Industry Playbook

The Law Firm Intake System

How PI firms 3x intake-to-signed-client conversion using sub-1-minute voice agents, conflict-check SMS, and automated follow-up sequences for cold leads.

26 min read Updated 2026 Q2 Jump to sections

01 / The Intake Gap

The math behind why good cases walk.

The reality of intake in 2026 is that the firm with the better case docket is not always the firm with the better attorneys. It is the firm with the better intake operation. Three numbers from our deploy data tell the whole story.

48%

of PI calls go to voicemail.

Median across 22 firms we audited before deploying Nirvani. After-hours, lunch hour, court days, depo days. The phone rings and no one is there.

4h 17m

average response time on web leads.

A claimant who fills out your contact form at 9:42 a.m. typically hears back at 2:00 p.m. By then they have already called two competitors and one of them answered.

70%

of qualified leads never sign.

Of leads firms classify as “good potential cases” on first contact, seven in ten never come back to sign a retainer. Most simply never get re-contacted past day three.

02 / The Outcome

What the stack actually changes.

Numbers below are the median (not the average, not the cherry-picked top) across the 22 law firms in the Nirvani deploy data set. The mix is roughly 11 PI shops, 5 family practices, 4 criminal defense practices, and 2 immigration firms. Your individual results will vary by case mix, geography, marketing spend, and ethics rules in your jurisdiction.

Median results across 22 deployed law firms

47s
Answer time on inbound calls
Down from 4h 17m
3.1x
Intake to signed-client conversion
Lift on qualified leads
+18%
Average case value
From higher-quality screen
90 days
Payback period on full stack
N1 deployment, median firm

03 / The Stack

Five modules. One intake operation.

The system is not one feature. It is five modules wired into your existing practice management software. Categories of PMS that integrate cleanly: cloud-based intake-friendly systems (Clio Grow / Clio Manage), all-in-one firm platforms (Smokeball, MyCase, PracticePanther), and the case-management layer of any firm running a CRM separate from time and billing. We do not endorse any single vendor. We integrate with the category.

Each module ships with a sample script, a workflow diagram, and the SMS templates a firm can adapt to its own voice. Everything stays inside the firm’s data boundary. The voice agent does not give legal advice, and the SMS system is A2P 10DLC registered with explicit consent capture and STOP / HELP keyword support out of the box.

01
Module 01 / Inbound Capture

24/7 Case Intake Voice Agent

Every PI firm in our deploy set was losing the same case at the same hour: the call that came in between 6 p.m. and 9 a.m., or while the front desk was on lunch, or while reception was on the other line walking a slip-and-fall client through medical history. The voice agent answers in under one second. It introduces itself by firm name. It captures the same intake form a paralegal would, in conversational language, and it never tries to practice law.

Answers in <1s 47 second median call SMS confirm sent Routes to attorney on-call

What the agent actually does

  • Greets by firm name and case type. If the inbound number is the PI line, it greets “Thanks for calling [Firm], personal injury intake.”
  • Captures eight core data points. Caller name, callback number, date of incident, location, injuries claimed, currently treating with a physician, currently represented by counsel, source of referral.
  • Screens for urgency. If the caller mentions an incident date and the matter type has a known statute of limitations window approaching, the agent flags it inside the system (see Module 03).
  • Schedules the consult. Either books a next-day attorney call onto the firm’s shared calendar, or warm-transfers if the on-call attorney is available now.
  • Sends an immediate SMS confirmation. Caller gets a text with the consult time, what to bring, and a single “Reply STOP to opt out” line at the bottom.
  • Notifies the intake paralegal. Email + SMS summary of the call goes to the intake desk inside 30 seconds of hangup.

Sample script: after-hours PI call

Inbound call. 7:42 p.m. Tuesday.
AGENT: Thanks for calling Hartwell & Boren, personal injury intake. My name is Maya. I’m the after-hours intake assistant. I can take down what happened and schedule you with an attorney first thing tomorrow. Are you okay to talk for about three minutes?
CALLER: Yeah, I got rear-ended on the 60 yesterday and my neck is killing me.
AGENT: I’m sorry to hear that. Let me get a few things down so the attorney has the full picture. Can I get your full name?
CALLER: Marcus Webb.
AGENT: Thanks, Marcus. Best callback number? And are you currently being seen by a doctor for the neck pain?
CALLER: Same as my caller ID. I went to urgent care last night. They told me to see a chiropractor.
AGENT: Got it. One more important question. Have you signed with any other firm or talked to an insurance adjuster about this yet?
CALLER: No, you guys are the first call.
AGENT: Perfect. Last thing, do you have a copy of the police report? Whether yes or no is fine, the attorney will walk you through next steps. I have an opening with Adam tomorrow at 9:15 a.m. Want me to lock that in and send you a text confirmation?
CALLER: Yeah, that works.
AGENT: Done. You’ll get a text in about ten seconds with the time and what to have ready. Adam will call you tomorrow at 9:15. (Quick reminder: I’m not able to give legal advice, but if anything urgent comes up tonight, the firm’s emergency line is in that confirmation text.) Take care of that neck, Marcus.
Quick stat

