AI Receptionist for Law Firms in 2026: Intake Tools, Verified Prices, and Where the Bot Must Stop

TL;DR: AI receptionists answer a law firm's phone around the clock, run scripted intake, and book consultations. Verified July 2026: Ruby runs $250 to $1,725 per month; LawDroid's intake builder is $99 per user per month; Smith.ai has pulled its public pricing behind a contact form. None of these tools can give legal advice, make a conflict determination, or accept an engagement, and every caller's information lands on a vendor's servers before it reaches you. That last part is the section vendors skip. We have not run our hands-on suite yet, and we sell none of these tools. Not legal advice.

Search "ai receptionist for law firms" and vendors wrote nearly everything you will find: listicles that rank their own product first, "resolves 90 to 95% of calls" claims that trace back to no study, and pricing pages that increasingly hide the price. Smith.ai , the biggest name in legal answering, gated both of its pricing pages behind a contact form as of our July 2026 direct fetch. Third parties now rank for "smith.ai pricing" by publishing the numbers Smith.ai will not.

This page does the missing work. It compares the intake vendors law firms actually shortlist, labels every price by where it came from, and spends its longest sections on the two questions no vendor page answers: where AI intake legally ends and a lawyer begins, and what happens to a prospective client's confidences after the bot writes them down. A misconfigured intake bot is not a customer-service problem for a law firm. It is a Rule 1.18 problem wearing a headset.

Best AI receptionist for law firms in 2026 (at-a-glance)

An AI receptionist for a law firm answers calls around the clock, runs a scripted intake, collects adverse-party names for your conflict process, and books consultations into your calendar or CRM. As of July 2026, it cannot give legal advice, run a conflict check, or accept an engagement. Those stay with a licensed lawyer.

Prices labeled by source: "vendor page" means we read it on the vendor's own pricing page in July 2026; "vendor page markup" means the numbers were extracted from the page's code before Smith.ai gated the page; "third-party" means a report we could not confirm against the vendor. Where two sources disagree, we show both.

Disclosure: we have no affiliate or business tie to any tool named here as of publication. If that changes, this line will say so. Our funding model is in our editorial policy .

Where AI intake ends and a lawyer begins

Every vendor sells "legal intake." None of them draws the line that keeps you out of a bar complaint, so here it is, drawn from what the tools actually do and what the profession's rules actually reserve.

The bot can collect. Caller name, callback number, matter type, a plain-language description of the problem, the names of other parties involved, how the caller found you, and a consultation slot on your calendar. Smith.ai 's legal intake runs custom screening scripts and pushes the answers into Clio, Lawmatics, or HubSpot (per Lawyerist's review ). LawDroid Builder lets you script the same question tree into a website chatbot yourself. This is data entry with a voice, and automating it is legitimate.

The bot cannot decide. Three decisions sit on the lawyer's side of the line, and no intake product moves them:

First, the conflict determination. An intake service collects adverse-party names; the conflict check itself is the lawyer's non-delegable call. Smith.ai 's own guidance frames its role as data collection feeding the firm's conflict process , and Lawyerist notes even Smith.ai 's human agents run conflict screening only "as part of a supervised workflow." A bot that tells a caller "no conflict, we can take your case" has made a legal judgment it is not licensed to make, and bound you to it.

Second, legal advice. The moment the script drifts from "what happened?" to "you likely have a strong claim" or "you should file before the deadline," the AI is practicing law under your firm's name. Configure intake scripts to gather facts and stop. Test this with adversarial calls before go-live, because callers will push: "do I have a case?" is the most natural question in the world, and the script needs a hard-coded deflection to a lawyer for it.

Third, the engagement. Fee quotes, scope of representation, and accepting the matter are the lawyer's. An AI that improvises a fee estimate has started negotiating a contract on your behalf.

The practical build: script the bot to the collection column, route every decision-column question to a human, and log the transcripts so you can audit where callers dragged it over the line. Not legal advice; your state bar's rules control.

The privilege problem: where a prospective client's confidences actually go

This is the section that makes a legal AI receptionist different from a dental or plumbing one, and it is the section every vendor page skips.

A person who calls your firm about representation is a prospective client, and prospective clients get confidentiality duties under Model Rule 1.18 even if you never take the case. Now trace the call. The caller describes their dispute, names the opposing party, maybe admits something damaging. The AI transcribes all of it. That transcript now exists on the intake vendor's servers, was processed by the vendor's speech-to-text provider, ran through an LLM API the vendor rents, and traveled over a telephony carrier. Four companies touched a prospective client's confidences before you read the summary.

