Write the disclosure script, the consent record, and the opt-out routing once, then apply them to every voice agent, chat widget, and SMS bot you run.
Every customer-facing agent an operator turned on in 2026 arrived with at least one disclosure duty attached, and the duties differ by channel and by where the customer is. In the EU, Article 50 of the AI Act (Regulation (EU) 2024/1689) has required systems that interact with people to disclose that fact since August 2, 2026 — the Digital Omnibus postponed the high-risk deadlines but left this one in place, with Article 99 fines of up to €15 million or 3 percent of worldwide annual turnover. In Texas, HB 149 (effective January 1, 2026) puts a disclosure duty on healthcare providers no later than the date service or treatment is first provided, enforced exclusively by the Attorney General with a 60-day cure period. Utah's AI Policy Act amendments (effective May 7, 2025) require generative-AI disclosure on clear and unambiguous request, and proactively, at the start of the interaction, where an individual in a regulated occupation uses generative AI in a "high-risk" interaction — one that collects health, financial or biometric data, or gives personalized recommendations or financial, legal, medical or mental-health advice. California has required companion chatbots to clearly disclose they are AI since January 1, 2026 (SB 243), and on September 28, 2026 the Governor signed AB 1609 (operative January 1, 2027), which applies only to businesses with more than $500 million in gross annual revenue nationally: they may not represent a customer-service chatbot as human, must clearly disclose that it is AI where a reasonable person would otherwise be misled, and must make a good-faith effort to connect a customer who asks to a human within 15 minutes or offer an appointment within one business day. Penalties run to $5,000 for a first violation and $10,000 for each subsequent one, enforced by public prosecutors. On top of all of that sits the TCPA: the FCC's declaratory ruling adopted February 2, 2024 (released February 8) confirmed that AI-generated voices, including voice cloning, are an "artificial or prerecorded voice," so every outbound AI call needs prior express consent, and 47 U.S.C. § 227(b)(3) allows $500 per violation, up to three times that for willful or knowing violations. Most operators are turning agents on without a disclosure script, a consent record, or a STOP-handling policy. This workflow produces all three once and reuses them across every agent.
List each AI touchpoint: voice receptionist (Vapi, Retell AI, Bland.ai, Goodcall, Rosie, Smith.ai), web chat (Intercom, Zendesk), SMS automation (Podium, Weave, Twilio-based flows). For each, record whether it is inbound or outbound, whether it can place calls or send texts without a human, and which states and countries your customers sit in. The duty attaches to the channel and the jurisdiction, not to the vendor.
Use the jurisdiction matrix in the assets. EU-facing: Article 50 disclosure, applied since August 2, 2026. Texas healthcare provider: disclosure before or at first service. Utah: disclose on request everywhere, and proactively at the start of the interaction where a regulated-occupation agent collects health, financial, or biometric data or gives personalized recommendations or financial, legal, medical, or mental-health advice. California: SB 243 if the agent is a companion chatbot; AB 1609 from January 1, 2027 if your business has more than $500 million in gross annual revenue nationally — no posing as human, disclose AI where a reasonable person would be misled, and a human within 15 minutes or an appointment within one business day. Outbound AI voice or text anywhere in the US: TCPA prior express consent. Have counsel confirm the mapping — this is a workflow, not legal advice.
The disclosure goes in the agent's opening line, not in a footer or a terms page. Voice: one sentence before the first question. Chat: the first message plus a persistent label on the widget. SMS: in the first message of every new conversation thread. Use the scripts in the assets; the vendor system-prompt field is where they live, and they must be impossible for the model to skip — put the disclosure in the fixed greeting, not in instructions the model may paraphrase.
Several of these rules (California AB 1609 explicitly, EU and Utah in spirit) expect the customer to be able to reach a person. Define the trigger words ("human", "person", "agent", "representative", "stop talking to a robot") and the route: live transfer during business hours, a callback request with a stated window outside them. Test it from the customer side on every channel. An agent that argues before transferring is the complaint a regulator will read first.
Prior express consent must exist before the agent dials or texts. Record who consented, to what (appointment reminders, review requests, marketing), on which channel, when, and how (web form, verbal on a recorded call, written). Store it in a consent log — Airtable or a table in your CRM is enough — keyed to the phone number. No log entry, no outbound message. Marketing consent is a separate, stricter record from informational consent.