The agent qualifies and schedules in 47 seconds. The same call going to voicemail loses roughly $8,400 of expected case value, before nurture recovery.

What the AI does not do

UPL Boundary
  • It does not give legal advice. If asked “do I have a case,” it responds: “I’m not able to give legal advice, but I can connect you with the attorney handling intake today.”
  • It does not quote contingency percentages. Fee discussions happen with the attorney.
  • It does not represent itself as an attorney. The opening line explicitly says “after-hours intake assistant.”
  • It does not promise outcomes. No language like “you’ll definitely recover” or “your case is strong” ever leaves the script.
  • It does not collect privileged content unnecessarily. Intake captures only what is needed to route to the right attorney. Detailed facts wait for the consultation.
  • It does not text without consent. The SMS confirmation is sent because the caller affirmatively scheduled. Marketing texts require a separate opt-in.

Where it integrates

The agent writes a contact and matter record into the firm’s practice management system (Clio, Smokeball, MyCase, PracticePanther, or the system you already run) the moment the call ends. It books the consult slot into the attorney’s calendar (Outlook, Google Workspace, or the PMS-native calendar). It triggers Module 02 (conflict check) and Module 03 (statute of limitations check) automatically. No paralegal touches anything until the morning standup.

02
Module 02 / Ethics-Layer

Auto Conflict-of-Interest Screening

The intake call ends. Forty seconds later, every name captured (caller, adverse driver, adverse business, prior counsel, treating physician where relevant) is matched against the firm’s existing client database, including closed matters, conflicts-only entries, and the firm’s ethical-wall list. If the system finds a hit, it pauses the consult booking and pings the firm’s designated conflicts attorney by SMS.

Matches all parties Includes closed matters SMS conflict alert Audit trail logged

The conflict-check flow

  1. Extract parties. The intake agent’s transcript is parsed for every named entity: caller, adverse parties, witnesses, prior counsel, treating providers, employers.
  2. Normalize. Names are normalized (e.g. “Bob” → “Robert,” “USAA” → “United Services Automobile Association”).
  3. Match. A query runs against the firm’s PMS contact and matter records, plus any standalone conflicts-only database the firm maintains.
  4. Decision tree. Clean → consult auto-confirms. Possible match → consult is marked “Conflict Hold,” SMS goes to conflicts attorney with the matched record and the new lead context.
  5. Audit. Every match attempt is logged with timestamp, parties checked, result, and the human who cleared it. This is the record you keep if a bar complaint ever asks how you screened.

Sample SMS to conflicts attorney

SMS dispatched 47 seconds after call end
[Hartwell & Boren Conflicts] Possible match on new intake. Marcus Webb (claimant), adverse driver Linda Carrasco. We have a closed 2024 matter (file 24-0381) where Linda Carrasco was a co-defendant. Hold consult tomorrow 9:15 a.m. Confirm clear or escalate. Reply CLEAR, HOLD, or DETAILS.

Why this matters more than firms realize

Most firms run conflicts checks the morning of the consult, by hand, against a static spreadsheet, in a 90-second flip-through. That works at 5 cases a week. It does not work at 25 cases a week. The miss rate goes up, the IOLTA-trust risk goes up, and the time to clear a clean case goes up too. Automating the screen lets the firm run a fuller, more thorough check on every single lead without adding paralegal headcount.

From a deployed firm

“We caught two conflicts in the first 30 days that our spreadsheet would have missed because the adverse party was a corporate parent we’d never written down by name.”

