A law firm’s first phone call can contain a deadline, a conflict, a sensitive family detail, or a description of a dispute involving someone the firm already represents. A caller may be seeking representation, asking for an update, trying to reach counsel, or requesting a referral. An AI receptionist for law firms can help capture and route routine inquiries, but it should not turn a preliminary intake call into an unsupervised legal consultation.

This guide lays out a safer design for law firm phone answering and client intake. It covers prospective-client information, conflict-screening order, practice-area routing, boundaries on legal advice, confidentiality and vendor review, and how to measure whether automation helps. It focuses on US firms. ABA Model Rules and formal opinions provide guidance, but each jurisdiction adopts and interprets its own rules; firms should have ethics counsel review the workflow.

Practical starting point: use the assistant to identify the caller’s broad need, gather only the minimum information for a human review, and route the matter to the right intake team. Before collecting a detailed narrative, explain that the call does not itself create an attorney-client relationship and that the firm must check whether it can speak with the caller. Do not promise representation, legal outcomes, deadlines, or confidentiality beyond the firm’s approved script.

Define the receptionist’s role in the intake process

A law firm AI receptionist should handle office logistics and structured intake, not legal analysis. It may state approved hours and locations, offer a callback, collect a name and safe contact method, identify a broad practice area, and create a review task. It should not tell a caller whether they have a claim, calculate a filing deadline, assess case value, recommend a legal strategy, or decide that the firm will take the matter.

Keep the assistant’s permitted actions explicit. A short allowlist might include routing a current client to their known team, collecting a prospective caller’s name and opposing-party names for conflict review, recording a high-level matter category, and scheduling an initial consultation only after firm criteria are met. Prohibit free-form legal answers, substantive document intake, fee promises, and statements implying that a lawyer has reviewed the facts.

Use firm-approved language to distinguish a request from a completed engagement. A caller may interpret “we’ll take care of it” as acceptance. The assistant should say that it has recorded a request and that the firm will review it; any representation begins only through the firm’s normal engagement process. The precise wording should come from the firm’s lawyers and align with engagement letters and local practice.

Consider whether the agent needs to handle current clients at all. A current-client flow may expose privileged context or misroute an urgent court issue. Some firms may limit automation to new callers and office logistics, while others use it for overflow with authenticated routing. Decide per practice group, not by a universal setting.

Protect prospective-client information from the first questions

ABA Model Rule 1.18 addresses duties to prospective clients, including limits on using or revealing information learned from a prospective client even when no attorney-client relationship follows. The rule’s comments recognize that a person may reveal information while discussing possible representation. This is why intake design should minimize unnecessary detail, restrict access, and avoid collecting an entire case story before the firm has decided who should review it.

Begin with a clear notice in language the caller can understand: the automated assistant is gathering preliminary information; the firm has not agreed to represent the caller; the caller should avoid sharing detailed confidential facts until the firm completes its review; and the firm will explain the next step. Counsel should tailor the script to local rules and the firm’s actual intake procedures. A disclaimer cannot cure careless data handling, but clarity helps prevent misunderstandings.

Ask for the caller’s name, contact details, adverse parties, related entities, broad matter type, jurisdiction, and any date that makes the request time-sensitive. Do not invite a detailed narrative before conflict screening. If the caller starts sharing sensitive facts, interrupt gently and redirect to a high-level summary. Store the minimum needed for the conflict check in a restricted intake system, not a general-purpose shared inbox.

Set access by role. Intake coordinators may need contact details and conflict-screening fields; an entire firm does not need every declined matter’s narrative. Define how the firm handles data from callers it cannot represent, how long it is retained, how deletion requests are handled, and how access is reviewed. Apply the firm’s retention and records policies to audio and transcripts, not only to the final matter record.

Put conflict screening before detailed case intake

A conflict check is not a conversational flourish; it is a control point. The intake workflow should collect names needed for the firm’s conflict system early, then wait for a human or approved conflict process before inviting detail. Include opposing parties, relevant organizations, related individuals, and aliases when the firm’s policy requires them. The assistant should not decide that a name is clear based on a weak fuzzy match.

Create states such as preliminary information received, conflict review pending, cleared for fuller intake, conflict concern, and unable to proceed. The caller-facing message should match the real status. Do not schedule a substantive consultation or encourage disclosure while the matter is pending unless the firm’s policy explicitly allows it. If a possible conflict appears, route the record to an authorized reviewer and avoid revealing the reason to the caller.

Test common identity complications: a caller uses a shortened name, a company has a parent and subsidiary, a former spouse is also a party, a matter involves a government agency, or the adverse party is known by a trade name. A human should resolve ambiguous names. A system that says “no conflict found” without checking the firm’s actual database may create false confidence.

