Guides
AI Legal Intake Compliance: Guardrails for Inbound Law Firm Calls
A practical rule boundary for first-party AI legal intake, with supervised fact collection, prospective-client safeguards, truthful scripts, and human legal judgment.
Guides
A practical rule boundary for first-party AI legal intake, with supervised fact collection, prospective-client safeguards, truthful scripts, and human legal judgment.
Last updated
AI legal intakeLegal intakeLegal intake is the process of capturing a potential client’s case details, urgency, location, eligibility, and preferred next step before routing them to the right legal team. compliance starts with a simple boundary: let the agent receive and structure a first-party inquiry, but do not let it practice law. A sound workflowWorkflowAn automated, multi-step process — usually triggered by an event (form fill, new lead) and orchestrating one or more voice / SMS / email actions. collects only approved intake facts, checks deterministic routing rulesRouting rulesLogic that determines where a lead, call, appointment, or task should go based on criteria such as location, product, urgency, language, licensing, or availability., schedules or transfers the caller, and leaves a reviewable record. Legal advice, case-value predictions, conflict decisions, fee negotiation, engagement terms, deadline calculations, and representation decisions stay with licensed lawyers.
Thoughtly's position is deliberately narrow. Use AI to cover demand the firm already earned, not to cold-call people inferred to need a lawyer. The platform should augment the intake team by handling response, structured fact collection, booking, and contextual handoff. It should not impersonate counsel or turn a qualification script into a legal opinion.
That boundary is practical, not theoretical. ABA Formal Opinion 506 says a trained and supervised nonlawyer may perform several prospective-client intake tasks, while questions about legal services, fee negotiation, and interpretation of an engagement agreement require a lawyer. AI does not get a broader job description because it speaks fluently.
Thoughtly's legal-intake workflow guide owns the operational path from an inquiry to a consultation, while the legal-intake buyer guide owns platform selection. This article owns a different question: what may the automated intake layer do, what must it refuse, and what evidence should the firm retain?
The distinction matters because legal intake sits beside professional-responsibility rules, prospective-client duties, advertising and solicitation rules, call-consent rules, privacy obligations, and the firm's own conflict process. A generic voice-agent disclaimer does not reconcile those systems. The workflow has to express the boundary in every branch.
The useful point of view is simple: faster response is valuable only when the firm can explain why the call happened, what the agent said, what it collected, where it stopped, and who took responsibility next. A booked consultation is not a successful outcome if the path to it created an ethics problem.
The ABA Model Rules of Professional Conduct serve as models for most jurisdictions. They are not a single nationwide code. The ABA also publishes jurisdictional comparison charts showing how local rules modify the model language.
Use the model rules to design the first control draft, then have qualified counsel map the actual states, practice areas, lead sources, call directions, scripts, and data flows. The tempting shortcut is to label a workflow compliant because it resembles the model rule. That label is not a substitute for jurisdiction-specific review.
| Intake moment | AI may handle | Human must own | Required control |
|---|---|---|---|
| New first-party inbound call | Identify the firm and automated assistant; collect contact details, general matter type, jurisdiction, and preferred next step | Legal assessment, advice, representation decision, and sensitive exceptions | Approved opener, minimum-data fields, advice stop, human-request transfer |
| Callback after the firm receives a form | Refer to the submitted request; confirm identity and interest; schedule or transfer | Consent interpretation, disputed source, channel permission, and cadence exceptions | Stored source, timestamp, form language, channel permission, and suppression check |
| Initial conflict intake | Collect only approved party names and send them to the firm process | Conflict determination, waiver analysis, screening, and acceptance | No merits discussion before the approved conflict path |
| Fee or process question | Read approved general process information verbatim | Fee negotiation, scope explanation, and engagement interpretation | Deterministic fee and scope escalation |
| Case merits, value, or deadline question | Capture the question and the caller's stated facts without evaluating them | Legal advice, case valuation, deadline calculation, and strategy | No-answer response plus priority lawyer handoff |
| Decline or referral | Deliver lawyer-approved neutral wording and next-step resources | Reason for decline, legal explanation, and referral policy | Approved disposition list, no invented explanation, auditable owner |
| Urgent or vulnerable caller | Recognize approved risk language and create a priority transfer or callback | Substantive advice and emergency judgment | Clear emergency language, immediate human route, no promise of representation |
ABA Model Rule 7.3 defines solicitation around a communication initiated by or for a lawyer and directed to a specific person known or reasonably believed to need legal services in a particular matter. Its commentary distinguishes a response to a request for information from prohibited live person-to-person solicitation under the model rule.