What this is NOT
  • Not a substitute for attorney judgment. The conflicts attorney still reviews any match and makes the call. The system is a flag, not a clearance.
  • Not a public-records search. The match runs against the firm’s own data only.
  • Not a replacement for new-client engagement letters. Standard intake paperwork still happens after the consult.
03
Module 03 / Urgency Layer

Statute-of-Limitations Watcher

The single most expensive intake mistake a firm can make is to let a viable case sit in the nurture queue past its statute of limitations window. Once SOL runs, the case is dead. No nurture sequence recovers a time-barred claim. The SOL watcher prevents this in three ways: at intake, in nurture, and at handoff.

Date-of-incident capture Per-jurisdiction SOL table Auto-flag at 60 days out Hard escalation at 30 days

How the watcher works

  1. Date-of-incident at intake. The voice agent captures the date as a required field. If the caller is unsure, the agent asks for an approximate week or month.
  2. SOL lookup. The matter type and jurisdiction are matched against a per-state SOL table the firm configures during onboarding (e.g. Arizona PI = 2 years from date of incident; California PI = 2 years; medical malpractice and government-entity claims have shorter windows).
  3. Urgency tag. Three tiers. Green = more than 6 months remaining. Amber = 60 to 180 days remaining. Red = under 60 days.
  4. Amber alert. Lead is flagged on the intake dashboard. Nurture sequence is escalated. SMS reminders to follow up are sent to the assigned attorney every 14 days.
  5. Red alert. Lead jumps the queue. Managing attorney is paged by SMS. The case must be signed, declined, or referred out inside 7 days.

Where SOL trips up firms

  • Government-entity notice of claim. Many jurisdictions require formal notice within 180 days (or shorter) for claims against a public entity. The agent flags any mention of a city / county / state employee, public vehicle, or government property at the scene.
  • Discovery-rule cases. Medical malpractice, latent-injury, defective-product matters where the clock starts on discovery, not on the act. These flag for attorney review, not auto-classification.
  • Minor tolling. Cases involving claimants under 18 typically toll until majority, but the firm still wants the date captured.
  • Out-of-state defendants and venue questions. If the captured details suggest a multi-jurisdiction case, the system tags it for venue analysis at the consult.

Sample red-flag SMS to the intake attorney

SMS, red SOL flag
[Hartwell & Boren SOL] RED FLAG. New PI intake, Marcus Webb. Date of incident 2024-06-04. Arizona PI 2-yr SOL runs 2026-06-04. 38 days remaining. Consult scheduled 2026-05-30 9:15 a.m. Confirm coverage or reassign.
Why we built it this way

The firm is responsible for the SOL table. We never ship default SOL values from cloud configuration. Each firm enters and verifies its own per-jurisdiction matrix during onboarding.

04
Module 04 / Nurture

30-Touch Nurture for Cold Leads

Of every 100 qualified intakes a PI firm receives, roughly 30 sign on the first consult. The other 70 do not, and the industry default is to abandon them. That is the seventy percent the data calls out. A 30-touch sequence across SMS, voicemail-drop, email, and occasional human follow-up recovers between 22 and 31 of those abandoned leads inside the first 120 days, in our deploy set. That is the deltas that turn N1 from a tool into a revenue line.

Multi-channel cadence A2P 10DLC compliant STOP / HELP enforced 120 day window

The 30-touch cadence

The cadence is calibrated to PI specifically. Family law and immigration use modified versions. Numbers below are touch-counts across the four-month window.

  • Days 0 to 7 (8 touches). Immediate consult confirmation, day-1 SMS check-in, day-3 educational link (“What to expect after a car crash, the first 14 days”), day-5 voicemail drop from the attorney by name, day-7 second consult offer.
  • Days 8 to 30 (10 touches). Weekly SMS with a single helpful resource (treating physician network, what to do when adjuster calls, lien basics, MMI primer). One human callback attempt around day 14. One email with the firm’s prior recoveries (anonymized, jurisdiction-appropriate, no improper solicitation language).
  • Days 31 to 60 (7 touches). Cadence slows. Bi-weekly SMS. One email on settlement vs. trial decision-making. One personal text from the attorney handling intake.
  • Days 61 to 120 (5 touches). Long-tail. Monthly SMS. Final “you have 60 days left on your statute” reminder if applicable, manually triggered by the attorney.

What gets messaged matters as much as cadence

The content rule we apply: every message either helps the claimant (without giving legal advice) or invites them to a no-pressure consult. We do not run “limited time” gimmicks. We do not run scarcity. The firm’s brand is its license. Pressure tactics are the fastest path to a bar complaint, especially in states with strict client solicitation rules.