ABA Formal Opinion 512 (July 29, 2024) speaks to exactly this. Before entering any information relating to a client's representation into a generative AI tool, a lawyer must assess whether it will be "disclosed to or accessed by" others, and self-learning tools require informed client consent beyond boilerplate engagement-letter language ( opinion PDF ; ABA announcement ). An intake bot is a generative AI tool ingesting representation-related information all day. The opinion's logic does not pause because the tool answered the phone instead of drafting a brief.

So the buying criterion vendors never volunteer: get four answers in writing before the bot takes a call. Is caller data used to train models, and is the exclusion contractual or a dashboard toggle? What is the retention period for recordings and transcripts, and can you set it? Who are the subprocessors (the LLM provider, the transcription provider, the carrier), and what are their retention terms? What happens to stored intake data if you cancel? A vendor that answers slowly is answering.

Two more consequences follow. Intake transcripts are records, and records are discoverable; assume opposing counsel could someday read the bot's summary of your first conversation with the client. And if a malpractice or disqualification fight ever turns on what a prospective client disclosed, the vendor's server logs become evidence you do not control. None of this means the tools are unusable. It means the data-practices addendum matters more than the feature list, which is roughly the opposite of how these products are sold.

The bar-rules box: disclosure, callbacks, and recording consent

Three regulatory threads wrap around a law-firm phone bot, and each one has a date and a citation. Not legal advice; verify current status with counsel in your jurisdiction.

The bot must identify itself as a bot. Florida Bar Ethics Opinion 24-1 puts client-facing chatbots under the lawyer-advertising rules and states that "a lawyer must inform prospective clients that they are communicating with an AI program and not with a lawyer or law firm employee" (wording per Clio's roundup of AI ethics opinions and Justia's 50-state survey ; the opinion itself is at floridabar.org). Disclosure-of-AI is the near-universal thread across state ethics opinions in that survey. If your intake bot opens with a human name and no disclosure, in Florida you already have a problem, and other states are converging on the same rule.

Outbound is a different legal universe than inbound. The FCC's declaratory ruling of February 8, 2024 (FCC 24-17) classified AI-generated voices as "artificial or prerecorded voice" under the TCPA: outbound AI-voice calls require prior express consent, written consent for marketing ( FCC ruling ). Answering an inbound call is outside that ruling. But the features vendors upsell — automated callbacks to missed leads, follow-up texting sequences — are outbound, and TCPA statutory damages run $500 to $1,500 per call, uncapped (per a third-party legal summary ). A law firm, of all businesses, should not be the defendant that learns this distinction in discovery.

Recording consent is not optional plumbing. AI receptionists record and transcribe essentially every call; that is how they work. Roughly a dozen states require all-party consent to recording (the count varies with how Michigan, Oregon and Connecticut are treated), including California, Florida, Illinois, Pennsylvania, Washington, Maryland, and Massachusetts (per a third-party state guide ). California Penal Code §632 requires all-party consent, with $5,000 statutory damages per violation under §637.2, and a wave of CIPA class actions is currently targeting AI call analytics specifically ( third-party analysis ). The safe configuration: a disclosure that precedes any transcription, applied to every call under the strictest state you serve. Your callers are in litigation-prone situations by definition; assume some of them will notice.

What lawyers actually report

The only named practitioner reviews we found live in Lawyerist's Smith.ai review, and they split cleanly.

The positive reports are about humans, not AI. "The phone service is on par with any receptionist... that would have worked at my old firm," per Austin B. "Having a live person answer all of my calls and gather specific, detailed information from my clients... is invaluable," per Rachel A. ( Lawyerist ). Note what both are praising: the live-agent tier, the one that costs roughly six times more per call.

The negative reports are about operations. Aaron H.: "After we paid, we were told they could not access our callrail and that it would be more for them to have spanish speakers answer." Nicholas P.: "They started letting a lot of our calls go to voicemail in the middle of the workday... not responsive and not acceptable" (same source; individual user reports, unverified).

