October 13, 2026 · LuMay AI
The Short Answer
AI legal intake uses a voice or chat agent to answer prospective-client inquiries, collect facts against firm-defined screening criteria, gather the names needed for a conflict check, request documents through secure channels, book consultations, and hand a structured summary to staff. Attorneys still clear conflicts, decide whether to accept a matter, and give all legal advice.
Most firms don't lose prospective clients to weak lawyering. They lose them in the first ten minutes, when a call rolls to voicemail or a web form sits unread overnight.
AI intake closes that gap without changing who makes legal decisions. This guide covers the full workflow, where the ethical lines sit, and the numbers that show whether it's working.
Why Legal Intake Is Worth Fixing First
Many firms simply aren't answering. For its 2024 Legal Trends research, Clio had a third-party research company contact 500 law firms as prospective clients. Forty-eight percent were unreachable by phone, and only 33% replied to email.
Intake is also where ethical exposure begins. Someone who consults your firm about a possible matter is generally a prospective client under ABA Model Rule 1.18, and what they share stays protected even if you never take the case.
So the goal isn't just answering more calls. It's answering every call consistently, collecting the right information (and not too much), and getting a clean record to a lawyer quickly.
The AI Legal Intake Workflow, Step by Step
A well-built AI intake flow runs through eight stages. Each one has a clear owner, and an attorney owns every decision that requires legal judgment.
1. Incoming Call and Disclosure
The agent answers immediately, day or night, and identifies itself as an automated assistant for the firm. Florida Bar Ethics Opinion 24-1 says client-facing AI chatbots must disclose that they are an AI program, not a lawyer or firm employee.
The opening also sets expectations: the call doesn't create a representation, the agent can't give legal advice, and the caller can ask for a person. Emergencies route straight to 911 or a staff line. A workable opening line:
“Thank you for calling [Firm Name]. I'm an AI intake assistant, not an attorney. I can collect basic information for our legal team and help schedule a consultation. Please share only what I ask for at first.”
2. Initial Screening
Next, the agent identifies the matter type: practice area, location, rough dates, and whether the caller already has a lawyer.
These questions filter out matters the firm can't take before anyone collects sensitive detail. Florida's opinion suggests screening questions that limit the conversation when a person is already represented.
3. Client Qualification
For matters that pass screening, the agent asks the firm's own qualification questions. In personal injury, that may be injuries, treatment, and insurance. In employment law, it may be employer size and termination date.
The output is a recommendation with reasons, not a verdict. Borderline, urgent, or unusual matters go to a person instead of being declined automatically.
4. Preliminary Conflict Information
The agent collects the names a conflicts check needs: the caller, opposing parties, related companies, insurers, and other known participants. It confirms spellings, because speech recognition can confuse similar names, trade names, and corporate abbreviations. It should not invite the full story yet.
Every automated search should end in one of three states:
- Possible match: stop further sensitive intake and route the record for lawyer review.
- No automated match: record the result as provisional, never as “conflict cleared.”
- Unable to search: flag the system failure internally and queue a manual check.
6. Secure Information Handling
Transcripts, recordings, and uploads all hold confidential prospective-client information. ABA Model Rule 1.6(c) requires reasonable efforts to prevent unauthorized disclosure or access.
ABA Formal Opinion 512 applies that duty to generative AI. It tells lawyers to understand a tool's terms of use and privacy policy before client information goes in. In practice, that means encryption, role-based access, set retention periods, and no vendor model training on intake data.
7. Appointment Scheduling
Qualified callers book a consultation from the attorney's real availability. The agent confirms the time, sends reminders, and logs the booking on the intake record.
Scheduling should never imply the firm has accepted the matter. And an urgent flag shouldn't disappear just because a slot was booked; it needs its own alert and an owner.
8. Structured Attorney Handoff
Instead of a voicemail, the attorney receives a structured record: contact details, matter type, key dates, screening answers, conflict names and match status, documents received and missing, qualification notes, and the transcript. A good record also separates what the caller actually said from what the system inferred, and lists low-confidence details and unanswered questions.
