1. ed.cafe
  2. AI safety guides
  3. Rules, Liability & MA Law
Rules, Liability & MA Law Includes: The ed.cafe Customer AI Disclosure Decision Tree & Channel Copy Snippets For Small-business owners, customer-support leads, marketers & clinic operators

Do You Need to Tell Customers You Use AI? A 2026 Channel Decision Tree

Wondering do you need to tell customers you use AI? Map chatbot, voice, email, state, and EU disclosure rules with copy-ready notices.

Constance Ibe-Whitmore The regulatory translator · ed.cafe editorial persona · Researched & verified by ed.cafe editors
Published · 14 min read (3,209 words)
ed.cafe guide banner: Tell Customers You Use AI? by Constance Ibe-Whitmore
Status as of October 2026: US federal & state rules (Massachusetts highlighted where distinct). Primary sources & sample sizes linked inline.

A composite scenario assembled from documented cases: In September 2025, a 28-person New Hampshire home-services firm turned on three customer tools in one week. Marketing launched an outbound AI voice dialer to call past HVAC leads in Massachusetts and Maine. Support renamed its chat widget "Maya, Customer Care" with a stock headshot and zero bot label. Billing connected a generative AI drafter to send auto-replies by email. Within thirty days, a Portland buyer flagged the unlabelled widget under Maine's chatbot law, effective September 24, 2025 (Maine LD 1727), while a Boston lead threatened a claim over an unconsented synthetic-voice call under the FCC's 2024 Telephone Consumer Protection Act ruling (FCC 24-17, Feb 8, 2024). Management assumed all AI use carried one rule, when the law actually turns on the channel, the state, and human pre-send review.

The short answer

If you are asking do you need to tell customers you use AI, the rule depends on whether the customer interacts directly with unreviewed AI or hears a synthetic voice. Outbound AI voice calls require prior express consent nationwide under FCC TCPA rules ($500 to $1,500 per call), commercial chatbots and voicebots require upfront labels in California, Maine, Utah regulated professions, and the EU, while human-reviewed emails generally need no label. Use the ed.cafe Customer AI Disclosure Decision Tree below.

Confusing internal drafting with live automation triggers steep fines. Calling cell phones with an AI voice without prior consent carries $500 to $1,500 per call (FCC 24-17). Utah licensed professionals face up to $2,500 per violation for unlabelled high-risk AI advice (Utah SB 226), and EU Article 50 sits in the EUR 15 million or 3% global annual turnover threshold tier for firms, lower for SMEs (Art. 50; Reg. (EU) 2026/1744). Meanwhile, KPMG and the University of Melbourne (survey of n > 48,000 people, April 2025) found 66% of employees rely on AI output without evaluating accuracy (KPMG). This October 29, 2026 guide is operational reference, not legal advice.

Why does AI customer disclosure depend on the channel rather than the model?

Customer AI disclosure statutes regulate the human boundary of the interaction rather than the model behind the screen. Regulators separate back-office drafting, where a named employee reads and owns an outgoing message, from front-stage automation, where software speaks, chats, or scores eligibility unattended.

Think of a bakery counter versus a lobby vending machine. Inside the kitchen, bakers run programmable ovens without stamping a machine notice on every loaf, because a human inspects the bread and hands it across the counter. Once you place an unattended kiosk in a hallway, consumer rules require a visible operator label and refund phone number. Where that vending analogy breaks is synthetic voice and adaptive text: nobody mistakes a metal snack box for a clerk, whereas modern chat widgets and cloned voices mimic human warmth and authority.

Four operational variables decide whether you must tell customers you use AI:

  1. Direct simulation vs human-reviewed drafting: Staff emails drafted in Claude or Copilot and edited before sending need no "AI-assisted" footer under US bot laws; autonomous replies trigger transparency rules.
  2. Voice telecom vs text screen: Spoken AI audio over phone lines activates federal TCPA consent rules and state wiretap laws before chatbot statutes apply.
  3. Clinical or licensed advice vs retail support: California clinics (AB 3030) and Utah licensed professionals (SB 226) face upfront duties for unreviewed clinical or high-risk advice.
  4. Consequential scoring vs general help: Colorado SB 26-189 (effective January 1, 2027) requires pre-decision notices and 30-day adverse explanations when AI materially influences lending, housing, insurance, healthcare, education, or employment.