The FCC revocation rule (47 CFR 64.1200(a)(10), effective April 11, 2025) lets consumers revoke by any reasonable means, and a revocation must be honored within a reasonable time not exceeding ten business days. On September 30, 2026 the FCC adopted FCC 26-67 (released October 1), which lets a caller treat a revocation made in response to an informational call or text — one with no advertising or telemarketing — as applying only to that category of informational messages, while a revocation made in response to a marketing message still revokes consent to all marketing from that caller. It also lets a caller designate an exclusive revocation method, provided the method is clearly and conspicuously disclosed on the call or in the text. The amendments take effect 30 days after Federal Register publication and supersede the January 31, 2027 waiver date, so the calendar entry moves earlier, not later — watch for the FCC public notice announcing the date. Until it takes effect, and whenever a customer's intent is unclear, treat a STOP as stopping everything from your number, and route every revocation through n8n to every sending system within the same business day.
Four dates are already on the calendar: December 2, 2026 (EU marking grace period ends), January 1, 2027 (AB 1609 operative, for businesses over $500 million in revenue only), July 1, 2027 (Utah AI Policy Act sunset), and the effective date of FCC 26-67, which falls 30 days after Federal Register publication and replaces the January 31, 2027 waiver date. Add any new agent or channel to the inventory the week it launches, and re-test the disclosure and handoff on every agent after a prompt or vendor change — those are the two moments disclosures silently disappear.
Use these templates as-is or customize for your business.
VOICE (inbound receptionist): "Thanks for calling [Business]. You're speaking with an automated AI assistant. I can book, reschedule, and answer common questions, and I can transfer you to a person at any time — just ask. How can I help?" VOICE (outbound reminder or follow-up — consent must already be on file): "Hi, this is an automated assistant calling from [Business] about your [appointment / request]. If you'd rather speak with a person, say 'person' at any time. Is now a good time?" WEB CHAT (first message + persistent widget label "AI assistant"): "Hi — I'm [Business]'s AI assistant, not a human. I can help with [scope]. Type 'human' at any point and I'll connect you with a team member." SMS (first message of every new thread): "[Business]: this is an automated assistant. Reply HUMAN to reach a person, STOP to opt out of messages." TEXAS HEALTHCARE ADDITION (HB 149 — before or at first service): Add to the new-patient intake form and the first-contact script: "We use AI tools to help answer calls and messages and to assist with scheduling. You can ask to speak with a staff member at any time." UTAH HIGH-RISK ADDITION (regulated occupation; the interaction collects health, financial, or biometric data, or gives personalized recommendations or financial, legal, medical, or mental-health advice): Disclose proactively — verbally at the start of a voice interaction, in writing before a written one — every time. Do not wait to be asked. An intake bot at a licensed practice that collects health data is high-risk even if it gives no advice. EU-FACING CHAT (Article 50): The disclosure must be present and obvious before the person relies on anything the agent says. First message plus the persistent label satisfies 'clear'; a footer link does not.
One row per consent event. Never overwrite; append. phone_e164 | +15551234567 contact_id | CRM record id consent_type | informational | marketing (separate rows; marketing is the stricter record) channels | sms | voice | both scope | e.g. "appointment reminders and confirmations" / "review requests" / "promotions" captured_at | ISO timestamp captured_via | web_form | verbal_recorded_call | written_signature | checkbox_at_checkout evidence_ref | form submission id / call recording id / scanned document id disclosure_shown | exact disclosure text version id shown at capture status | active | revoked revoked_at | ISO timestamp or blank revoked_via | STOP_text | verbal | email | web | other_reasonable_means revocation_scope | all | informational_category:<name> | marketing_all (FCC 26-67 lets you scope an informational STOP to its category once it takes effect; a STOP in reply to a marketing message always revokes all marketing; default to all when intent is unclear) propagated_to | list of systems the revocation was pushed to, with timestamps RULES: 1. No outbound AI call or text without an active row matching phone + channel + consent_type. 2. A revocation received on any channel is logged within one business day and pushed to every sending system. The rule allows up to ten business days to honor it; same day is the standard here. 3. Keep rows for at least five years from the last message sent; TCPA claims are brought years later. 4. If you designate an exclusive revocation method under FCC 26-67, disclose it on every call and in every text; if you do not, honor any reasonable means.
