AI clones for legal intake: prenup, PI, clinical research use cases.

Why intake is the right wedge for legal AI

Legal work divides roughly into two phases: intake (qualifying the prospective client, gathering facts, deciding whether the firm will take the case) and representation (the actual legal work). AI is dangerous in the second; it's useful, properly bounded, in the first.

Intake has a specific shape:

• Structured. Nearly every firm runs the same intake checklist across cases of the same type. Prenup intake has its fields; PI intake has its fields.
• High-volume. Many inquiries; only some become clients.
• Time-sensitive. Prospects shop. Slow responses lose cases.
• Non-advisory. Intake is fact-gathering, not counsel. The answer to "should I do X?" during intake is "let's set up a consultation with the attorney."

That shape fits an AI protégé far better than, say, drafting a brief or interpreting case law. An AI protégé can do intake with the speed of a 24-hour answering service and the structure of a paralegal, without attempting to give legal advice. The right architecture makes this safe; the wrong one makes it a bar complaint.

Intake is fact-gathering with clear handoff points. That's exactly what a well-built AI protégé does. Everything past intake needs a human.

Use case 1: Prenuptial agreement qualification

The intake shape. A prospect fills out a form or calls a family-law firm to ask about a prenup. The firm needs to know: engaged or married, jurisdiction, significant assets, significant debts, prior marriages, children, business ownership, inheritance considerations, timing of the wedding. Most of those questions have the same phrasing across every inquiry.

What the AI protégé does:

• Asks the structured questions in an order that mirrors the firm's paper intake
• Captures answers into a structured-output JSON schema so the attorney sees structured fields, not a wall of transcript
• Flags specific indicators (international assets, recent cohabitation property, a wedding in under 30 days) that change which attorney in the firm should handle the case
• Refers to the firm's published prenup-relevant content when the prospect asks general questions, not giving legal advice, but citing the firm's own explainers
• Ends with a scheduling handoff: Calendly link or a call-back request

What it doesn't do:

• Give legal advice on whether a prenup is "worth it"
• Draft or preview language
• Opine on whether a specific asset is separate or marital property
• Estimate fees (that requires attorney review)

We have prenup-attorney as a dedicated campaign landing page, and the legal intake pattern is live with Matt Rossetti. The structured-data-on-intake pattern is what makes this usable; the attorney gets a clean form on first glance, not a transcript they have to read.

Use case 2: Personal-injury case screening

The intake shape. PI firms screen cases aggressively. Most inquiries don't qualify: incident outside the statute of limitations, no liable party identifiable, injuries too minor, state jurisdiction issues, prior representation conflicts.

The conventional firm runs a long, expensive human triage. Most calls end in "we can't take your case," which is both time-consuming for the firm and discouraging for the caller.

What the AI protégé does:

• Walks through the incident, capturing: date, location, type of incident, injuries, medical treatment, police report, insurance involvement, other parties, prior legal action
• Identifies the dispositive facts early, so a date outside the statute of limitations ends the call compassionately in minutes, not after a 30-minute human intake
• Captures contact information and medical-treatment timeline in structured form
• Flags high-value indicators (commercial defendant, clear liability, significant medical treatment) for immediate attorney routing
• Respects privilege by never opining on fault or recovery
• Ends with either a booked consultation (qualified) or a kind close (not qualified)

What it doesn't do:

• Estimate settlement values
• Advise on whether to accept an insurance offer
• Take positions on fault
• Promise outcomes

The scale economics are strong here. PI firms commonly run call centers to triage. An AI protégé handles the initial sort, captures structured data, and hands off only the qualified cases to human staff. Our personal-injury-attorney landing page is live for this use case.

Use case 3: Clinical research participant screening

The intake shape. Clinical trials screen prospective participants against protocol-specific inclusion and exclusion criteria: age ranges, medical history, current medications, geographic proximity to the trial site, ability to attend required visits. The criteria are written into the protocol; screening is a translation from participant answers to protocol yes/no.

This use case shades into healthcare, which means HIPAA considerations are gating. What's below assumes you have HIPAA-compliant infrastructure; our current posture on that (stated plainly in our PII handling post) is partial, not certified.