Pair transparency notices with strict authority limits using our chatbot promises liability test and escalation script.

What are the rules for website chatbots, customer-service bots, and companion bots?

Commercial website chatbots need an upfront bot label whenever a US or EU visitor could mistake the widget for a live human. Small firms should also separate retail chat rules from $15,000/day "companion chatbot" headlines aimed at emotional-attachment apps.

Customer-service bots handle pricing, orders, scheduling, and support. Companion chatbots simulate sustained human friendship or emotional care. New York's companion-model law (N.Y. Gen. Bus. Law Art. 47, up to $15,000 per day) explicitly excludes standard customer-service bots (Nelson Mullins). California SB 243 (effective January 1, 2026, $1,000+ private suits and three-hour minor break reminders) likewise targets social-need companion chatbots rather than retail support widgets (FPF).

Four enacted laws govern ordinary commercial chat widgets as of October 29, 2026:

  • California BOT Law (Cal. Bus. & Prof. Code s. 17940-17943): Bars online bots from misleading Californians about their artificial identity to incentivize a commercial sale (CA LegInfo); s. 17200 fines reach $2,500 per violation unless a clear upfront label appears (s. 17941(b)).
  • Maine Chatbot Law (LD 1727 / P.L. 2025, ch. 294, eff. Sep 24, 2025): Bars text chatbots or voice "callbots" in trade that could mislead a reasonable consumer into thinking they are human without clear disclosure (Maine Legislature).
  • Utah AI Policy Act (SB 149 amended by SB 226, eff. May 7, 2025): General retail bots must disclose AI only upon a clear consumer ask, though upfront labels earn a safe harbor against $2,500 fines (Utah SB 226); Utah regulated occupations face mandatory upfront disclosure for high-risk interactions.
  • EU AI Act Article 50(1) (live Aug 2, 2026): Requires interactive AI systems whose output is used in the EU to inform users unless obvious (Art. 50). Check scope via our EU AI Act five-question check.
Jurisdiction & statuteEffective dateCommercial triggerRequired timingPenalty or remedy
California (B&P s. 17940).Live (2019)Online bot incentivizing CA sale.Upfront clear label.Up to $2,500/violation (s. 17200).
Maine (LD 1727, Tit. 10).Sep 24, 2025Text chatbot or voice callbot.Upfront clear notice.UTPA AG fines + actual damages.
Utah retail (SB 226).May 7, 2025General consumer GenAI chat.On clear ask (upfront safe harbor).Up to $2,500/violation fine.
Utah licensed pros (SB 226).May 7, 2025Licensed pro + sensitive data/advice.Prior written or upfront verbal.Up to $2,500/violation + board review.
CA / NY companion (SB 243).Jan 1, 2026Social/emotional companion bots.Upfront + minor reminders.CA $1,000+ suit; NY $15,000/day.
EU AI Act (Art. 50(1)).Aug 2, 2026Interactive AI output used in EU.At or before first interaction.EUR 15M or 3% turnover threshold (SME lower).

Why do AI outbound and inbound phone agents carry the highest legal risk in the US?

AI voice agents trigger federal telecom statutes and state wiretap laws with strict per-call damages, making synthetic phone calls far riskier than web chat. Saying "Hi, I am an AI assistant" at the start of an outbound call does not cure a missing consent record.

On February 8, 2024, the FCC unanimously issued Declaratory Ruling FCC 24-17 (CG Docket No. 02-278), classifying AI-generated voices as "artificial or prerecorded voices" under the Telephone Consumer Protection Act, 47 U.S.C. s. 227(b)(1) (FCC 24-17 PDF; FCC Release):

  1. Outbound AI sales calls require prior express written consent signed pre-dial, naming artificial/prerecorded voice and stating consent is not a condition of purchase.
  2. Outbound informational AI calls (appointment or delivery alerts to cell phones) require prior express consent, upfront caller ID, and an opt-out key.
  3. Inbound AI receptionists avoid outbound TCPA rules if they never place unconsented return voice calls, yet must satisfy Maine LD 1727, Utah SB 226, CA AB 3030, and all-party wiretap laws in states like Massachusetts (M.G.L. c. 272, s. 99, $1,000 minimum damages) and California (Penal Code s. 632, $5,000/violation) before recording. See our Massachusetts all-party consent script.

