AI Intake for Family Law Firms: Consultations and Conflict Checks

An AI intake assistant for a family law firm answers first contact, collects the structured facts a lawyer needs to triage a consultation, and sends party names to the firm's conflict check before anyone books time. It does not advise on custody, support, property or fault, and it should not hold the emotional story of a separation in general memory. The useful design is a narrow role with written approval boundaries, a restricted record for sensitive text, and a clear handoff to staff.

Key takeaways

• An intake assistant should collect facts for triage, not the full story of the separation.
• Conflict checks work best when names and roles are captured in a standard format and the firm's own conflict process makes the decision.
• Any answer that sounds like legal advice belongs to a lawyer, and the assistant should stay on process.
• Sensitive narratives should be captured only when needed, kept in a restricted view, and deleted on a retention schedule the firm sets.
• Success is measured by complete, accurate handoffs and conflicts caught before scheduling, not by the number of chats the assistant handled.

What the intake assistant should collect

A family law inquiry usually begins with a painful life event, but the firm needs a short set of facts before a lawyer can decide whether to meet the prospect and in what time window. The assistant's job is to gather those facts consistently, accept the prospect's own words where the form allows them, and say plainly what happens next.

A sound intake record usually includes:

• Contact details, preferred language (English or Portuguese), and consent to be contacted.
• The type of matter as the prospect describes it, chosen from the firm's own list, such as divorce, parenting arrangements, support, or a marriage agreement.
• The jurisdiction where the parties live, since this decides whether the firm can take the matter at all.
• The names and roles of other people involved, such as a spouse, co-parent, new partner, employer or business partner, for the conflict check.
• Any court date, order or deadline the prospect knows about, recorded as stated and not verified.
• Whether the other party already has a lawyer, and that lawyer's name if known.
• Preferred consultation format and availability.

The list is deliberately short. Questions about fault, likely outcomes or what the prospect should do next belong to the lawyer. The assistant can ask whether a deadline exists, but it should not interpret one or calculate one.

How to run a conflict check at intake

Family matters create dense webs of relationships. The same relatives, employers and businesses can appear across several files, and a name that looks unfamiliar may belong to a former client or an adverse party. A conflict check should happen before scheduling and before the firm learns more than it needs. The assistant's role is to collect party names in a standard format and pass them to the firm's conflict process. The firm's rules and conflict records decide whether there is a conflict, never the assistant.

A workable sequence looks like this:

• Collect each party name, known alternate spellings, and role, using the same fields in every conversation.
• Ask whether the prospect knows of other names that may matter, such as former partners or related companies, and record them as given.
• Match the names against current and former clients, adverse parties and related entities in the firm's conflict system.
• If there is a possible match, hold the scheduling step and route the request to the conflicts reviewer with the matched fields only, not the prospect's narrative.
• Tell the prospect, in neutral language, that the firm is completing a review before scheduling, and give a realistic response window that staff have approved.
• Log the result, the reviewer's decision and the time it was made.

The assistant must never tell a prospect that there is no conflict. That determination belongs to the firm. Its message should describe the process, not the outcome. For a broader view of how intake and deadlines fit together in a law practice, AI Agents for Law Firms: Setup Guide for Intake and Deadlines describes the surrounding workflow.

Keeping sensitive narratives out of the system

People in a separation often want to explain everything at once. Their messages can contain allegations, health details, children's names, finances and fear. A well-designed intake keeps the structured record narrow and gives the free-text field a clear purpose. It tells the prospect what is needed and what is not, and it keeps that text away from general memory, reports and shared channels.

A practical control set includes:

• The intake record stores structured fields and a short matter summary, not the full narrative.
• Free-text content sits in a restricted view that only named staff can open.
• Any mention of violence or an immediate safety concern triggers a fixed message with the firm's emergency guidance and an immediate human alert.
• Children's details are limited to what scheduling and conflict checks need.
• The assistant's context does not carry earlier conversations unless the firm has approved that use.
• Retention periods are set by the firm and applied to transcripts, rather than left to a vendor default.
• Privacy obligations under the laws that apply in each jurisdiction the firm serves are reviewed with counsel before launch.