Keep an audit record of who reviewed the conflict, which names were searched, what decision was made, and when fuller intake became permitted. If the firm updates its conflict system or intake rules, update the voice flow and retest. The AI should never silently bypass the gate because a caller sounds urgent or a calendar has an open slot.

Route by practice area without giving legal advice

Routing can make the first interaction more useful when the firm serves multiple practice areas. Ask a short, neutral question about the type of help the caller is seeking, then send the inquiry to the relevant intake team. Avoid questions that lead the caller toward a legal conclusion or that require the assistant to determine eligibility. “Is this about a residential lease, a business contract, or something else?” is safer than “Were you illegally evicted?”

Allow the caller to choose “not sure” or “something else.” A routing model should not reject a caller because the phrasing does not match its taxonomy. If the matter is outside the firm’s scope, create a human review task or provide a firm-approved referral resource. Avoid asserting that another firm will accept the case or that a deadline is unimportant.

Capture venue and timing only to the extent the firm wants those details for triage. If a caller mentions a hearing, notice, response deadline, arrest, custody event, or imminent harm, use an approved urgent escalation path. The system should not calculate how many days remain or tell the caller that a limitation period has expired. Those judgments require jurisdiction-specific legal review.

Design separate paths for current clients, opposing counsel, courts, vendors, media, and prospective clients. Current-client calls should route to known staff with identity checks. Court or opposing-counsel calls may need immediate staff handling. A single “press one for intake” tree cannot safely serve every caller type.

Set boundaries on deadlines, emergencies, and promises

A legal deadline can turn a routine intake into a time-sensitive event. The assistant may ask whether a date exists and record the caller’s own words, but it should not calculate a deadline or assure the caller that there is time. Define a firm-approved trigger phrase for time-sensitive events and send it to a live intake lead with acknowledgement tracking. If no one accepts, the workflow needs a backup person or clear instruction.

Avoid promises such as “an attorney will call you today” unless a named team actually owns that service level. Use an expectation the firm can meet: the inquiry is recorded, the relevant team has been notified, and the firm will contact the caller if it can proceed. Do not imply that the firm is monitoring an emergency line when it is not. Update after-hours wording around court holidays, closures, and staffing changes.

If the caller indicates immediate danger or a medical emergency, the firm’s script should direct them to the appropriate emergency service according to counsel-approved policy. The receptionist should not attempt crisis counseling or evaluate danger. Give the staff member who receives the escalation enough information to act, while avoiding a broad distribution list that exposes the caller’s details unnecessarily.

Test contradictory and ambiguous calls. A person may say they have no deadline and then mention a court date. A caller may be unsure whether a notice is real. A speech model can miss “tomorrow” or confuse a year. Use confirmation: repeat dates exactly, ask whether the caller wants to correct them, and mark them as caller-reported. Never treat a transcription as verified legal fact.

Review the voice agent like a supervised nonlawyer workflow

ABA Model Rule 5.3 discusses lawyer responsibilities regarding nonlawyer assistance. ABA Formal Opinion 512 discusses lawyers’ duties when using generative AI, including competence, confidentiality, communication, and supervision. These sources do not mean every AI receptionist is automatically a legal assistant or that a single rule resolves every deployment. They do make it prudent for firms to define responsibilities, review outputs, understand limitations, and supervise the workflow.

Assign a responsible lawyer and an operational owner. The lawyer approves what the system may say and do; the operational owner maintains schedules, routing, vendor contacts, and review queues. Create a change log for prompts and knowledge sources. Require approval before changing legal intake language, deadline questions, conflict gates, or matter acceptance criteria. Keep vendor configuration access limited and reviewed.

Sample calls across each matter type and caller experience, including callers who are upset, have limited English, use an interpreter, or ask a complex question. Check whether the assistant identifies itself, avoids legal advice, records names accurately, preserves a caller’s correction, and routes urgent matters. Review transcripts as sensitive firm records and restrict the audience. Do not use raw client calls as general model-training material without an explicit, lawyer-reviewed basis.

Create an incident process for wrong-party disclosure, missed escalation, inaccurate promise, unauthorized recording, or a conflict-gate bypass. Decide who investigates, who contacts the caller, whether a lawyer must review the facts, how the system is paused, and what records are preserved. A vendor’s customer support ticket should not be the firm’s entire incident plan.

Evaluate confidentiality and vendor data terms

Map the full chain of vendors: telecom carrier, voice platform, transcription service, foundation-model provider, hosting and storage providers, analytics, CRM, calendar, support tooling, and any integration partner. Ask which vendors receive audio, transcript text, caller ID, matter names, and metadata. Ask where data is stored, how long it remains, whether it is used for training, and what happens when support staff troubleshoot an issue.