Worked numerical example (assumptions labelled): Suppose a 15-person solar firm loads 400 web leads from 2023 (submitted for human quotes, zero artificial-voice consent box) into an outbound AI voice dialer, and 300 calls reach cell phones. Even with an upfront AI greeting, all 300 calls violate 47 U.S.C. s. 227(b)(1). Base TCPA exposure is $150,000 ($500 x 300), rising to $450,000 ($1,500 x 300) if willful. Keeping AI voice inbound with an upfront bot and recording prompt drops that $150,000 TCPA exposure to $0.

When do clinics, licensed pros, and consequential scorers face extra rules?

Healthcare practices in California, licensed professionals in Utah, and eligibility scorers in Colorado face disclosure rules across portals, intake forms, and adverse notices. Across all three, human pre-send review is the primary legal dividing line.

  • California AB 3030 (Health & Safety Code s. 1339.75, eff. Jan 1, 2025): Covers clinics and physician offices using GenAI for patient clinical information (CA AB 3030). Scheduling, billing, and messages reviewed pre-send by a licensed healthcare provider are exempt; unreviewed clinical AI requires a GenAI notice plus human contact instructions (start of emails, throughout chat, both start and end of audio).
  • Utah SB 226 Regulated Occupations (eff. May 7, 2025): Utah licensed professionals (financial, legal, medical, mental health, real estate, CPA) must give upfront verbal or prior written GenAI notice before collecting sensitive data or giving personalized advice (Utah SB 226).
  • Colorado SB 26-189 (eff. Jan 1, 2027) & EU Article 50(4): SB 26-189 requires pre-decision ADMT notice, a 30-day plain-language adverse explanation with human review, and 3-year logs for consequential lending, insurance, housing, healthcare, education, and employment decisions (Legal 500). EU Article 50(4) labels deepfakes and public-interest AI text unless editorially reviewed; see EU Article 50 labelling and our FTC AI-washing checklist.
TouchpointHuman reviews pre-send?Disclose AI?Governing statute
Staff drafts email in AI.Yes (edited)No (check NDAs).Exempt under CA AB 3030 & EU Art. 50(4).
Auto-reply AI chat/ticket.NoYes, upfront.Maine LD 1727, CA B&P 17940, EU Art. 50(1).
CA clinic portal clinical note.NoYes + contact.CA AB 3030 (audio needs start + end).
Utah licensed pro intake bot.NoYes, prior.Utah SB 226 ($2,500 fine).
Outbound AI voice call.NoPrior consent.FCC 24-17 TCPA ($500-$1,500/call).
CO loan / housing score.EitherPre-notice + 30d.Colorado SB 26-189 (eff. Jan 1, 2027).

How do you run the ed.cafe Customer AI Disclosure Decision Tree across every channel?

Teams of 10 to 500 people can standardize compliance in one afternoon by routing every customer tool through the ed.cafe Customer AI Disclosure Decision Tree & Channel Copy Snippets. The ed.cafe Customer AI Disclosure Decision Tree defaults live channels to Maine's upfront rule and Utah's safe harbor while keeping human-reviewed email friction-free.

Remember one key boundary: bot labels stop impersonation fines, not contract liability. In Moffatt v. Air Canada (2024 BCCRT 149, Feb 14, 2024), Air Canada owed CAD $812.02 after its bot invented a bereavement-fare rule (BCCRT), and in 2026 the FTC hit Air AI with an $18 million judgment over false conversational-AI claims (CFO Dive). Pair every snippet with policy limits and a human handoff.

The ed.cafe Customer AI Disclosure Decision Tree & Channel Copy Snippets
THE ED.CAFE CUSTOMER AI DISCLOSURE DECISION TREE & CHANNEL COPY SNIPPETS
Verified: Oct 29, 2026 | Publisher: ed.cafe by No Worker Left Behind (EIN 88-3268312)
Operational guide for US & EU customer channels; not formal legal advice.