That is why the first control is provenance. An inbound legal-intake workflow should begin with a direct call to the firm, a website form submitted to that firm, a requested callback, or another source counsel has approved. A purchased list, scraped accident record, shared marketplace lead, or unexplained CRMCRMThe system of record for leads, contacts, deals, and activity. Thoughtly reads from and writes to your CRM continuously. import is not magically inbound because it arrived in the CRM.
Store the source URL or campaign, timestamp, exact request or consent language, seller identity, permitted channel, and any expiration rule before the first automated callback. If the record cannot answer who asked whom to communicate, the workflow should stop for review rather than improvise permission.
Professional-responsibility rules are only one layer. The FCC treats AI-generated outbound voices as artificial or prerecorded voices under the TCPATCPAUS federal law governing telemarketing calls and SMS. Thoughtly enforces consent capture, time-of-day windows, and DNC scrubbing automatically., and the FTC explains that the Telemarketing Sales Rule treats unsolicited consumer-initiated calls differently from covered telemarketing. A form-fill callback is still initiated by the firm, so counsel should review the exact consent and campaign instead of assuming every warm lead is exempt.
The safest commercial posture is also the clearest one: Thoughtly is for rapid response to first-party demand and approved lifecycle events, not speculative cold outreach to people in distress.
Formal Opinion 506 provides a useful task boundary. Under the opinion, trained and supervised nonlawyers may obtain initial matter information, perform an initial conflict check, determine whether the assistance sought is germane to the practice, answer general questions about the fee agreement or representation process, and obtain a signature when the prospective client is offered an opportunity to communicate with the lawyer.
The same opinion says specific questions about what legal services the person should obtain, negotiation of fees or expenses, and interpretation of an engagement agreement require a lawyer. Model Rule 5.5 separately prohibits a lawyer from assisting another in the unauthorized practice of law.
Translate that line into behavior, not a broad prompt. The agent may ask what happened, where it happened, who was involved, which practice area appears relevant, and when the caller can speak. It should not tell the caller whether a claim is valid, what it is worth, whether a deadline has passed, what legal strategy to follow, or whether the firm represents them.
A disclaimer is useful, but it is not a control if the next node confidently answers the prohibited question. The better design uses a fixed response, captures the question, sets a structured escalationEscalationMoving a conversation to a human, specialist, supervisor, or alternate workflow when the agent detects risk, uncertainty, urgency, or a request it should not handle alone. reason, and transfers or schedules the right licensed person.
Model Rule 1.18 protects certain information learned from prospective clients even when no client-lawyer relationship follows. The rule commentary says a consultation can arise through oral or electronic communications when a firm invites information about potential representation, and recommends limiting the initial consultation to information reasonably necessary to decide whether to proceed.
The operating consequence is minimum-data intake. Before conflict and fit checks, prefer party names, contact details, general matter category, relevant location, the caller's own description of urgency, and a requested next step. Do not invite a full privileged narrative, medical record, financial account record, evidence upload, or strategy discussion simply because the model can summarize it.
Do not promise that an AI conversation creates attorney-client privilege or that a disclaimer defeats prospective-client duties. Privilege, confidentiality, and relationship formation are related but distinct legal questions that depend on facts and jurisdiction. Design for the stricter information-handling posture and let the firm's counsel decide the legal characterization.
Before sending prospective-client narratives into any platform, review the executed order, data-processing terms, retention settings, access controls, and downstream integrations. Thoughtly's public Terms of Service place responsibility for lawful customer data, third-party consent, notices, call recordingCall recordingCapturing audio from a phone conversation for review, QA, training, compliance, dispute resolution, or supervised retention., telemarketing, and privacy compliance on the customer. The right move is a documented firm decision, not an assumption based on a security badge.
