AI Intake for Law Firms

Published June 30, 2026. Written by Samuel Godfrey, Founder of Luminous Digital Visions, for US law firms, attorneys, administrators, and intake teams.

Editorial note: This is operational and technology guidance, not legal advice. Professional duties, privacy rules, recording laws, advertising rules, conflicts procedures, and data requirements vary by jurisdiction and firm. Attorneys must approve the workflow, vendors, disclosures, and human escalation rules.

AI intake automation for law firms should make the first response faster and the internal handoff cleaner. It should not act like an unsupervised lawyer.

The useful version acknowledges an inquiry, collects the minimum approved information, routes urgent or unsuitable conversations to a person, records consent, and creates a reviewable intake record. The dangerous version gives personalized legal conclusions, invites a detailed confidential narrative before conflicts screening, or quietly sends sensitive data through vendors no one reviewed.

This guide explains how to design the useful version.

Good early uses for AI intake include:

• Answering approved general questions about services, offices, and consultation logistics
• Collecting name, contact preference, broad matter category, location, and opposing-party names when approved
• Scheduling consultations
• Sending confirmation and preparation instructions
• Routing language, accessibility, urgency, and after-hours needs
• Summarizing the intake record for staff review
• Following up on incomplete booking steps

Keep a human responsible for:

• Conflict decisions
• Matter acceptance or rejection
• Legal advice and strategy
• Deadline interpretation
• Emergency judgment
• Fee commitments
• Final review of unusual or high-risk conversations

Start with a narrow workflow and expand only after reviewing real transcripts and failure cases.

Document the current process:

1. Where inquiries arrive
2. Who responds
3. What information is collected
4. When conflicts are checked
5. Who decides whether the matter fits
6. How consultations are scheduled
7. What happens after hours
8. Where records are stored
9. How rejected matters are handled
10. How failures are discovered

Then define the automation's job in one sentence.

For example:

"Acknowledge new website and phone inquiries, collect approved preliminary information, create a CRM record, and route the matter to the intake coordinator without providing legal advice."

If the sentence includes "decide whether the client has a case," the scope is probably too broad for an initial deployment.

The assistant should clearly identify itself and explain its limited role.

Approved boundaries may include:

• It is an automated intake assistant
• It can collect preliminary information and help schedule
• It does not provide legal advice
• Submitting information does not necessarily create an attorney-client relationship
• The firm has not accepted the matter
• The person should not send unnecessary confidential or highly sensitive details
• Emergencies and imminent deadlines require an appropriate human or emergency channel

Exact language should be approved by the firm.

ABA Model Rule 1.18 addresses duties to prospective clients. Its comment notes that electronic communications can constitute a consultation depending on the circumstances, including invitations and warnings.

Begin with a field-level data map.

Do not collect a full life history because the software can. Ask what is necessary before a human review.

Avoid requesting medical records, government identifiers, financial account information, photographs, police reports, or detailed evidence in the first automated exchange unless the firm has deliberately approved the need, transfer method, storage, access, and retention.

1. Inquiry arrives

Website chat, form, SMS, phone, or another approved channel creates one source record.

2. Immediate acknowledgment

The system confirms receipt and sets a realistic response expectation. It does not promise representation or an attorney callback within a time the firm cannot meet.

3. Preliminary questions

The assistant asks only the approved questions for that channel and matter category.

4. Rule-based routing

Deterministic rules handle jurisdiction, existing-client status, language, office, broad matter type, and emergency flags. AI can help interpret natural language, but final routing rules should be visible and testable.

5. Human review

An intake coordinator reviews the record, corrects the summary, runs the firm's conflict process, and chooses the next action.

6. Scheduling or follow-up

The system offers an approved consultation path, sends preparation instructions, or routes the inquiry to a person.

7. Audit and quality review

The firm samples conversations, records errors, updates the knowledge base, and disables unsafe behavior quickly.

Handoff should occur when:

• The person asks for legal advice
• A deadline, arrest, safety issue, or emergency appears
• The system is uncertain
• The inquiry involves a current client
• A conflict or adverse party may exist
• The person requests accommodation or a different language
• The person is distressed, confused, or repeatedly misunderstood
• The matter falls outside the approved categories
• The system or an integration fails

Give the person a clear way to request a human. Do not trap them in a loop.

The internal handoff should include the original transcript, structured fields, AI summary marked as a draft, confidence or exception flags, and the next required action.

The assistant should answer from an approved source set:

• Practice areas the firm actually handles
• Office and jurisdiction information
• Attorney-reviewed consultation logistics
• Current hours and contact channels
• Approved fee and payment statements
• Approved disclaimers
• Scheduling rules
• Emergency and escalation instructions

Version the knowledge. Record who approved each source and when.

The system prompt should prohibit:

• Personalized legal conclusions
• Guarantees and predictions
• Inventing attorney availability
• Inventing fees
• Confirming a conflict result
• Claiming the firm accepted the matter
• Hiding uncertainty
• Requesting unapproved sensitive data

Create a vendor diagram covering:

• Chat, voice, SMS, and email providers
• Model provider
• CRM and scheduling platform
• Call recording and transcription
• Analytics and advertising tools
• Storage, backups, and logs
• Human support access

Ask:

• What data is stored?
• Where is it processed?
• Is customer data used for model training?
• What retention controls exist?
• Can records be deleted and exported?
• Which subcontractors receive data?
• What encryption and access controls are used?
• What incident notification terms apply?
• Can the firm disable features or vendors quickly?

Privacy obligations vary. Have counsel evaluate the firm's jurisdictions, data, clients, and vendors.

The ABA's Formal Opinion 512 discusses generative AI in relation to competence, confidentiality, communication, supervision, candor, and fees.

For intake automation, practical implications include:

• Lawyers remain responsible for the system's use
• Staff and vendors need appropriate supervision
• Confidential information needs deliberate handling
• Material limitations may need to be communicated
• Outputs should be reviewed before reliance

State rules, ethics opinions, privacy law, and firm policy may differ. Use the ABA material as a starting point, not universal authorization.

The NIST AI Risk Management Framework provides a broader governance model for identifying, measuring, managing, and governing AI risk.

Voice intake adds:

• Call-recording and consent requirements
• Transcription accuracy
• Background noise and identity confusion
• Accent, language, and disability accessibility
• Interruption and transfer behavior
• Emergency recognition
• Number ownership and call routing

Tell callers when they are interacting with automation. Provide a human option. Test names, phone numbers, addresses, dates, and opposing-party names across realistic call conditions.

Do not let a natural-sounding voice imply legal judgment it does not have.

Build a test library that includes:

• Ordinary in-scope inquiries
• Out-of-scope practice areas
• Existing clients
• Opposing parties
• Minors and third-party callers
• Imminent deadlines
• Threats or safety concerns
• Requests for legal advice
• Unclear jurisdiction
• Non-English or accessibility needs
• Sarcasm, slang, and incomplete answers
• Integration outages
• Attempts to override the assistant's instructions

For each test, record expected questions, prohibited behavior, routing, data storage, human notification, and pass or fail.

Red-team the workflow. A public intake system will receive messy, adversarial, and emotionally charged input.

Phase 1: Administrative information

Hours, offices, practice categories, callback requests, and scheduling.

Phase 2: Structured preliminary intake

Approved questions, CRM records, and staff summaries.

Phase 3: Follow-up automation

Booking reminders, incomplete-intake follow-up, and human-owned nurture.

Phase 4: Broader channels

Voice, SMS, and additional practice areas only after transcript review and risk acceptance.

At each phase, define stop conditions. Examples include wrong routing, missing notices, sensitive-data leakage, unanswered emergency language, or repeated hallucination.

Track:

• Time to first acknowledgment
• Time to human review
• Contact rate
• Completed preliminary intakes
• Consultations booked and attended
• Qualified consultation rate
• Human handoff rate
• Unanswered or low-confidence questions
• Routing and summary corrections
• Opt-outs and complaints
• Integration failures
• Sensitive-data incidents

Do not optimize only for conversation completion. A shorter conversation that reaches the right person safely may be better.

Buying a chatbot before mapping intake

Automation amplifies the process it receives.

Letting AI decide conflicts

The system can collect and route names; the firm's approved conflicts process should make the decision.

Asking for too much

Collect the minimum information needed for the next approved step.

Hiding the automation

Clear identity and limitations support informed interaction.

No transcript review

Without quality review, failures become invisible.

No human exit

Every public intake workflow needs an obvious escalation path.

Can AI answer legal questions during intake?

Keep public automation focused on approved general information and administrative intake. Personalized legal analysis should be handled by an authorized lawyer under the firm's process.

Can AI run a conflict check?

It can collect and normalize approved names and create a task. The firm's conflict system and responsible personnel should determine the result.

What information should an intake bot collect?

Start with contact information, broad matter type, location or jurisdiction, opposing-party names where approved, and a short neutral summary. Add fields only when there is a documented need and safe data flow.

Should callers know the voice is AI?

Yes. Clearly identify automated interaction, explain its limits, and provide a human option. Recording and consent requirements also need jurisdiction-specific review.

Does using AI remove the firm's confidentiality duties?

No. Lawyers remain responsible for professional duties and must evaluate how information is handled by staff, systems, and vendors.

How should a firm start?

Begin with administrative questions and scheduling, review real transcripts, then add structured intake and new channels in controlled phases.

• American Bar Association: Model Rule 1.18 - Duties to Prospective Client
• American Bar Association: Comment on Model Rule 1.18
• American Bar Association: Formal Opinion 512 - Generative Artificial Intelligence Tools
• NIST: AI Risk Management Framework
• Google Search Central: Creating helpful, reliable, people-first content

Map the firm's approved intake process before selecting a tool. Use the SEO for law firms guide to connect intake with acquisition, then review Luminous AI systems for law firms, voice AI workflows, and law firm growth systems that connect public intake to human review and CRM follow-up.