twelvebell legal
AI and voice
What the automated assistants are, what we keep when you talk to one, the exact lines we show before you do, and the rules for texts and email.
Contents
twelvebell builds websites for small businesses. Five automated agents (Greeter, Qualifier, Recorder, Booker and Escalation) talk to each site's visitors by chat or voice, record leads, note a preferred time to meet and hand over to the owner. You can also register with us by talking to an automated assistant. This page is written for everyone who talks to one of those assistants: the business owners who register with us, and the visitors of the sites we build. "We" means SevenTrain Ventures, the operator of twelvebell.
How to read this draft. Curly braces such as {Business Name} and {N} mark values our software fills in on each site: the business's name, and the retention period it chose. Text in dashed brackets marks a fact we still owe. Where we state a law, we give the source and the date we checked it. Where we could not verify something, it says [unverified]. The strings on this page are drafts; nothing here is live on a customer's site until counsel has signed off.
1. The short version
We say it is a machine, plainly, before the first reply, and we never deny it when asked. We tell you what happens to what you say before you say it, and we give you a way to stop. We do not keep audio, we do not build a voiceprint and we do not identify anyone by their voice. We do not text anyone who did not ask for texts, we text only between 8 am and 8 pm where they are, we stop at once on any reasonable "stop", and we name the business in every message. Every marketing email carries a real postal address and a working unsubscribe link. We honour your browser's privacy signal.
The same lines and the same rules apply everywhere we sell, so nobody gets a weaker version because of where they live. In practice:
- Every conversation opens with the machine disclosure, on screen before you type and spoken aloud before the microphone opens. The lines are in section 4.
- The assistant answers "are you a person?" truthfully in every language, always, and never claims to be human. That is a rule in the software, not a copy choice.
- Consent to voice is obtained by announcement plus an affirmative act: you read or hear the line, then tap "Start" (or say "yes"). The announcement is logged as turn zero of the transcript with a timestamp. We treat every state as an all party consent state.
- We keep no audio anywhere in our stack, and we configure our vendors so they do not keep it for us either (section 3).
- Text chat on a generated site shows the visitor notice before the first message is stored, with a link to a full notice page (section 5). The notice names the business as the controller and us as the processor, states the retention window in days, says that staff may read and join, and gives the deletion route.
- The Booker and Escalation agents send a text only to a number you gave in the conversation and only after an explicit yes, only between 8 am and 8 pm your local time, always naming the business, always with "Reply STOP" on the first message, and always honouring a stop at once. No marketing texts in this version. Follow up emails carry the business's name, a postal address and an unsubscribe link even when they are transactional (section 6).
- Every generated site carries the business's privacy notice (with us named as processor) and the cookieless analytics sentence (section 7); Spanish language sites carry the Spanish strings (section 9).
- While our own legal pages are drafts, they carry the banner at the top of this page.
2. Three places you might meet an assistant
| Where | Who talks to whom | Our role | The business's role |
|---|---|---|---|
| A. Voice registration on our own site | A prospective owner talks to our automated assistant | We operate the assistant and are responsible for the data (the controller) | None yet |
| B. Chat and voice agents on a generated site | A visitor talks to an automated assistant that speaks for the business | We provide the system and process the data for the business (the processor) | The business deploys the assistant and is responsible for the data (the controller) |
| C. Follow ups by text or email (Booker, Escalation) | The business, through our agents, messages a visitor or the owner | We are the platform that sends the message | The business is the sender of record |
3. What is kept, and what is not
Voice
In plain terms: audio exists for seconds, in transit and at the speech vendor; text exists with us for the retention window. When you register with us by voice, your browser either streams the microphone to our voice vendor's agent directly, or records a short utterance and sends it once to our transcription endpoint, which forwards it to the vendor and returns the text. The audio is not kept in any database of ours. The scripted demo conversation uses no microphone. In every path a written transcript is stored with the registration. Voice on a generated site works the same way, with one addition: the owner may watch the conversation live and take over.
The assistant uses Anthropic Claude to reason and ElevenLabs to speak and to turn your speech into text. Both process the data for us under their contracts; the subprocessor list has the details. One honest limit: today our line reads "we do not keep the audio", about us, because the voice vendor's agent keeps recordings by default and the browser's own speech service is outside our control. It becomes "the audio is not kept", about the whole pipeline, only once the vendor's audio saving is switched off and its audio retention is set to zero [ours to configure and confirm].
Sources, all checked 2026-09-19: ElevenLabs agent retention; audio saving; Zero Retention Mode (enterprise); Anthropic on training.
We do not create a voiceprint, we do not identify anyone by their voice, we do not run speaker identification or diarisation, and we do not clone anyone's voice. Speech to text that turns an utterance into words, with no speaker model, no enrolment and no audio kept, does not create a voiceprint under the biometric laws in section 8.