Sample SMS at day 7

SMS, day 7 in the sequence
Marcus, this is Adam at Hartwell & Boren. Just checking in. The treating physician network we use takes care of the billing up front so you don’t pay anything out of pocket. If the neck pain is still bothering you, want me to make an intro? Text back YES or NO. Reply STOP to opt out. Reply HELP for help.

Compliance baseline (A2P 10DLC)

  • Consent on record. Every recipient affirmatively opted in at intake or web form. Consent timestamp, source, and IP are stored.
  • Brand registered. The firm registers its brand with the campaign registry and its messaging use case (Customer Care, Account Notifications, plus 2FA where applicable).
  • STOP keyword. Replies of STOP, END, CANCEL, UNSUBSCRIBE, QUIT immediately suppress all messaging.
  • HELP keyword. Replies of HELP trigger an auto-response with firm name, contact number, and a link to the SMS terms.
  • Quiet hours. No sends before 8 a.m. or after 9 p.m. local-to-recipient.
  • No third-party sharing. Phone numbers are not shared with third parties for marketing purposes. Period.
  • Suppression list inheritance. If a former client opted out historically, they remain on the suppression list for any new campaign.
What we never do in nurture
  • No outcome promises. “You will win” never appears.
  • No fee quotes. Contingency is discussed with the attorney, in writing.
  • No comparative attacks. We do not denigrate other firms by name.
  • No artificial scarcity. “Only 3 slots left” is not used. The firm’s calendar is the firm’s calendar.
  • No re-targeting after STOP. Opt-out is permanent.
Recovery rate, our deploy set

The 30-touch sequence recovers between 22 and 31 leads per 100 that would otherwise have gone cold. The median PI firm in our set saw $11,800 in additional signed-case value per recovered lead.

05
Module 05 / Referral Loop

Past-Client Referral Engine

The cheapest signed case is the referral from a satisfied past client. Almost no firm has a working system for this. They mean to send the “hope you’re doing well, who do you know” note. It never goes out. The referral engine automates it, on a cadence calibrated to the case lifecycle: 30 days after settlement, 90 days after settlement, 6 months later, and annually thereafter. Each touch includes one specific ask (review, referral, or no-ask check-in) and at most one polite reminder.

Post-settlement triggers Review request flow Referral attribution Anniversary touches

The post-resolution sequence

  • Day +30 from settlement / disposition. A personal SMS from the assigned attorney. “How are you doing one month out?” No ask.
  • Day +60. If positive sentiment was registered, a review request goes out. We route the request through Nirvani’s reputation flow: green-rated clients get the public-review link, red-rated clients get a private feedback form.
  • Day +90. First referral ask. “If anyone you know ever needs help after a wreck, here’s a card you can pass along.” Includes a unique referral link the firm tracks.
  • Day +180. Holiday or seasonally-relevant check-in. Still no ask.
  • Anniversary year 1, 2, 3. Memory-jog message. “One year ago we helped you close out the case. Hope this finds you well.”

Referral attribution

Every former client gets a unique referral code or short URL. When a new intake comes in through the voice agent and the caller mentions a referral source by name, the agent writes that back to the original case’s referral count. Firms learn which past clients are sending the most cases. We do not pay referral fees to non-lawyers (that runs afoul of most rules of professional conduct), but we do help firms create ethics-compliant recognition: handwritten thank-you, charitable donation in the referrer’s name, or simply a personal call from the partner.

Sample message: day +90

SMS, day +90 from settlement
Marcus, Adam here at Hartwell & Boren. Hope you’re feeling better than you were a year ago. No reason for this text, just wanted to say if anyone you know ever ends up in a wreck, this is my direct line. We’ll take care of them the same way we took care of you. Reply STOP to opt out.
Why this module pays for the stack

In our deploy set, firms that turned this module on saw their organic referral volume rise 38% in the first 12 months. Referrals close at roughly 2.4x the rate of cold leads.

04 / Implementation

The 30 / 60 / 90 day deploy roadmap.

The full stack takes 90 days to land cleanly. We do not recommend trying to ship everything in week one. Sequence is intentional. Voice agent first, because it generates the data the other modules feed on. Conflict and SOL layers next, because they are the ethics floor. Nurture and referral last, because they require enough data volume to tune well.