The attorney accepts, declines, refers, or asks for more. People whose matters are declined should receive a clear non-engagement message.
| Stage | AI agent handles | Firm staff or attorney handles |
|---|---|---|
| Answer and disclose | Answers 24/7, states it is AI, offers a person | Approves disclosure wording |
| Screening | Asks scripted eligibility questions | Writes the criteria |
| Qualification | Scores against firm rules, shows reasons | Reviews and decides |
| Conflicts | Collects names, runs a preliminary match, flags hits | Clears, declines, or seeks consent |
| Documents | Sends secure links, tracks receipt | Reviews contents |
| Scheduling | Books from real availability | Holds the consultation |
| Handoff | Delivers structured record and transcript | Accepts, declines, or refers |
AI-Assisted Conflict Screening vs. Attorney-Led Conflict Determinations
An AI agent can support conflict screening. It cannot decide whether a conflict exists. This is the line firms most often blur.
The agent can capture party names and spellings, related entities, and aliases, then query the firm's conflicts database and flag exact or near matches.
It can't judge whether a match is a real conflict, whether matters are substantially related, or whether consent or screening would cure the problem. Those questions turn on Model Rules 1.7, 1.9, 1.10, and 1.18. They belong to a lawyer.
| AI-assisted screening | Attorney-led determination | |
|---|---|---|
| Purpose | Surface possible matches early | Decide whether representation is permitted |
| Inputs | Names, entities, matter type | Full conflicts report and rule analysis |
| Output | “Possible match, review needed” | Clear, decline, or seek informed consent |
| Nature of the work | Administrative support | Professional judgment |
| Accountable | Supervising lawyer | Lawyer |
Two habits keep this boundary safe.
Collect only what's needed before clearance. ABA Formal Opinion 510 advises gathering only information reasonably necessary to decide whether the firm can and will take the matter. Over-collecting early can disqualify the whole firm from a later adverse matter.
Log every prospective client promptly. Oregon State Bar Formal Opinion 2026-208 notes that disqualifying information received through a chatbot can bar adverse representation. It advises entering prospective clients into the conflict system promptly.
What AI Intake Does Not Do
Three boundaries hold no matter which tool you choose.
It Doesn't Establish the Attorney-Client Relationship
Only the firm can agree to represent someone. But an agent that sounds too much like a lawyer can create the impression that it already has.
Florida's opinion warns that a welcoming chatbot can create a prospective-client or lawyer-client relationship without the lawyer knowing. Oregon's opinion notes that the relationship can turn on the client's reasonable belief. Clear disclaimers and attorney review of every intake record reduce that risk.
It Doesn't Give Legal Advice
“Do I have a case?” and “What's it worth?” are legal questions. Opinion 512 states that lawyers may not leave it to generative AI alone to advise clients. The agent should acknowledge the question and save it for the consultation.
It Doesn't Replace Judgment or Supervision
ABA Formal Opinion 506 lets trained, supervised nonlawyers handle intake tasks such as gathering information and running an initial conflict check. The prospective client must still be able to speak with the lawyer.
Apply the same standard to an AI agent. Opinion 512 also expects managing lawyers to set firm AI policies and vet vendors' security and confidentiality practices. Oregon adds a human point: a chatbot may miss signs that a caller is in distress, so build an escalation path for it.
Example: One After-Hours Call, End to End
At 8: 40 p.m., Dana calls an employment law firm after being let go that afternoon. Here's how a well-configured agent handles it.
- The agent answers, says it's an AI assistant that can't give legal advice, and offers a staff callback.
- Screening confirms the job was in the firm's state and Dana has no lawyer.
- Qualification captures the employer, its approximate size, Dana's role, and a two-sentence summary of why Dana believes the firing was retaliatory.
- Dana names the employer, its parent company, and a former manager. The preliminary check flags the parent company as a possible match.
- The agent texts a secure link for the termination letter.
- Because of the flag, it doesn't book a consultation. It tells Dana someone will follow up the next business day.
- At 8:30 a.m., the supervising attorney reviews the record and clears the flag: the existing client is an unrelated affiliate.