Touchpoint: __________ Channel: voice / chat / sms Direction: inbound / outbound JURISDICTION | APPLIES? | DUTY | WHERE IT LIVES IN OUR SETUP -------------|----------|------|---------------------------- EU (Art. 50, since Aug 2, 2026) | y/n | disclose AI interaction clearly | first turn + widget label Texas HB 149 (Jan 1, 2026, healthcare) | y/n | disclose before/at first service | intake form + first-contact script Utah AI Policy Act (May 7, 2025) | y/n | disclose on request; proactively if high-risk | greeting (high-risk) / trained response (on request) California SB 243 (Jan 1, 2026) | y/n | companion chatbot must disclose AI | greeting (only if companion-style) California AB 1609 (Jan 1, 2027; >$500M revenue) | y/n | no posing as human; disclose AI if a reasonable person would be misled; human within 15 min or appointment within 1 business day | greeting + HUMAN trigger + staffed queue TCPA 47 U.S.C. 227 + 64.1200(a)(10) as amended by FCC 26-67 | y/n | prior express consent; honor revocation within 10 business days; informational STOPs may be category-scoped once the order is effective | consent log + STOP routing REVIEW DATES: 2026-12-02 (EU marking grace ends) · 2027-01-01 (AB 1609 operative; >$500M businesses only) · FCC 26-67 effective 30 days after Federal Register publication, superseding the 2027-01-31 waiver (date once announced: ______) · 2027-07-01 (Utah sunset) Signed off by counsel: __________ Date: ________
Get a new AI workflow every week. Prompts, tool stacks, and ROI math included.
AI does the categorization or first-draft work, a human approves before action is taken. The pattern of choice for anything irreversible, externally visible, or financially sensitive.
Learn the agentic glossary →Where this workflow tends to break in production — and what to put in place before you ship it.
Disclosure placed in model instructions gets paraphrased away or skipped on short calls
Mitigation: Put the disclosure in the fixed, non-model greeting on every channel; test that it fires on 100% of sessions including one-turn calls.
STOP received on one channel, messages continue from another system
Mitigation: Single consent log of record; n8n pushes every revocation to every sending system within one business day; monthly reconciliation of sends against the log.
Agent retains or argues when a customer asks for a human
Mitigation: Trigger words transfer immediately with no retention attempt; tested from the customer side after every prompt change.
Category-scoped STOP handling switched on before FCC 26-67 takes effect, or applied to a marketing message
Mitigation: Revoke-all stays the default until the Federal Register effective date has passed; marketing revocations are always all-marketing; counsel signs off on the category mapping first.
New agent launched without being added to the inventory
Mitigation: Inventory update is a launch checklist item; quarterly re-run of the jurisdiction matrix.
Do not treat this workflow as legal advice or as a substitute for counsel — it organizes the duties so the conversation with your lawyer takes an hour instead of a day. Skip the EU rows if you have no EU customers and your chat is geofenced; do not skip them because you assume nobody from Europe visits. Skip outbound consent capture only if your agents are strictly inbound and never place a call or send a text on their own — the moment you add a reminder flow, come back. And do not switch to category-scoped STOP handling under FCC 26-67 until its Federal Register effective date has passed and counsel has confirmed how your message categories map to it; until then a STOP stops everything, because the cost of over-honoring an opt-out is a missed reminder and the cost of under-honoring one is a statutory claim.
A phased approach to get this workflow running and delivering ROI.
Days 1–30
Foundation
Days 31–60
Optimization
Days 61–90
Scale
On 30 September 2026 the FCC voted to let a STOP sent to a reminder stop only reminders, while a STOP sent to a promotion still stops all marketing. Here is how that sorts your reminder, review-request and reactivation texts into buckets, and what to keep on file.
There are 500+ AI tools marketed to small businesses. These are the 8 that actually drive revenue for most SMBs — plus what to skip.
Salesforce bills actions or conversations, Microsoft bills credits and switches you off at 125%, Google bills seats, OpenAI bills tokens plus a sandbox, Anthropic bills tokens plus session-hours. One 900-conversation-a-day agent, priced on each where the published numbers allow.
One practical AI workflow per week. No fluff.
Get the full guide with step-by-step setup, workflow templates, and copy-paste assets.