Days 1 to 30

Foundation

  • Voice agent live. Inbound number ports or twins. After-hours coverage active by end of week 2.
  • PMS integration. Contact + matter sync to Clio / Smokeball / MyCase / PracticePanther category.
  • Calendar wiring. Attorney consults book directly into shared calendars.
  • A2P 10DLC registration. Brand registered, campaign approved, STOP / HELP wired.
  • Intake script approved. Firm’s ethics counsel reviews and signs off.
  • SOL table loaded. Firm enters per-jurisdiction matrix, verifies twice.
Days 31 to 60

Ethics Layer

  • Conflict screen live. Matches running against full PMS contact database including closed matters.
  • SOL watcher live. Amber and red flags routing to assigned attorneys.
  • Reporting baseline. Weekly dashboard: intakes captured, conflicts flagged, SOL flags raised, consults booked.
  • Nurture sequence v1. Days 0 to 30 cadence turned on. Content reviewed by firm.
  • Recorded-call review. Firm spot-checks 10 calls weekly. Script tuned.
Days 61 to 90

Compounding Layer

  • Nurture v2 live. Full 30-touch sequence active for cold leads.
  • Referral engine live. Post-resolution sequence triggering for resolved matters.
  • Review flow live. Green-routed public, red-routed private feedback.
  • Performance review. 90-day data pull. Conversion rate, average case value, response time, signed-case delta.
  • Optimization round 1. Script tuning, cadence tuning, SOL threshold review.

05 / The Math

ROI snapshot: the deltas that matter.

Three numbers tell most of the story for a typical four-attorney PI shop with $2.4M in annual fee revenue. These are deploy-set medians, not single-firm bests. They assume the firm runs the full stack and tunes the nurture sequence by the end of month two.

What the numbers move.

Median four-attorney PI shop, $2.4M annual fee revenue baseline, 90 days post-deploy.

+9.2

Signed cases per month

Net new retained matters per month, blended across new intake and recovered nurture leads.

+18%

Average case value

Higher-quality intake screen produces a better case mix. Fewer minor-impact, fewer hard-to-prove damages.

14 hrs

Attorney time saved / week

Per attorney. Less intake call drift, fewer triage meetings, faster conflict clearance.

Stack runs on Nirvani N1 at $444 / mo per firm seat baseline. Firms with 6+ attorneys or higher volume typically upgrade to N2 at $1,500 / mo for the multi-line, multi-location configuration. Firms running multi-state or multi-practice deployments size to N3 at $4,000 / mo.

06 / Case Study

What a real deploy looks like.

Composite case based on three Phoenix-metro PI firms in our deploy set. Names changed. Numbers anchored to the median of the three. The firm is intentionally typical, not cherry-picked.

Composite case study

Hartwell & Boren, Phoenix, AZ

4 attorneys / ~$2.4M annual fees / 60% auto, 25% premises, 15% other PI / deployed Q1 2026

Hartwell & Boren came to Nirvani after a brutal quarter. The managing partner had run a referral audit and discovered that 38% of the firm’s after-hours leads were going to voicemail and never being returned. Two of those leads had subsequently signed with a competitor across town for cases the firm estimated at six figures of contingency value. The decision to deploy was made the same week.

What we shipped

Voice agent live day 9. PMS integration to their Smokeball-category practice management stack live day 16. A2P brand registration cleared day 22. Conflict-check and SOL watcher live day 41. Full 30-touch nurture sequence live day 67. Post-resolution referral engine live day 78.

What changed

By day 90, after-hours capture went from 22% of inbound calls returned, to 96% of inbound calls answered live by the agent. Median time-to-consult dropped from 3 days to 14 hours. The conflicts attorney caught two true matches inside the first month that would have slipped through the manual spreadsheet check. Most importantly, the firm’s signed-case count for Q2 ran 31% ahead of the comparable Q2 in the prior year.

What the partner said

“The voice agent paid for itself in 11 days. The nurture sequence is doing something I could never get a paralegal to do consistently, which is just be patient with the 70% who don’t sign on the first call. We’re seeing leads come back in month two and three that we would have written off a year ago.”

96%
Inbound calls answered live
14 hr
Median time to consult
+31%
Signed cases vs. prior Q2

07 / Companion Calculators

Run the numbers for your firm.

Three calculators in the Operator Toolkit pair directly with this playbook. Each takes under 60 seconds and outputs a number specific to your inputs.

Implementation Pack

Want the full deploy bundle?

We’ll send you the full intake script (PI version), the 30-touch nurture sequence text-for-text, the conflict-check workflow diagram, a sample SOL table, and a 30-minute call with the founder if you want one. No fluff, no signup gate beyond what’s here.