Model Rule 7.1 prohibits false or misleading communications about a lawyer or legal services. Model Rule 7.3 also bars certain solicitation involving coercion, duress, harassment, or contact after a person has said they do not want solicitation.
The agent should never imply that a lawyer reviewed the matter, the firm accepted the case, a consultation guarantees representation, a result is likely, or immediate signature is necessary to preserve rights. It should not manufacture urgency or keep persuading a distressed caller after the person asks to stop.
Use approved factual language for the firm name, practice areas, office locations, scheduling process, and general fee model. When the answer depends on facts, law, lawyer judgment, or a missing source, the agent should say so plainly and route the question. Honest uncertainty is a trust feature.
AI identity and recording notices require their own review. Use the AI disclosure guide and call-recording consent guide as adjacent control layers, then have counsel approve the exact opener for each call direction and jurisdiction.
Model Rule 5.3 requires reasonable efforts by managers and supervisors to ensure nonlawyer conduct is compatible with professional obligations. The rule is a useful governance principle for an automated intake layer: somebody must own the instructions, review behavior, and take remedial action.
The escalation design should be deterministic for legal advice, fee negotiation, engagement scope, conflict concerns, deadline questions, complaints, emergency language, unsupported practice areas, consent disputes, and any request for a person. These are not edge cases to leave to model creativity. They are the workflow.
Thoughtly's current Agent Builder supports Start, Speak, Transfer, and End nodes, along with variables, actions, testing, and prompt-based or rule-based outcomes. The documentation specifically recommends rule-based outcomes for compliance or precision and fixed verbatim messages for must-say language.
Use a Transfer node or a clearly owned callback task, and pass the captured intake context so the caller does not have to repeat the entire story. The handoff succeeds only when a named person or queue accepts responsibility within a defined service level.
An audit trail is not a transcriptTranscriptThe text record of a voice conversation, used for review, training, compliance audit, and search. dump. It is a structured explanation of what happened: lead sourceLead sourceThe channel, campaign, marketplace, referral partner, or form that generated a lead. Lead source often determines routing, compliance rules, and follow-up cadence., timestamp, contact direction, approved script version, notices delivered, data fields captured, outcomes selected, advice-stop reason, transfer attempt, destination, final disposition, opt-outOpt-outA recipient’s request to stop receiving calls or messages. Compliant systems must capture opt-outs and suppress future outreach where required. state, and human owner.
Thoughtly Variables can capture information for branching and later CRM updatesCRM updatesCRM updates are the automatic changes an agent writes back after a conversation, such as status, notes, next step, owner, appointment, or disposition.. The On Call Completed triggerTriggerThe event or condition that starts an automated workflow, such as a new lead, missed call, CRM status change, calendar booking, or completed call. exposes outcomes, variables, transfers, and transcript entries for post-call workflows, while webhooks can push the approved subset into the firm's system of recordSystem of recordThe authoritative system where customer, lead, policy, loan, appointment, or account data is stored and updated..
Send only fields the receiving system needs. A full transcript may contain much more than intake operations require, and duplicating it across a CRM, spreadsheet, inbox, and analytics warehouse multiplies exposure. The firm should define retention and deletion by data class, not by whatever the default connector happens to keep.
Review the evidence by exception and by sample. Every advice stop, failed transfer, consent dispute, potential conflict, decline, complaint, and unsupported answer should be visible to a human. Random sampling should also inspect ordinary calls, because a workflow that only reviews failures can miss confident but wrong success labels.
List the exact inbound numbers, forms, campaigns, referral sources, CRM states, and requested-callback events that may start the workflow. Anything else routes to review. A clean source taxonomy is more valuable than a clever opening prompt.
Identify the firm and automated assistant, explain the limited intake purpose, give any approved notice, and offer a human path. Keep the wording short enough to understand under stress. Counsel should approve variants by call direction and jurisdiction.