Two patterns from the wider market apply directly to law firms. First, third-party roundups of Smith.ai reviews report the AI auto-transferring calls to live agents without consent, which inflates per-call bills on hybrid plans (user reports via G2/Trustpilot summaries, unverified; Trustpilot overall ~4.3/5 across 336+ reviews as of January 2026). On a per-call contract, the escalation logic is a billing decision, and you do not control it. Second, callers hang up on bots: a 2026 OnePoll survey of 6,000 consumers found 31% would hang up if connected to AI, and 85% prefer a real person — with the caveat that the survey was commissioned by AnswerConnect , a human answering service with an obvious interest in the result. Discount it accordingly, but a personal-injury caller comparing three firms from a hospital bed is exactly the caller most likely to bail on a robot.

The cost math: per-call, per-minute, and the platform floor

Three billing models compete for your phone line, and the arithmetic between them is the closest thing to an honest comparison this market allows.

Per-call ( Smith.ai ): every answered call costs the same whether it is a 30-second solicitation or a 20-minute intake. Smith.ai argues per-minute vendors are incentivized to stretch calls; the counter-incentive is that per-call vendors profit from short ones. Its earlier vendor-page markup put in-tier AI calls at $1.50 to $3.00 each, and its own human tiers at roughly $7 to $11.50 per call — about a 6x gap between Smith.ai 's AI and Smith.ai 's humans for the same phone ringing.

Per-minute (Ruby, Rosie ): Ruby's $250/month buys 50 minutes — $5.00 per minute of human-with-AI-assist receptionist time, with billing starting when the receptionist takes the call, hold time included (per Ruby's help center ). Rosie 's AI-only $49/month buys 250 minutes — about $0.20 per minute, a 25x spread that tells you what the human in the loop costs.

The platform floor ( Retell ): the infrastructure many white-label "legal AI receptionists" resell prices at $0.07 to $0.31 per minute all-in on its public pricing page . A reseller charging $2 to $3 per roughly three-minute call is marking up raw platform cost around 5 to 10x. Some of that markup buys real things: legal scripts, CRM integrations, escalation logic, support. Some of it buys a logo. Knowing the floor tells you which question to ask on the sales call.

The law-firm-specific twist: your intake calls are long. A real potential-matter call runs 10 to 15 minutes of facts, parties, and dates. On per-minute billing that is $2 to $4.50 of AI time or $50+ of Ruby time per lead; on per-call billing it is one flat unit. Firms with high spam-to-lead ratios lean per-call; firms with few but long intake calls should price both models against a month of real call logs before signing anything.

Where these tools fall short

The catches, on the record, tool by tool — plus the category-wide ones.

• Smith.ai withdrew its public pricing behind a contact form (July 2026 direct fetch), which means every number in circulation is either stale markup or a third-party report. User reports describe unconsented AI-to-human transfers inflating per-call bills, mid-day calls reaching voicemail, and post-sale surprises on CallRail access and Spanish-language coverage (Lawyerist reviews; individual claims unverified).
• Ruby publishes plan prices but not overage rates, bills hold time, and draws recurring cost complaints on Capterra, where reviewers report finding competitors at half the price (third-party report). It is a human service with AI assist, so it never gets AI-cheap.
• LawDroid Builder is a toolkit, not a service: nobody answers your phone, and the intake logic is only as good as the question tree you build and maintain.
• Rosie and the generic AI-only tier know nothing about legal intake: no conflict-data fields, no practice-management integrations named, no bar-rules awareness in the scripts. You would be building the legal layer yourself on a $49 product, with overage reportedly $0.25/min (third-party review; verify on heyrosie.com).
• Retell-based white-labels inherit every compliance question on this page with an extra hop: now the subprocessor chain includes the reseller too.

And the category-wide gaps. No vendor publishes an independent accuracy study for legal intake — how often the bot mis-captures a name, a date, or an adverse party, which in this vertical is the difference between a lead and a missed conflict. No vendor publishes its escalation-decision logic, even though on hybrid plans that logic is also billing logic. And none of the pricing above includes the cost that dominates the real total: a lawyer's time reviewing every intake summary, because after ABA 512 and Rule 1.18, unreviewed AI intake is not a corner you can cut. We have not run our hands-on test-call suite yet; when we do, dated transcripts and a changelog will land here.

All guides in this topic

• AI Receptionist: the category hub — what these tools actually handle, the billing models, and the consent-law checklist that applies to every vertical.
• AI receptionist for dental offices — PMS integrations, the HIPAA business-associate problem, and the patient who will not repeat their date of birth a third time.
• AI answering for plumbers and home services — after-hours dispatch, the 2am burst-pipe misroute, and ServiceTitan-native options.

For the rest of the legal stack — research, drafting, contract review — see AI for lawyers .

Frequently asked questions

Частые вопросы