It’s on the way.

Check your inbox in the next 2 minutes for the implementation pack. You’ll get one confirmation text. Sammy will personally reach out within 24 hours if you flagged the strategy call.

See what N1 includes →

Compliance FAQ

The seven questions ethics counsel always asks first. Short answers below. The implementation pack goes into each one with citations.

Does the voice agent cross the UPL line?
Short answer: no, by design. Unauthorized practice of law is generally defined as giving legal advice or representing a client without a license. The agent does neither. It performs an intake function: identification, basic facts, conflict screen inputs, calendar scheduling. If a caller asks “do I have a case,” the scripted response is, “I’m not able to give legal advice, but I can connect you with the attorney handling intake today.”

The agent is also explicitly identified as a non-attorney assistant. The opening line uses “intake assistant,” never “lawyer” or “attorney.” We script in a closing reminder for any caller who has lingered on a borderline-advice question.

That said, UPL definitions vary by jurisdiction. Run the script by your state bar’s rules of professional conduct or your ethics counsel before launch.

How is attorney-client privilege protected during intake?
Two layers. First, the agent collects only what is needed to triage, route, and book. Detailed facts wait for the consultation with the attorney. Second, intake transcripts are stored inside the firm’s own data boundary inside the practice management system, behind the same access controls that protect any client communication.

A prospective-client privilege typically attaches the moment a person reaches out seeking legal services in good faith, even before retention. The intake transcript is treated as protected from the moment the call connects. Internal access is logged. We do not use intake content for model training. We do not share it across firms.

What about IOLTA and trust account interactions?
The stack does not touch IOLTA or operating accounts. Payment workflows for retainer deposits remain on whatever your firm uses today (LawPay, Clio Payments, PracticePanther Payments, etc.). The intake agent does not collect payment information.

The conflict and SOL layers are purely informational and do not interact with funds. Where the system pays for itself is on the revenue side: more signed cases, higher average value, faster intake-to-retainer cycle. The financial mechanics of the firm stay untouched.

Does this conflict with ABA Model Rules 7.1, 7.2, 7.3 on solicitation and advertising?
The system is built to stay inside the lines of Model Rules 7.1 (false or misleading communications), 7.2 (advertising), and 7.3 (solicitation of clients).

Inbound calls and inbound web inquiries are not solicitation under any reading we are aware of. The claimant initiated contact. Nurture messaging is sent only to contacts who affirmatively consented at intake. We do not send unsolicited SMS to cold lists. Past-client outreach is permitted under most rules as long as it stays truthful, non-misleading, and respects opt-outs. We script accordingly.

State variations matter. Some states (Florida and Texas, for example) have stricter rules. Your ethics counsel should review the configured templates before they go live.

Do we need bar association approval?
In most states, no. The system is an internal operations tool for the firm. Advertising-content rules typically apply to the firm’s own marketing materials, not to the technology stack the firm uses to manage intake and follow-up.

However, several states (including New York and California in certain practice areas) require filing of certain advertising content. If the nurture sequence is going to include any content that could be characterized as advertising past results, your firm’s standard advertising compliance process should apply. We do not push past-results content by default.

A handful of states (notably Florida) have specific rules about lawyer referral services and online lead-generation that can implicate certain deployment patterns. Have your ethics counsel review.

How does the agent handle the disclosure question (“am I talking to a robot?”)?
Honestly and immediately. If a caller asks any version of “is this a real person,” the scripted response is, “I’m an AI assistant the firm uses for after-hours intake. I can take your information and schedule you with an attorney, and I can transfer you to a person on call now if you’d prefer.”

We never script denial of the AI nature of the agent. We also never script impersonation of an attorney or named staff member. The agent uses a clearly distinct first name (e.g., Maya) and identifies as the firm’s intake assistant on the opening line. Several states have explicit AI-disclosure requirements that we either meet or exceed by default.

What happens if a caller mentions a crime in progress or an emergency?
The agent has a hard-coded emergency branch. If the caller mentions an active threat to themselves or another person, an injury requiring immediate care, or any version of “this is happening right now,” the agent responds:

“If you’re in danger or someone is hurt, please hang up and call 911 right now. I can help with the legal side once you’re safe. The firm’s emergency line is also in the text I’m about to send you.”

The agent then sends the SMS, logs the call with a high-priority flag, and notifies the on-call attorney immediately. We do not let an AI play crisis counselor.