Review contract terms for confidentiality, permitted use, retention, access, subprocessors, breach notice, audit cooperation, data return, deletion, and termination. Confirm whether the firm’s engagement and privacy policies permit the call recording and processing proposed. If callers may share confidential or privileged information, counsel should assess whether the arrangement preserves the firm’s duties and whether the vendor’s terms are adequate.

Use the least data needed for each function. The scheduling tool may need a caller’s preferred time and contact details, but it may not need the legal narrative. The CRM may need a preliminary matter category but not a full transcript. Keep intake data in a system with role-based access, audit logging, and retention controls rather than duplicating it across email and spreadsheets.

Ask how the vendor handles model updates and changes in service providers. Require notice before material changes affect confidentiality or data processing. Make sure the firm can export its records in a usable format and revoke vendor access promptly. If the product cannot honor the firm’s necessary restrictions, narrow the use case or choose a different arrangement.

Write an intake script that is brief and operational

A strong script follows the caller’s need instead of reading a long questionnaire. Start with the firm name and automated-assistant disclosure. Ask whether the caller is a current client or seeking new representation. For a prospective caller, collect identity and adverse-party names before a narrative. Ask only the broad category and urgency signal required to route the request. Then state what the firm will do next.

For a current client, ask for a safe identifier and the existing team or matter reference. If identity cannot be established, take a callback request without disclosing case status. For a new caller, do not say the firm can help until conflicts and availability are reviewed. For a caller who wants a lawyer immediately, transfer if the policy allows; if not, explain the next step without suggesting that the firm has accepted the matter.

Use confirmation on every critical field: spell a name when needed, repeat the callback number, confirm adverse-party names, and read back any caller-reported date. Let the caller correct the information. Mark uncertainty explicitly rather than forcing a confident value into the record. A well-structured “unknown” field is better than an invented answer that looks complete.

Keep a human option visible. Callers may be distressed, have a disability, distrust automation, or need an interpreter. Offer a direct route to staff, not a maze of repeated prompts. If all staff are unavailable, record a concise request and explain when the firm expects to respond. Do not conceal the limits of the process.

Measure the intake pipeline, not just answered calls

Track the complete path from call attempt to human review: answer rate, completed intake, conflict-review time, qualified consultation scheduled, missed urgent escalations, callback completion, intake abandonment, duplicate records, and staff correction time. Break results out by practice group, daypart, language, and referral source. A high answer rate is not useful if callers leave before the conflict check or if the right team never sees the inquiry.

Use a small, structured call-review form. Score identity and party-name accuracy, clear non-engagement language, no legal advice, correct matter routing, appropriate urgency, caller understanding, and whether the next step occurred. Have a lawyer review high-risk categories and an intake lead review routine operations. Keep the scorecard focused so that review is practical and sustainable.

Compare the pilot to a baseline with the same definitions. If the firm previously counted every call as an intake but now counts only fully screened records, the metrics are not comparable. Document exclusions, holidays, marketing campaigns, and staffing changes. Use call logs and CRM events where possible rather than relying on a vendor’s headline dashboard alone.

Set stop conditions before launch: any conflict data exposed to an unauthorized user, any unacknowledged urgent deadline alert, repeated false representation promises, or material errors in party names. Decide who can switch the system off and how calls revert to the previous route. A reversible pilot is safer and easier to learn from than a firmwide cutover.

Plan a phased launch for a law office

Phase one can cover office hours, directions, and transfer requests, with no case narrative captured. Phase two can add basic prospective-client fields and conflict screening, while a human approves every record. Phase three may add approved consultation scheduling for cleared matters. Keep legal questions and case evaluation with lawyers. Each phase should have a written scope, an approver, and a rollback route.

Train lawyers and staff before the first live call. Explain what the assistant can and cannot do, where its summaries appear, how to correct a record, how to request a pause, and which alerts require immediate response. Staff should not assume the AI handled a caller simply because a notification was generated; define acknowledgement ownership and escalation if the alert is unread.

Run tabletop exercises for a current client with an urgent court deadline, a potential client naming a represented adverse party, a caller who begins sharing sensitive facts, a person seeking advice, a wrong number, a transfer that fails, and an outage. Observe whether the system follows the intended path and whether staff know how to respond. Fix the workflow before increasing coverage.

Review the arrangement after launch and after material changes. Ask whether callers understood the assistant, whether staff trusted the intake records, whether conflict screening happened early enough, and whether declined callers’ information was handled appropriately. Keep a documented decision about whether to continue, narrow, or expand. A firm should be able to explain why each automated step is appropriate.