For each field, record why it is needed, who may see it, where it is written, how long it is retained, and what happens when the caller declines. Do not use an open-ended request such as "tell me everything" before conflict and fit controls run.
Create explicit outcomes for advice, merits, value, deadline, scope, fee negotiation, conflict, emergency, complaint, consent dispute, opt-out, unsupported practice area, and human request. Each outcome should have one approved response and one owned next step.
Test a caller who demands legal advice, names a current client as the opposing party, asks whether the limitation period expired, tries to negotiate the fee, provides an unapproved document type, requests a person, withdraws channel permission, and switches jurisdiction mid-call. A demo that only books a consultation proves very little.
Assign owners for script changes, sample review, incident response, source suspension, and workflow shutdown. If the agent crosses the boundary, stop the affected path, preserve the evidence, correct downstream records, and let counsel determine notice or remediation.
Thoughtly should own fast, consistent coverage for first-party legal inquiries: identify the request, collect approved facts, apply the firm's routing rules, book or transfer, continue only permitted follow-up, and write structured outcomes back to the system of record.
Lawyers should own the practice of law and the decisions around it. That includes legal advice, case merit, strategy, conflicts, fee negotiation, scope, engagement, deadline interpretation, referral policy, and exceptions. The platform is strongest when it makes that division visible instead of pretending fluent conversation removes it.
Teams evaluating the broader workflow can use the legal-intake implementation guide for lifecycle design and the legal-intake platform comparison for selection criteria. The compliance layer should remain a separate approval gate across either path.
Measure response time, completed intake records, qualified consultations booked, transfer acceptance, consultation completion, and signed matters. Those are the revenue outcomes. Pair them with boundary-violation rate, unsupported-statement rate, conflict-path completion, failed-transfer recovery, opt-out propagation, duplicate-contact rate, and human review time.
Do not optimize the agent solely for more completed scripts. A shorter conversation that recognizes a legal question and gets a lawyer involved is often better than a longer one that sounds helpful while making the wrong judgment. The best pilot proves both throughput and restraint.
A supervised agent can perform a narrow set of intake tasks, but the allowed line depends on jurisdiction and task. ABA Formal Opinion 506 supports delegation of initial information gathering and other defined intake work to trained, supervised nonlawyers while reserving specific legal-service, fee-negotiation, and engagement-interpretation questions for lawyers. Treat that as a design baseline and have local counsel approve the actual workflow.
Do not let the agent answer that categorically. Model Rule 1.18 and its commentary show that duties to a prospective client can arise before representation and depend on the circumstances of the consultation. Use clear wording, collect only necessary information, avoid promises of representation, and route relationship questions to a lawyer.
It can apply lawyer-approved routing rules to stated facts, but it should not render a legal merits opinion, predict value, calculate a deadline, or make the final representation decision. Use categories such as likely practice-area fit, outside approved scope, and lawyer review required, then keep acceptance with licensed counsel.
Yes, if the firm has approved the source, request language, channel permission, jurisdiction, and cadenceCadenceA cadence is the planned sequence of calls, texts, emails, delays, and retries used to follow up with a lead until they answer, opt out, book, or become inactive.. Do not assume that a form submission erases TCPA, telemarketing, state bar, recording, privacy, or opt-out obligations. The TCPA guide covers the separate calling-law layer.
Use a short, counsel-approved opener that identifies the firm and automated assistant, states the limited intake purpose, gives required notices, avoids implying representation, and offers a human option. The exact language should vary only through approved jurisdiction and call-direction rules, not free-form improvisation.
Collect the minimum necessary before conflict and fit checks, limit who can access the data, review vendor and integration terms, define retention by data class, and keep sensitive documents out until the firm has approved the channel. Rule 1.18 is the baseline reminder that a person who never becomes a client may still create duties for the firm.
This article is operational guidance, not legal advice. The ABA Model Rules and opinions are persuasive planning sources, while binding duties come from the rules and law that apply to the firm, lawyer, jurisdiction, practice area, lead source, and communication. Review the complete workflow with qualified counsel before production and whenever those inputs change.