Prompt injection is also a real concern. Prospects can type instructions into a chat window, and an assistant that reads external text can be steered by it. The OWASP guidance on prompt injection prevention describes why instructions and data should be kept separate and why sensitive actions should not depend on the model alone. For a related problem in a clinical setting, AI for Behavioral Health Clinics: Intake and Scheduling Limits covers how intake limits are set when the content is especially sensitive.

Where the assistant stops and a lawyer starts

The boundary should be written down before launch, reviewed by the lawyers who will own it, and tested with scripted examples. Ambiguity at this line causes most of the risk. A simple decision guide helps staff and the assistant apply the same rule:

• If the prospect asks what the law says, the assistant points to the firm's approved general information and offers a consultation.
• If the prospect asks whether they will keep the house, receive support or obtain custody, the assistant declines to predict and routes the question to a lawyer.
• If the prospect describes an immediate safety risk, the assistant shows the approved safety message and alerts a staff member at once.
• If a conflict check returns a possible match, the assistant pauses scheduling and routes the case to the conflicts reviewer.
• If the prospect requests a document, a deadline calculation or a status update on a matter, the assistant records the request and sends it to staff.
• If the record is complete and no flag applies, the assistant offers consultation times the firm has approved.

Permissions should follow the same discipline. The assistant needs read access to approved calendar slots and write access to the intake record. It should not see case files, billing or prior matter narratives. Access should be scoped by role and reviewed whenever staff change, in line with the OWASP authorization guidance. The NIST AI Risk Management Framework is a useful reference for documenting these risks and their owners.

An illustrative scenario

The following scenario is illustrative and does not describe a real client. A parent contacts a family law firm after office hours through the website chat. The assistant collects contact details, the matter type as a parenting arrangement, the region where the child lives, and the other parent's name, along with whether that parent has counsel. The parent then writes a long message describing the other parent's behavior. The assistant thanks them, explains that a lawyer will review the details, and asks whether anything described is an immediate safety concern. The parent says it is not.

The assistant stores the structured fields, marks the message as restricted, and submits only the names and roles to the conflict check. The check returns a possible match with a former client. The assistant tells the parent the firm is completing a review before scheduling, and that staff will reply within the stated response window. The conflicts reviewer makes the decision the next business day. The long message never appears in the summary sent to the reviewer, and the scheduling step stays closed until the review is complete.

How a managed approach keeps intake reliable

Be Human designs and operates this kind of intake as a managed service. The first engagement defines one role, the intake assistant, the fields it collects, the memory it may use, the approval boundaries described above, and one observable success criterion, such as whether every complete intake reaches the conflict step with all required fields filled and no narrative attached. Ongoing operations include monitoring of failed lookups and escalations, recurring evaluations against scripted conversations that include tricky messages, permission reviews when staff roles change, and controlled updates to wording. Every change is logged so the firm can see what changed, when and why. The firm keeps the legal decisions, and the assistant keeps the process consistent. If you want to find the first intake task worth handing to an agent, the free Be Human AI diagnostic is a practical place to start.

Frequently asked questions

Can an AI intake assistant give legal advice to a prospect? No, and it should be configured so it cannot. The assistant can explain the firm's process, share approved general information and offer a consultation. Questions about outcomes, strategy or what a prospect should do belong to a lawyer, and the assistant should route them there with a clear message.

Should the assistant store the full story a prospect writes? Usually not in the general record. The structured fields and a short matter summary are enough for triage and conflict checks. Longer narratives can be kept in a restricted view with named access and a retention period the firm sets, and they should never flow into shared reports or the assistant's general memory.

How does a conflict check work if the assistant does not make the decision? The assistant collects party names and roles in a standard format and submits them to the firm's conflict process. The firm's conflict system and reviewers decide whether a conflict exists. Until that decision is recorded, the assistant holds scheduling and tells the prospect only that a review is underway.

Can the intake assistant serve Portuguese-speaking prospects? Yes, the service operates in English and Portuguese. Each language needs its own approved scripts, safety messages and boundary examples, reviewed by staff who read that language well. Translation alone is not enough, because the boundary between information and advice has to hold in both languages.

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