- Staff call Dana and book the consultation.
Nobody promised representation, nothing was advised, and a lawyer answered the conflict question. (Illustrative scenario, not a client case study.)
Implementation Considerations
The technology is the easy part. Most of the work is deciding, in writing, how your firm wants intake to run.
- Map today's process. Note who answers, what they ask, and where records land. Automating an unclear process only makes it consistently unclear.
- Write criteria per practice area. Define eligibility, qualification factors, and escalation triggers, and have a supervising attorney approve them.
- Define escalation triggers. At minimum: the caller asks for a lawyer; an emergency, safety concern, or imminent deadline; a criminal, detention, or restraining-order matter; a distressed caller; a conflict match; a legal question outside approved responses; or a request the agent can't understand.
- Have ethics counsel review the scripts. Disclosures, non-engagement language, and call-recording notices vary by state.
- Connect your systems. The handoff should land in your practice management system or CRM and query your conflicts database. Confirm which integrations are native and which are custom work.
- Vet the vendor. Opinion 512 points to enforceable confidentiality terms, breach notification, and data retention after the contract ends. Oregon adds the underlying software and APIs a vendor relies on.
- Pilot narrowly and test failure modes. Start with after-hours calls or one practice area, and have staff review every transcript for the first few weeks. Test what happens when an integration is down, a caller refuses recording, or a name can't be understood. A system that fails safely beats one that finishes every call by guessing.
Key Risks and the Controls That Address Them
Every risk below has a practical control. None of them require avoiding AI; they require designing for it.
| Risk | What can go wrong | Control |
|---|---|---|
| Implied representation | Caller believes the firm took the case | Disclaimers, non-engagement letters, attorney review |
| Legal advice | Agent answers “do I have a case?” | Scripted deflection to the consultation |
| Over-collection | Detailed facts taken before conflicts clear | Lean script, conflict names first |
| Data exposure | Intake data used for vendor training or leaked | Contract terms, encryption, access controls |
| Inaccurate records | Misheard names or invented details | Transcript links, uncertainty flags, staff QA sampling |
| Missed urgency | Distressed or vulnerable caller not escalated | Escalation triggers, human callback |
| Silent rejection | Good matters declined by rigid rules | Human review of declines and edge cases |
Operational KPIs to Track
Measure a baseline for at least two weeks before launch, then compare. These metrics show whether AI intake is improving speed, quality, and control.
| KPI | How to measure it | What it tells you |
|---|---|---|
| Answer rate | Inquiries answered ÷ inquiries received | Whether calls still go unanswered |
| After-hours capture | Completed intakes outside office hours | Value of 24/7 coverage |
| Intake completion rate | Completed records ÷ started conversations | Where callers drop off |
| Time to attorney review | Intake complete to first attorney action | Handoff speed |
| Conflict flag rate and clearance time | Flags per 100 intakes; hours to clear | Conflicts workload and bottlenecks |
| Consultation booking and show rate | Booked ÷ qualified; attended ÷ booked | Scheduling quality |
| Signed-matter conversion | Engagements ÷ qualified intakes | Business impact |
| Attorney override rate | Recommendations changed by attorneys | Whether criteria need tuning |
| Record accuracy | Errors found in sampled QA reviews | Data quality |
| Escalation rate | Calls handed to a person | Coverage gaps and caller needs |
A rising override rate is a useful signal, not a failure. It usually means your criteria need adjusting.
Test AI Intake Against Your Own Script
LuMay AI Voice Agents handle inbound calls with configurable call flows, a knowledge base, custom guardrails, and handoff to a person. LuMay's Legal AI Receptionist template applies that to personal injury intake. It gives a non-attorney disclaimer, checks for existing counsel, screens the case, books consultations, and texts secure document upload links.
Firms that need attorney approval before a matter opens can look at LuMay Intake Flow, part of the LuMay Legal Agents platform. Governance controls are described on LuMay's security and governance page. For a comparison of options, see LuMay's guide to AI client intake tools for personal injury firms.
The best test is your own intake script. Book a voice agent demo and walk through a real call scenario with the team.