Create a sample law firm intake decision tree

A practical first branch is whether the caller is an existing client, a prospective client, a court or opposing party, a vendor, or someone seeking general office information. Current clients move to an identity-aware team route. Court and opposing-counsel calls go to a staff path. General questions use approved office information. Prospective matters proceed to minimum party details and a conflict review before a narrative is collected.

The second branch records the broad practice area and jurisdiction. Give the caller an “other” choice and a way to speak with a person. The system should not reject an inquiry because it cannot classify the caller’s description. If a potential client names an adverse party, the assistant should collect the name accurately, confirm spelling when needed, and stop before asking for sensitive facts until the firm clears the conflict step.

The third branch checks urgency without interpreting the law. Ask whether there is a date on a court document, a scheduled hearing, an immediate safety issue, or another time-sensitive event. Record the date as caller-reported and route according to a lawyer-approved rule. Do not calculate a deadline, tell the caller which filing is required, or state that the firm has time to respond.

The final branch confirms the next step: a callback request, a conflict review, an appointment request, or an inability to assist. Read back contact details and say whether the request is confirmed or pending. Do not state that the firm has accepted a client or matter. A human intake coordinator should review the record and decide whether to proceed.

What good human handoff looks like in legal intake

A handoff should preserve the caller’s effort. The staff member needs the caller’s name and safe callback method, the adverse parties, broad matter category, jurisdiction, caller-reported dates, language preference, and what the caller wants next. The summary should distinguish facts from the system’s interpretation and identify fields the caller could not confirm. Do not forward a long transcript when a concise structured record is sufficient.

The caller should know what happens after transfer. If a staff member answers, the assistant should briefly introduce the context and then stop speaking. If no one answers, the call should not silently end or return to the main menu. Offer a callback request and explain when the firm is open. For urgent matters, follow the approved escalation path and verify acknowledgement.

Give intake staff a correction mechanism. They should be able to flag an incorrect name, misheard date, wrong matter type, or unsupported promise. Route the flag to the system owner and keep a record of the fix. If staff routinely need to rewrite the entire summary, revisit what the assistant asks and how it structures the handoff.

Measure the time from first call to reviewed intake and from conflict clearance to fuller consultation. This shows where the process actually slows down. An AI answer in seconds does not improve client service if the record sits unassigned for two days. Define a human response owner and monitor unclaimed tasks.

Treat intake prioritization as a lawyer-approved policy

A firm may need to prioritize calls by urgency, practice area, or existing-client status, but the criteria should be explicit and reviewed by lawyers. Do not let an opaque model decide that a caller is credible, likely to win, or financially attractive based on voice, accent, emotional tone, neighborhood, disability, or other sensitive proxies. If the firm uses a case-fit rubric, staff should understand and own it.

Keep caller-reported dates and events separate from the system’s priority label. A person who says a hearing is tomorrow may be urgent even if the model fails to recognize a common court term. A low-confidence classifier should route to a human, not downgrade the call. Use clear operational triggers and allow intake staff to override them with a recorded reason.

Audit whether comparable callers receive comparable review and response. Look at routing outcomes by practice area, language, channel, and location, and ask whether a rule has an unintended effect. The goal is not to automate legal judgment but to make sure administrative triage does not block access to the people who can evaluate the matter.

The firm should be able to explain its intake logic to a caller and to its own staff. Document why a field is collected, who sees it, and what decisions it can influence. Remove questions that do not help the firm take a safe and timely next step.

Frequently asked questions

Can an AI receptionist give legal advice?

A firm should configure a receptionist for office information, structured intake, and approved routing. Legal advice, case evaluation, and decisions about representation should remain with lawyers under the firm’s procedures.

Should an AI receptionist collect a full case story?

Usually not before the firm’s conflict process and review. Collect only the minimum needed to identify parties, understand the broad matter type, and route the caller. Restrict access to any prospective-client information.

Can a law firm use AI to schedule consultations?

It can schedule within firm-approved rules, but a request should not be represented as acceptance of the matter. Define whether conflict review must happen first and what confirmation language staff have approved.

Does ABA Formal Opinion 512 specifically regulate AI receptionists?

The opinion addresses lawyers’ use of generative AI tools and related professional duties. It is not a product-specific certification. Firms should have ethics counsel assess how it and their jurisdiction’s rules apply to the intended workflow.

What should law firms ask an AI receptionist vendor?

Ask about audio and transcript retention, model training, subprocessors, access controls, confidentiality terms, deletion, security incidents, audit trails, integration permissions, and how urgent transfers are acknowledged.

Related Sysevo guides

Sources and further reading

Explore Sysevo’s industry workflows for connected calls, CRM records, and team follow-up.