What the AI protégé does:

• Walks a prospective participant through the inclusion/exclusion criteria, adapting questions based on prior answers (skip further medication questions once a disqualifying medication is identified)
• Captures structured output matching the protocol's screening fields
• Handles the high-volume "am I eligible?" inbound that clinical trial recruiting produces
• Refers qualified candidates to a human coordinator for informed-consent and formal enrollment
• Respects the fact that pre-enrollment screening is fact-gathering, not medical advice or trial-specific counseling

What it doesn't do:

• Diagnose
• Recommend medication changes
• Provide the informed-consent conversation (that's regulated)
• Promise enrollment

The clinical-research use case has been tested in pilots; it's promising but depends on the firm's underlying HIPAA posture. Don't skip that.

What the AI protégé should and should not do

Across all three use cases, the pattern is the same:

Should:

• Run structured intake
• Capture fields into structured output
• Refer to the firm's published content (website pages, FAQs, attorney bios, blog posts)
• Flag cases for specific human routing
• Schedule handoffs (Calendly, call-back request)
• Say "I don't have that information" when asked something outside scope (zero-hallucination architecture)

Should not:

• Give legal advice
• Opine on case outcomes
• Quote fees as binding
• Diagnose, in the medical context
• Execute autonomous actions without human review (if your rules engine is wired to send documents, that send should be human-approved in a legal context)
• Pretend to be a human

The last point matters. The protégé's opening greeting should identify it as an AI intake assistant and name the attorney or firm behind it. Jurisdictions vary on disclosure requirements for AI in legal contact; be explicit.

Privilege, confidentiality, and conflicts

Three ethics considerations come up repeatedly:

Privilege

Attorney-client privilege generally attaches when a prospective client communicates with an attorney or the attorney's agent for the purpose of seeking legal services, even if no representation ultimately results. Most jurisdictions extend this to intake conversations. An AI protégé operating as an agent of the firm likely falls within the privileged sphere. Consult your bar's guidance; this is a jurisdiction-specific question.

Practical implication: Session transcripts of intake conversations should be treated as privileged material. Our session-record access controls scope transcripts to the expert (attorney) and their firm admins. PII redaction on ingestion means sensitive details do not enter the protégé's public-retrieval knowledge base; the transcript is private to the session record.

Confidentiality

Distinct from privilege but related. The firm owes a duty of confidentiality to prospective clients even when no representation follows. A platform that stores intake transcripts in an auditable, access-controlled way supports this; a platform that muddles data across clients does not. Ask your vendor explicitly about cross-tenant data isolation.

Conflicts

Before the firm can take a case, a conflicts check runs. The AI protégé can capture the information the conflicts check needs (parties named, adverse counsel, prior representations in the prospect's memory) but should not attempt to clear the conflict; that's a firm-level process. The protégé hands off the captured information for human conflicts review before any substantive engagement.

FAQ

Can my AI protégé represent me in court? No. The protégé does legal intake and fact-gathering. Representation is the attorney's job, and nothing in current rules permits an AI to appear or advocate.

Is this different from a chatbot on my website? Yes, in three ways: voice-first vs text-only, architectural accuracy constraint instead of prompt-level grounding, and structured data capture into your firm's intake fields. A well-built chatbot can do 20% of what a structured intake protégé does.

What about unauthorized-practice-of-law concerns? The rule in every US jurisdiction is some variant of "non-lawyers cannot practice law." AI systems that give legal advice may cross the line; AI systems that do intake generally don't, provided they (a) disclose they're an AI, (b) don't opine on legal questions, and (c) route to a licensed attorney for any advisory component. Your state bar may have specific guidance on AI in legal practice; check. The ABA has published guidance but state interpretations vary.

Can I use this for family-law matters other than prenups? Divorce and custody intake follow the same structured-intake pattern. We haven't built a specific case-study flow for divorce but the infrastructure applies. For custody-adjacent matters involving minors, add additional disclosure and care around data handling.

How does my existing law firm software fit in? Most firms already run intake through Clio, MyCase, PracticePanther, or similar. Our structured output can format captured fields in whatever schema your PMS expects. Webhook integration is supported; direct API sync per firm is on a case-by-case basis (ask support).

What's the compliance posture for healthcare / clinical research use? Our general compliance posture is partial, not certified. HIPAA-gated work requires a BAA, which we evaluate on a case-by-case basis. Start the conversation early if this is your use case; the procurement lead time for named certifications is longer than the product lead time.

Are there named law firms using this today? Matt Rossetti's practice is the reference case for legal intake on Apex Replicant: qualitative reference, with metrics being collected through current client engagements. Case-study publication pending the paid-campaign data cycle.

Related reading

• How do I deploy an AI expert clone for a law firm?
• What is a zero-hallucination AI architecture?
• How do AI digital proteges handle PII?