PART 1: 5-STEP CHANNEL DECISION TREE
1. OUTBOUND AI VOICE CALL? -> STOP until TCPA consent is logged (FCC 24-17; $500-$1,500/call).
   Sales calls need signed Prior Express Written Consent naming artificial voice.
2. NAMED HUMAN REVIEWS BEFORE SEND? -> YES: No bot label needed under CA B&P 17940, CA AB 3030,
   Maine LD 1727, or EU Art. 50(4) (check B2B NDAs). Still check Step 5 for scoring.
3. CA CLINICAL HEALTHCARE OR UTAH LICENSED OCCUPATION?
   - CA Clinical (AB 3030): GenAI notice + human contact (start of email; throughout chat;
     BOTH start and end of audio).
   - Utah Licensed Pro high-risk data/advice (SB 226): Upfront verbal or prior written notice.
4. REASONABLE USER COULD THINK IT IS HUMAN, OR OUTPUT USED IN EU? -> Persistent upfront AI
   label + human handoff (CA B&P 17940, Maine LD 1727, Utah SB 226 safe harbor, EU Art. 50).
   Add all-party recording consent BEFORE capture (MA c. 272 s. 99; CA Penal Code s. 632).
5. MATERIALLY INFLUENCES CONSEQUENTIAL DECISION (CO SB 26-189, JAN 1, 2027)? -> Pre-decision
   ADMT notice, 30-day adverse explanation + human review, and 3-year compliance logs.

PART 2: COPY-READY DISCLOSURE SNIPPETS
1. WEB CHAT HEADER (CA/ME/UT/EU): "Automated AI Assistant: You are chatting with [Company]'s
   AI bot, not a live person. Replies cannot alter written policies or pricing. Type 'HUMAN'
   anytime or call [Phone] to reach our staff."
2. INBOUND AI VOICE GREETING (MA/CA/ME/UT): "Thank you for calling [Company]. You are speaking
   with our automated AI voice assistant on a recorded and transcribed line. Stay on the line
   to continue, or press 0 now for a team member on an unrecorded line."
3. WEB FORM TCPA VOICE CONSENT (Unchecked; FCC 24-17): "[ ] Optional: I authorize [Company] to
   call my number above about [Topic] using automated dialing and an AI-generated (artificial)
   or prerecorded voice. Consent is not a condition of purchase."
4. CA CLINIC UNREVIEWED PORTAL / AUDIO (AB 3030): "Notice: Generated by generative AI without
   prior clinician review. Call [Phone] or reply 'PROVIDER' to reach a human healthcare
   provider." (Audio end line: "Reminder: Generated by generative AI. Call [Phone] for staff.")
5. UTAH LICENSED PRO BANNER (SB 226): "AI Disclosure: This tool uses generative AI, not a
   licensed professional. Do not rely on replies as [legal/financial/medical/mental health]
   advice; contact a licensed professional at [Contact]."
6. AUTO-EMAIL HEADER: "Automated AI Reply: Sent automatically by [Company]'s AI assistant
   without prior human review. Reply 'AGENT' to reach a human specialist."

Through the Massachusetts AI Hub, Massachusetts residents can earn the Google AI Professional Certificate free (Mass AI Hub). At ed.cafe, run by Boston nonprofit No Worker Left Behind (EIN 88-3268312), applications are open at ed.cafe/start/ for Circle #1, a free 30-chair cohort closing December 1, 2026 (pre-launch; zero cohorts completed yet).

Where this stops being true

  • B2B contracts and public procurement: Client MSAs and state vendor policies (like MA Policy AI.001) often require written approval before using AI on client work, even with human review.
  • Companion, therapy, and youth bots: Illinois HB 1806 ($10,000 fine) bars AI therapy decisions, California SB 243 requires minor reminders ($1,000+ suits), and New York GBS Art. 47 reaches $15,000/day.
  • Credit, securities, and election calls: FCRA/ECOA notices, FINRA rules, and political robocalls carry separate mandates.
  • 2027 rules and not legal advice: Colorado SB 26-189 starts January 1, 2027 amid federal litigation, and EU marking grace ends December 2, 2026. Consult counsel before launching outbound voice or clinical tools.

Questions people ask

Do I legally have to tell customers if my staff uses ChatGPT or Copilot to help write emails?

Generally no under US state bot laws and EU AI Act Article 50, provided a real employee reads, edits, and sends the email under their own name. California AB 3030 exempts clinical messages reviewed by a licensed provider, and EU Article 50(4) exempts text published under human editorial responsibility. Check B2B client contracts for private AI clauses.

