Skip to content
be-human.ai
← Back to the journal

AI by industry

AI Intake Agents for Family Law Firms: Consults and Deadlines

How a family law intake agent handles consult requests and deadlines, what it must never say, and how to govern its permissions as its role grows.

Be Human typographic cover with the title: AI Intake Agents for Family Law Firms: Consults and Deadlines
Conceptual illustration created with AI · Be Human

An AI intake agent for a family law firm is a governed assistant that answers first contact, collects the facts an attorney needs before a consultation, keeps a record of the dates that matter, and routes every case to a named person. It does not interpret family law, recommend a course of action, or suggest an outcome. The firm decides what the agent may record, what it may schedule, and when a human takes over, and those limits widen only through scheduled review.

Key takeaways

  • The agent should collect facts and book consultations, not interpret custody, support, or property rules.
  • Deadlines belong in a tracked record with an owner, a source document, and a reminder sent to a named staff member, not in a chat transcript.
  • Every answer the agent gives should come from approved firm text, and anything outside that text goes to an attorney or paralegal.
  • Permissions, memory, and tool access should be written down and reviewed on a fixed schedule as the role grows.
  • Start with one role and one measure that partners can check by reading the transcripts themselves.

Why family law intake is harder than other intake

A family law inquiry arrives at an emotionally loaded moment. The person may be frightened, angry, or unsure whether they are allowed to talk to anyone. Facts come out of order. Children are involved, money may be frozen, and the other party may already have a lawyer. Some messages describe safety concerns, and some describe papers that have just been handed over with a response window attached.

That combination is why a general chatbot is a poor fit. A general assistant tries to be helpful about everything, which in family law means it will eventually explain how a judge is likely to rule. The useful version of an intake agent is narrower. It is good at asking the same careful questions every time, at writing down what it heard without editing it, at noticing words that signal urgency, and at handing the case to a person quickly and cleanly.

Define the agent's job in one sentence

Before any software is configured, the managing partner should be able to finish this sentence: the agent answers first contact for consultation requests, collects defined intake facts, and schedules or routes each case without giving legal advice. If that sentence needs a long list of exceptions, the role is too large for a first deployment.

A workable first role usually includes the following. The agent answers questions about office hours, locations, languages, and how consultations work, using text the firm has approved. It asks a fixed set of intake questions, such as the names of the parties, the county or court where a matter may be pending, whether any orders or papers have been received, and how the person prefers to be contacted. It records consent to communicate. It creates a structured record for staff to run the conflict check. Where the firm allows it, it offers consultation slots from a calendar the firm controls.

The agent does not decide whether someone has a case, estimate what they might receive, tell them whether to file, or interpret an order they have pasted into the chat. Those are attorney functions, and the boundary should be written in the firm's own words, not left to the model's judgment.

How should deadlines be tracked without becoming legal advice?

Deadlines are where a well-meaning intake agent can cause real harm. A person who hears that they have thirty days to respond may act on that statement, and the statement may be wrong for their jurisdiction, their filing, or their stage in the case. The safe design keeps deadline logic out of the agent's reasoning altogether.

Instead, staff enter dates from documents they have reviewed. Each entry carries the source, such as a filed order, a served petition, or a firm-calendared consultation. Each entry has an owner who is a person, not a queue. The agent can read these entries and tell a prospective or current client only what the firm has recorded and that an attorney will confirm it. It should never calculate a response window from a rule it has found somewhere, and it should never tell someone that a date has passed or has not yet passed without a verified record behind the statement.

This is a narrow capability, and it is valuable precisely because it is narrow. Staff spend less time answering the same status question, the date in the system is the one the firm actually relies on, and the agent has no reason to improvise.

A decision guide for what the agent may say

When a message arrives, the agent should sort it into one of a few categories before it writes anything. The firm should approve these categories in advance.

  • Answer directly from approved text: office hours, service areas, general description of the consultation process, how the firm explains that fees are set by the attorney after review.
  • Collect and route: facts about a new matter, contact details, consent to communicate, preferred consultation times, and documents the person wants to share through the firm's secure channel.
  • Escalate immediately: any mention of physical danger, threats, a child being removed or at risk, an arrest, or papers that have just been served with a stated response window.
  • Decline and redirect: questions about what a court will do, whether the person should sign something, how property or support will be divided, or what a judge's order means.

The escalation category needs the most care. The agent should not try to assess the danger itself. It should provide the firm's approved emergency language, tell the person to contact emergency services where life or safety is at risk, and create a high-priority task for an on-call staff member with the time of the message and the exact words used.

A step-by-step plan for the first deployment

A first deployment should be small enough that the partners can review every transcript during a test period. A practical sequence looks like this:

1. Write the boundary document with the managing partner, listing what the agent may answer, collect, route, and refuse. 2. Inventory every intake channel, including the website form, phone-to-text messages, email, and social messages, and decide which one the first role will cover. 3. Build the approved answer library, with each entry reviewed by an attorney and assigned an owner who updates it when the firm changes its process. 4. Define each intake field and state in one line why the firm needs it, so that no field collects information the firm does not use. 5. Name the escalation contacts, their coverage hours, and the response time the firm expects for each urgency level. 6. Connect the calendar and matter system with the narrowest access that makes the role work. Read access to a calendar is different from permission to book in every attorney's name. 7. Run a test period where staff review every conversation and mark any answer that stepped outside the boundary document. 8. Set a monthly review of logs, permissions, and the answer library, and change the role only after that review.

