, highest-gap window; the team answers in-hours as normal.
- Review every transcript for the first days. Listen for script drift: skipped questions, misspelled names, a caller unsure what happens next.
- Week two — add weekday overflow. Calls unanswered after a set number of seconds roll to the AI instead of voicemail.
- Run planted test calls. Staff call in as fictional clients: a routine booking, a caller whose adverse-party name matches an existing matter, and an “emergency” — testing the conflict flag and escalation path, not the small talk.
- Score it, then decide. Use a scorecard agreed before the pilot starts, so “it feels fine” is not the acceptance test.
| Pilot metric | How to measure it | Pass bar to agree in advance |
|---|---|---|
| Answer rate | Routed calls answered by the AI | 100% of routed calls |
| Disclosure and consent delivered | Transcript check on sampled calls | 100% of calls |
| Intake completeness | Intakes containing every required field | Set your bar — 95% is a common starting target |
| Conflict flags | Planted matching names flagged for review | 100% — no exceptions |
| Booking accuracy | Test bookings on the right calendar and time | 100% |
| Escalation | Emergency tests reaching a human or the agreed fallback | Within the limit you set, every time |
Fail the conflict or escalation rows and it does not go live, whatever the other scores say. Pass, and you can widen the hours gradually, keeping transcript reviews weekly for the first quarter.
Build, Buy or Settle for a Hybrid?
Off-the-shelf legal answering tools are fastest, but you accept their script logic and their integration depth. A hybrid service adds humans behind the AI for calls needing judgment, at higher cost. A custom-built agent — an AI calling agent configured around your intake rules, conflict workflow and practice-management stack — takes more setup but gives you control of the script, data flows and escalation rules, inside your wider AI solutions rather than beside them. For most small and mid-size firms: pilot first, and let the scorecard, not the sales call, set the scope.
At AI Dev Expert, we build AI receptionists for professional-services firms as pilot-first projects covering intake script design, conflict-check data rules, disclosure wording, practice-management integration and the two-week scorecard above. If your website must also convert the callers the AI books, our web design for law firms team works alongside the automation team. Tell us about your firm via our contact page, or get a quote for a scoped pilot.
Frequently Asked Questions
How much does an AI receptionist cost for a law firm?
It depends on the model — flat monthly, per-minute or per-call — plus call volume and coverage hours. Published AI-only plans start under $100 a month; hybrid services cost several hundred.
Can an AI receptionist perform a conflict check?
It can collect the names and search your firm’s conflict database, then flag matches for attorney review. It cannot decide a conflict or clear a matter.
Does it need to disclose it is AI?
Yes — disclose promptly at the start of the call, and announce recording. The requirements vary by jurisdiction, but a single compliant opening script works across them.
What if a caller is in crisis?
The script screens for danger and emergencies, then transfers immediately to a human. If no one answers, it follows the fallback you agreed — never leaves a person in crisis to voicemail.
Will it integrate with Clio or MyCase?
Most legal tools promise it; pilot-test the depth. Real integration creates or updates contact records, attaches summaries and recordings, and searches the live conflict database.
Sources and Further Reading
Sources and background for the rules and figures cited above: CloudTalk — Best AI voice agents for law firms; Phone2 — AI receptionist for law firms; Nexwin — AI receptionist for law firms.