Is it legal to use an AI voice agent to call sales leads if the bot immediately says it is AI?

No, an upfront AI greeting does not cure an unconsented outbound sales call. Under FCC Declaratory Ruling FCC 24-17 (February 8, 2024), AI-generated voices count as artificial voices under the TCPA (47 U.S.C. s. 227). Outbound AI telemarketing calls require signed prior express written consent before dialing, or callers face $500 to $1,500 per call.

What states require website chatbot disclosures for small businesses in 2026?

As of October 29, 2026, California (Bus. & Prof. Code s. 17940, commercial sales bots) and Maine (LD 1727, effective September 24, 2025, text chatbots and voice callbots that could mislead a reasonable consumer) require upfront notices. Utah (SB 226) mandates upfront notice for licensed occupations in high-risk interactions and grants a retail safe harbor.

Does New York's $15,000-a-day AI chatbot law apply to my ecommerce support widget?

No. New York's companion-model law (General Business Law Article 47) regulates adaptive systems built to simulate ongoing emotional relationships and expressly excludes ordinary customer-service bots handling business transactions. Deceptive pricing or refund claims by a support bot still violate general consumer-protection law (N.Y. Gen. Bus. Law s. 349).

Does disclosing that a chatbot uses AI protect my business if the bot quotes the wrong price or refund policy?

No. Bot-identity labels only satisfy transparency rules against human impersonation; they do not excuse negligent misrepresentation or false policy promises. As shown in Moffatt v. Air Canada (February 2024), companies remain bound by fare and refund statements on their support channels, so pair labels with strict policy guardrails.

When does EU AI Act Article 50 require a US website to label its chatbot?

EU AI Act Article 50(1) took effect August 2, 2026 and covers a US business under Article 2(1)(c) whenever the chatbot's output is used inside the European Union, such as EU shoppers on an EU storefront. Unless obvious from context, inform users at or before the first message that they are chatting with AI.

Key terms

  • Artificial voice (FCC 24-17): Synthetic or AI-cloned voice on a phone call, regulated under TCPA (47 U.S.C. s. 227) prior express consent rules.
  • California BOT Law (Cal. Bus. & Prof. Code s. 17940): Statute barring online bots from misleading Californians about their artificial identity to incentivize commercial sales.
  • Companion vs customer-service bot: Split separating social-attachment bots (CA SB 243, NY GBS Art. 47) from transactional retail widgets (CA B&P s. 17940, Maine LD 1727).
  • Human-review exemption (CA AB 3030 / EU Art. 50(4)): Safe harbor waiving AI disclaimers when a licensed clinician or accountable human editor reviews drafts pre-send.
  • Prior express written consent (TCPA): Signed consumer agreement authorizing telemarketing calls to a stated number using an artificial or prerecorded voice.
  • Utah high-risk AI interaction (SB 226): Generative AI exchange in a Utah licensed occupation collecting sensitive data or giving financial, legal, medical, or mental health advice.

Sources

  1. FCC, "Declaratory Ruling FCC 24-17" (Feb 8, 2024): FCC PDF and Release
  2. CA Legislature, "B&P s. 17940" & "AB 3030": B&P 17940 and AB 3030
  3. Utah Legislature, "SB 226" (May 7, 2025): Utah SB 226
  4. Maine Legislature, "LD 1727" (Sep 24, 2025): Maine LD 1727
  5. FPF & Nelson Mullins, "CA SB 243 & NY Companion Bots": FPF and Nelson Mullins
  6. Legal 500, "Colorado SB 26-189" (Jun 18, 2026): Legal 500
  7. EU, "AI Act Art. 50 & Reg. (EU) 2026/1744": Art. 50 and EUR-Lex
  8. BCCRT, "Moffatt v. Air Canada" & CFO Dive (FTC Air AI): BCCRT and CFO Dive
  9. KPMG & Univ. of Melbourne, "Trust in AI (n > 48,000)" (Apr 2025): KPMG
Free for Massachusetts learners

Finish Google's AI certificate with 30 neighbors in Boston.

Circle #1 is forming now (closes December 1, 2026; zero cohorts have run yet, and every completion count will be published two ways). Educators can run a Finish Line cohort in the Staff Room; L&D leaders can inspect license waste in the Council.