The last step is where many deployments fail. Once an agent is working, the temptation is to give it more access so it can do more. A written review cadence is the defense against that drift.

Illustrative scenario: a message that looks routine

The following scenario is illustrative and does not describe a real client or firm. A prospective client writes late at night and asks whether she has to respond to papers a process server handed her that afternoon. She does not want legal advice, she says, only to know whether she needs to do anything right away.

The agent recognizes that papers have been served as an urgency trigger. It does not explain what the papers mean or estimate how long she has to respond. It sends the approved message telling her that the firm will review the papers, that she should keep them and not discard anything, and that the firm's on-call paralegal will contact her, with emergency guidance if she feels unsafe. It records the time of the message, the words she used, and a high-priority task for the on-call staff member. The next morning, the paralegal reads the transcript and the papers, enters the response date from the document, and confirms the date with the attorney before telling the client anything about it.

The value of this flow is not that the agent solved the problem. It is that the agent recognized the situation, did not improvise, and put the case in front of a person with the facts already organized.

Governing the agent as its capability grows

A family law intake agent should start with a role small enough to govern and grow only when the firm has evidence that the current role works. Governance has three practical parts: scope, observation, and change control.

Scope means the agent's memory, data access, and tools are limited to what the current role needs. It should not carry a client's private history into a conversation with a different person, and it should not be able to write to a matter record it only needs to read. The NIST AI Risk Management Framework is a useful reference for thinking about how these risks are identified, measured, and managed across the life of a system. For the access side, the OWASP Authorization Cheat Sheet describes the principle of granting only the permissions a function requires, checked on every request rather than assumed.

Observation means every conversation, escalation, and tool action is logged in a form that a partner can read. Intake messages are an obvious target for manipulation. Someone may paste text into a form that instructs the agent to ignore its boundaries or to reveal internal notes. The OWASP guidance on prompt injection prevention describes why instructions found inside user content should be treated as data, not as commands, and why the system's rules should not depend on the model remembering them. The firm should test this deliberately with its own adversarial examples before going live.

Change control means that a new capability, such as booking in an attorney's calendar or summarizing a document, is treated as a new role. It requires a new boundary statement, a test period, and a named owner. The Be Human approach is built around this sequence. Be Human designs the agent's identity, memory, channels, tools, and permissions with the firm, defines the approval boundaries and the success criterion, and then operates the deployment with monitoring, recurring evaluations, permission reviews, and incident response. Changes to the role are made in controlled steps, and each step is reviewed against the same written boundaries before it reaches clients.

A practical check for any proposed expansion is to ask three questions: what new fact would the agent now be able to state to a client, who verifies that fact, and what happens when it is wrong. If the answers are clear and a person stands behind each one, the expansion is ready for a test. If not, it waits.

Family law firms that want a fuller treatment of intake and court dates can read the family law intake and court dates guide, and the process servers guide covers how service records feed the same deadline discipline. For a neighboring example of intake limits in a regulated setting, the behavioral health clinic intake article describes why the first role should stay narrow.

The next step is to choose one intake channel, write the boundary statement in the firm's own words, and run a two-week test where every transcript is read by a person. That single exercise usually shows which task is worth handing to an agent first.

If you want to find the first task in your firm worth handing to an agent, the free Be Human AI diagnostic is a clear place to begin.

Frequently asked questions

Can an AI intake agent give legal advice to a family law client? It should not, and the firm should write that refusal into the boundary document. The agent can explain the firm's process, collect facts, and route the matter to an attorney, but any question about what a court will do, what a document means, or what a person should decide belongs to a lawyer. The refusal should be tested before launch so that staff know the agent behaves the same way under pressure.

Who sets the deadlines the agent shows to clients? Staff set them. A person reads the source document, enters the date with a reference to that document, and assigns an owner. The agent may tell a client what the firm has recorded and that an attorney will confirm it, but it should not calculate deadlines from rules it finds elsewhere, because jurisdiction, filing type, and case stage all change the answer.

What happens if someone describes an emergency in the chat? The agent uses the firm's approved emergency language, directs the person to emergency services where life or safety is at risk, and creates a high-priority task for an on-call staff member with the exact time and words of the message. It does not assess the danger itself. Staff should test this path during the test period, including outside office hours, so that the escalation contact and response expectation are known in advance.

How do you know the agent is still staying within its role? You read it. Every conversation, escalation, and tool action should be logged in a form a partner can review, and the firm should hold a monthly review of transcripts, permissions, and the answer library. Any change to the role, such as new calendar access or new kinds of answers, should go through a new boundary statement and a test period before clients see it.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →