Your AI intake chatbot probably meets more potential clients than your front desk ever will. That’s great—until the wording in that chat creates ethics headaches, privacy issues, or (worst) a surprise client relationship you never meant to start.

In 2025, bars and privacy regulators want a few basics up front: say it’s a bot, say it’s not legal advice, say no attorney‑client relationship is formed yet, and explain what you do with the info people type in.

Below, you’ll see what to say, where to say it, and how to get real consent without tanking conversions. We’ll cover privacy and retention, emergencies, accessibility and language options, short‑ vs long‑form examples, state‑specific lines, multichannel tweaks, and a checklist you can run tomorrow. We’ll also show how LegalSoul makes these disclaimers dynamic and audit‑ready.

Key Points

• Tell users up front: it’s a bot, it’s general info, it’s not legal advice, no attorney‑client relationship yet, what not to share, how and when a human follows up, and (when required) “Attorney Advertising.” Always give a non‑AI contact option.
• Get real consent: show a short disclaimer before any questions, use a clear “I agree,” record timestamp/IP/text version, and add state lines where needed (CA bot + CPRA notice, NY advertising label, TX specialization, WA MHMD health‑data caution).
• Treat the chat as data collection: list what you collect, why, how long you keep it, who sees it; discourage sensitive uploads; set transcript retention and purge rules; keep it ADA‑friendly and offer the same disclosure across chat, SMS, voice, and DMs.
• Build in safety: emergency/deadline warnings and escalation, practice‑area tweaks, version control, regular reviews, QA to prevent advice, and staff training. LegalSoul can handle the dynamic disclaimers and consent logs so you don’t have to babysit it.

Why your AI intake chatbot needs a disclaimer (2025 context)

If your chatbot starts the conversation, your disclaimer is doing the risk work your receptionist used to do. Bars usually treat intake chat as a communication about legal services, which brings attorney advertising rules into play (think New York Rule 7.1, Florida 4‑7.11–4‑7.20). California requires a bot disclosure when you’re promoting services (Bus. & Prof. Code §§17941–17943), and the CPRA wants a notice at collection for personal data. Washington’s My Health My Data Act can kick in fast if users share health details. ABA opinions also remind lawyers to mind confidentiality and supervision around tech.

The upside: better leads. Clear expectations mean fewer walls of text and more usable contact info. Put “not legal advice” and “no attorney‑client relationship” above the first question, then ask for consent. That small order change cuts reliance risk and tends to boost completion. Multi‑office? Use geo to add state‑specific lines to a national base so you don’t scare folks off with a novel‑length warning they won’t read.

Core elements every AI intake disclaimer must include

• Automated agent: “You’re chatting with an automated intake assistant.”
• No legal advice: general information only; not a substitute for a lawyer’s advice.
• No attorney‑client relationship: nothing starts until conflicts are cleared and you sign an engagement.
• Confidentiality limits: don’t send sensitive or urgent details; staff may review submissions.
• Human follow‑up and timing: when you’ll hear back and a phone number for urgent matters.
• Privacy and consent: link to Terms/Privacy; explain data categories, purpose, retention, and service providers.
• Attorney advertising label where required.
• Accessibility and alternatives: ADA‑friendly, plus phone/email options.

Try this short intro: “We’ll ask 5 quick questions to see if we can help. This assistant is automated and can’t give legal advice. Using this chat doesn’t create an attorney‑client relationship. Deadline today or not sure? Call us at [number]. By continuing, you agree to our Privacy Policy and Terms.”

Tailor by practice. PI? Add a health‑data caution. Criminal? Add after‑hours routing. Keep it tight: short form first, link to the long form, then the consent button. Users keep moving, you get enforceable assent.

Placement, timing, and consent mechanics that hold up

Court cases keep saying the same thing: clickwrap wins, browsewrap loses. People need conspicuous text and an obvious “I agree.” Nguyen v. Barnes & Noble (9th Cir. 2014) and Berman v. Freedom Financial (9th Cir. 2022) both hammer passive footer links. So, put your short disclaimer before any questions, show your links to Terms/Privacy, and get a real click.

A simple flow:

1. Welcome line
2. Short disclaimer
3. Links to Privacy/Terms
4. “I agree—continue” button
5. First intake question

Log timestamp, IP, device, user agent, and the exact text version shown. If you change terms, refresh consent. If a chat switches to a human, flash a mini‑banner that repeats no relationship until conflicts and signature.

One small tweak that helps: first ask “I understand this isn’t legal advice,” then “Continue.” You’ll see fewer drop‑offs and a cleaner consent record.

Privacy, data use, and retention disclosures

Treat the chat like a form that collects data. In California, the CPRA wants a notice at collection listing categories, purposes, and retention. The Colorado Privacy Act needs consent for sensitive data. Washington’s MHMD adds strict rules on health data and even geofencing for clinics.

Write it plainly: what you collect (text, uploads, metadata), why (to evaluate representation), who can see it (staff and vetted processors), and how long you keep it. Explain how to request access or deletion. Many firms keep non‑client transcripts 12–24 months, shorter for sensitive uploads, with an auto‑purge. Don’t brag about “bank‑level security” unless you can back it up; just say you use reasonable safeguards.

Add a guardrail before any file upload: “Please don’t upload IDs, Social Security numbers, or medical records here.” It prevents a lot of unnecessary risk and nudges those items into a secure portal after engagement.

Avoiding inadvertent attorney‑client relationships and reliance risk

Websites can create duties you didn’t intend. Keep your no attorney‑client relationship message clear and visible. Spell out the conditions: conflicts check and a signed engagement. Keep it close to the consent button so users can’t miss it.

• Use direct language: “Using this chat does not create an attorney‑client relationship.”
• Don’t give specific advice or deadlines in chat; keep it general and hand off to a person.
• Texas needs a specialization disclaimer unless a lawyer is certified by the Texas Board of Legal Specialization.
• Avoid claims like “we win almost every case.” NY and FL see those as misleading without context.

If someone asks for advice, fire a “reliance breaker”: “I can’t provide legal advice here. If you have a deadline, call [number] now.” Train staff to repeat the same line during handoffs so your record is consistent.

Emergencies and time‑sensitive matters

Missed deadlines cause claims. Don’t let urgent matters sit in a chat queue. Add an obvious warning and a phone number. For example: “If your hearing is within 48 hours or you have an imminent deadline, don’t use this chat—call [number] now. If this is a safety emergency, call 911.”

• Detect time words like “tomorrow,” “hearing,” “deadline,” and offer a call option or live agent.
• After hours, use voicemail with the same disclosure and an email auto‑reply that repeats it.

For PI or criminal, consider listing relevant hotlines (like domestic violence help lines) when appropriate. It helps people and shows you tried to prevent reliance on a slow channel.

Track how often the emergency rail triggers. It tells you if you need more on‑call coverage or if your ads are attracting too many last‑minute cases to a bot.

Accessibility and language access

Accessibility isn’t optional. DOJ points to WCAG 2.1 AA as a practical standard and many ADA settlements use it. Your disclaimer should work with screen readers, be keyboard‑friendly, and pass color contrast checks. If the chat starts in Spanish, show the disclaimer in Spanish. Always offer a phone or email alternative.

• Keep the short‑form at an 8th–9th‑grade reading level.
• Offer audio read‑outs on mobile if you can.
• Use expandable sections so assistive tech doesn’t drown in text.

Serve immigrant communities? Note that transcripts may be reviewed by staff and language service providers under confidentiality. Log which language version was accepted. If consent is challenged later, that detail helps.

Short-form vs. full-form disclaimer examples

Short‑form (above the first question):
“Attorney Advertising. This automated intake assistant provides general information and can’t give legal advice. Using this chat doesn’t create an attorney‑client relationship. Don’t send confidential or time‑sensitive info. For emergencies or deadlines, call [number]. By continuing, you agree to our Privacy Policy and Terms.”

Full‑form (expandable) should include:

• Who you are and advertising status in the user’s state.
• Bot disclosure and the limited scope (general information only).
• No attorney‑client relationship until conflicts are cleared and an engagement is signed.
• How info is used, who sees it, retention length, and transcript logging.
• Sensitive data cautions (no SSNs, medical records, bank numbers).
• User rights (access/deletion) and how to request them.
• Accessibility options and non‑AI ways to contact you.

Use both. Short‑form for clarity and speed, full‑form for details and compliance. On mobile, aim for 60–80 words in the short‑form and a bold “I understand—continue” button so consent is crystal‑clear.

State-specific requirements and sample language (2025)

California
Requirements: Bot disclosure (Bus. & Prof. Code §§17941–17943); CPRA notice at collection.
Sample: “You’re interacting with an automated chatbot. This is attorney advertising. See our Notice at Collection for the personal information we collect and why.”

New York
Requirements: Attorney advertising label; avoid misleading claims (Rule 7.1).
Sample: “Attorney Advertising. This automated intake tool provides general information only and is not legal advice. No attorney‑client relationship is formed.”

Florida
Requirements: Rules 4‑7.13 and 4‑7.14; no promises of results.
Sample: “This AI intake assistant can’t provide legal advice or predict outcomes. Representation begins only after conflicts clearance and a signed agreement.”

Texas
Requirements: Specialization disclaimer unless certified.
Sample: “Unless stated otherwise, attorneys are not certified by the Texas Board of Legal Specialization. Using this chat does not create an attorney‑client relationship.”

Illinois
Requirements: Consent for recording calls/voice notes.
Sample: “If we record a call or voice note, we’ll ask for your consent first.”

Washington
Requirements: My Health My Data Act (heightened health‑data rules).
Sample: “Please don’t submit detailed health information here. We’ll review and contact you.”

Colorado
Requirements: Colorado Privacy Act.
Sample: “We collect limited personal data for intake and share it with service providers as needed; see our Privacy Policy.”

Always run final language by local ethics counsel, especially if your practice mix or audience shifts.

Practice-area nuances that change your disclaimer

Personal Injury/Medical: Add a health‑data caution and tell people not to upload medical records here; Washington’s MHMD and similar laws are sensitive about that.

Criminal Defense: Put the phone number front and center, include after‑hours help, and discourage confessions in chat. Escalate fast.

Immigration: Offer multilingual options, note that timelines depend on agencies, and avoid promises.

Family Law: Mention minors’ privacy and protective‑order issues; discourage posting children’s PII or orders in chat.

Employment/Wage: Flag class/collective action risks, and don’t invite confidential employer documents before engagement.

Across the board, repeat the conditions: conflicts check and a signed agreement before representation starts. Add practice‑specific trigger words (like “restraining order,” “ICE,” “statute runs Friday”) that pause intake or escalate to a person. Save proof when the bot nudges users to non‑AI channels.

Multichannel intake: chat, SMS, email, voice, and social DMs

Each channel needs its own twist. SMS must follow TCPA/CTIA: get explicit consent, say who you are, how often you’ll text, and include STOP/HELP. Keep the no‑relationship/no‑advice line short. Voice needs an audible bot disclosure and (in states like IL or CA) recording consent. Email auto‑replies should include advertising labels where needed and repeat that nothing starts without conflicts and a signed agreement.

Social DMs are screenshot land. Pin a short disclosure with a link to your full policy, then route folks to a secure form. If you do call‑backs or voice notes, log verbal consent and store transcripts under the same retention rules as chat. Keep the phrasing consistent across all channels so users don’t get mixed messages.

Implementation checklist and audit trail

• Version control: give every national and state variant an ID and date.
• Consent logs: timestamp, IP, user agent, geo, language, and the exact text shown.
• Data retention: set different windows for non‑client vs. client transcripts; auto‑purge; map your processors.
• Links: show Privacy/Terms at the point of collection and link to the versions in effect at consent.
• Accessibility: WCAG 2.1 AA checks, keyboard nav, color contrast.
• Escalation: emergency and after‑hours playbooks; SLAs for callbacks.
• QA: test prompts so the bot refuses individualized advice.
• Change management: re‑consent when terms materially change.

Align your transcript and deletion policy with malpractice carrier expectations and any recordkeeping rules. A simple monthly “consent ledger” sent to your privacy lead and ethics counsel catches gaps early—especially after a site redesign or CRM swap.

Governance, testing, and training

Treat the disclaimer like living text. Review quarterly for rule and privacy updates. A/B test placement and phrasing—“I understand—continue” vs. “I agree,” one sentence vs. two. Track acceptance and drop‑offs and adjust.

Run red‑team tests that try to bait the bot into advice (“Should I plead guilty tomorrow?”). It should refuse and escalate. Save these tests with production logs to show supervision (see ABA guidance on virtual practice). And keep your clickwrap approach consistent across web, SMS, and voice. These artifacts help at cyber insurance renewals too.

Common mistakes to avoid

• Hiding the disclaimer in a footer. Courts don’t like browsewrap.
• Collecting facts before consent. Ask first, then start intake.
• Over‑promising outcomes or security. No “we win 99%” or vague security boasts.
• Skipping state lines. Think TX specialization, NY advertising label, CA bot disclosure.
• Ignoring accessibility. WCAG gaps are low‑hanging risk.
• No emergency routing. Deadlines shouldn’t sit in a queue.
• Inconsistent language across channels. Keep the message uniform.
• No retention plan. CPRA and CPA won’t love that.
• Untrained staff. One off‑script handoff can undo careful wording.

Easy win: show a one‑line Texas specialization disclaimer to Texas visitors by default unless you detect a certified specialist profile. Fewer misses, less maintenance.

Copy-and-paste templates you can adapt

Baseline short‑form (web)
“Attorney Advertising. You’re chatting with an automated intake assistant. This is general information, not legal advice. Using this chat does not create an attorney‑client relationship. Don’t share confidential or time‑sensitive info. If it’s urgent, call [number]. By continuing, you agree to our Privacy Policy and Terms.”

Baseline long‑form (key clauses)

• Automated agent disclosure; scope and limits.
• No legal advice; no attorney‑client relationship until conflicts cleared and engagement signed.
• Data categories, purposes, retention; sharing with service providers; access/deletion rights.
• Sensitive data caution (no SSNs/medical records).
• Accessibility and non‑AI alternatives.

SMS
“[Firm]: Automated intake. Not legal advice; no attorney‑client relationship. Msg/data rates may apply. Reply STOP to opt out. Urgent? Call [number].”

Voice (pre‑recorded)
“This is an automated intake assistant. This call may be recorded for quality and training. If you consent, press 1. Don’t share confidential or time‑sensitive details. This is not legal advice, and no attorney‑client relationship is formed. For emergencies, hang up and dial 911.”

How LegalSoul supports compliant, state-aware disclaimers

LegalSoul helps you put all of this in place without duct‑taping tools:

• Dynamic geo‑targeting: adds California bot disclosure and CPRA notice, New York advertising labels, Texas specialization language, and Washington MHMD cautions automatically.
• Clickwrap consent capture: clear “I agree” with immutable logs (timestamp, IP, text version, language) and easy exports for audits or carrier reviews.
• Privacy sync: pulls your current Privacy Policy/Terms into the widget and prompts users when versions change.
• Data controls: different transcript retention windows for non‑clients vs. clients, auto‑purge, and legal‑hold options.
• Accessibility and language: WCAG 2.1 AA‑friendly widgets with multilingual short‑ and long‑forms.
• Escalation workflows: emergency routing, after‑hours playbooks, and human handoffs that repeat the key disclosures.

End result: intake that’s fast, compliant, and easier to monitor. Many firms see fewer “ghost” leads and cleaner conflicts data once the disclaimer sets expectations.

FAQs

• Is an AI intake chatbot considered attorney advertising? Often. Many states treat site and chat content as communications about legal services. Use “Attorney Advertising” where required and avoid unjustified expectations.
• Do I need user consent before any data collection? Best practice: yes. Show the short disclaimer and links to Privacy/Terms first, then get clickwrap consent.
• How do I handle users who refuse the disclaimer? Offer a phone number and email form. Log the refusal so you can show you didn’t take data without consent.
• What if the chatbot collects sensitive health data? Use just‑in‑time warnings and require explicit consent. In Washington, MHMD adds extra obligations.
• Can I use one national disclaimer for all states? Use a national base plus geo‑layered state lines (e.g., Texas specialization, California bot disclosure). Clearer and safer.
• What consent format holds up? Clickwrap. Courts want conspicuous notice and affirmative assent. Apply it across channels.

Conclusion

A solid AI intake disclaimer is basic risk control now. Say it’s a bot, say it’s not legal advice, say no attorney‑client relationship yet, and explain privacy, retention, accessibility, emergencies, and any state‑specific rules (CA bot/CPRA, NY advertising, TX specialization, WA MHMD).

Done right, it lowers reliance risk and improves lead quality. Want it handled without a rebuild? Book a quick LegalSoul demo and launch state‑aware disclaimers with audit‑ready consent logs, multilingual/ADA support, and sensible retention controls.