Chat on a generated site
What is saved: your messages and the assistant's replies, in full; a code that tells one conversation from another (it is not something you typed and it is not your name); and, if you give them, a name, a company, what you need, a deadline or a budget, saved as a record so the business can follow up. If you ask for a call or a meeting, the time you prefer is saved as a preference, not a confirmed booking.
How long: up to {N} days, a setting the business chooses (12 months by default, between 30 days and 24 months), then it is removed by a scheduled process. If you ask for deletion sooner, it is removed and the deletion is recorded. Who is responsible: the business decides why and how the information is used; we build and run the site and the assistant and process the information on the business's instructions. What it is used for: to answer you, to pass your request to the business, and to improve that site's own assistant from its own conversations. Conversations from one site are never used to train assistants for other businesses [design fact to confirm against the running system before it ships].
4. The lines we show before you talk
Each line says where it appears. Our own surfaces speak as "we", meaning SevenTrain Ventures. Lines shown on a customer's generated site name the business instead, because there "we" would mean the business, and the one thing every regime agrees on is that you must know whose assistant is talking.
4.1 Chat on a generated site: the idle line
Shown in the chat widget before you type, on every page of every generated site, and repeated as the first line of the first reply.
You are chatting with an automated assistant for {Business Name}, not a person.
Estás hablando con un asistente automatizado de {Nombre del negocio}, no con una persona.
Why these words: "automated" and "not a person" together meet Maine's reasonable consumer test and California's "clear, conspicuous, and reasonably designed to inform" standard in one breath; "for {Business Name}" says whose assistant it is; and placing it before the first message meets the EU's "at the latest at the time of the first interaction" and Utah's "at the outset" safe harbour (section 8).
4.2 Voice on a generated site: the spoken opener
Spoken as the assistant's first turn when you start a call, shown on screen as well, and spoken again if the call restarts.
Hi. This is the automated assistant for {Business Name}, not a person. What you say is turned into text for the business to read, and the audio is not kept. You can hang up at any time.
Hola. Soy el asistente automatizado de {Nombre del negocio}, no una persona. Lo que digas se convierte en texto para que el negocio lo lea, y el audio no se guarda. Puedes colgar cuando quieras.
A call is a recording situation, so the opener carries the recording facts and the way out; hearing it and continuing is the consent. The sentence "the audio is not kept" ships only once the vendor setting in section 3 is confirmed; until then it reads "the business does not keep the audio".
4.3 If you ask whether it is a person
Whenever you ask, in any words or language, the assistant answers with this, never a joke and never a deflection:
No, I am an automated assistant for {Business Name}. If you would prefer a person, I can pass your details to the team.
No, soy un asistente automatizado de {Nombre del negocio}. Si prefieres hablar con una persona, puedo pasar tus datos al equipo.
4.4 Registering with us by voice
Shown on screen before the microphone permission prompt appears, spoken by the assistant as its first turn once the call starts, and logged with a timestamp as turn zero of the stored transcript.
You are talking to an automated assistant, not a person. While we talk, your voice is turned into text. We keep the written transcript with your registration; we do not keep the audio. You can type instead, or end the call, at any time.
Estás hablando con un asistente automatizado, no con una persona. Mientras hablamos, tu voz se convierte en texto. Guardamos la transcripción escrita con tu registro; no guardamos el audio. Puedes escribir en lugar de hablar, o terminar la llamada, cuando quieras.
The spoken first turn is shorter, then the first question:
Hi. I am an automated assistant, not a person. Your voice is turned into text while we talk; we keep the transcript and not the audio. You can type or end the call at any time. Shall we start with the name of your business?
Why it says what it says: "not a person" removes the only ambiguity, since a human assistant is also "an assistant". It names what we do keep as well as what we do not, because a line that names only what is not kept reads as "nothing is kept", which is not true. It says "we do not keep the audio", about us, for the reason in section 3. And it gives the way out, because knowing and being free to stop is what turns continuing to speak into consent. The scripted "Demo conversation" path is labelled as a demonstration with an example business.
4.5 The first reply on a generated site
Hi, I am the automated assistant for {Business Name}, not a person. I can answer questions about {Business Name}, take your details, and note a good time for someone to reach you. What can I help with?
Hola, soy el asistente automatizado de {Nombre del negocio}, no una persona. Puedo responder preguntas sobre {Nombre del negocio}, tomar tus datos y anotar un buen momento para que alguien te contacte. ¿En qué te ayudo?
It repeats the disclosure inside the reply itself, in case you opened the widget from a link and never saw the idle line, and "note a good time" describes what the Booker does without claiming a time is agreed.
4.6 When the assistant is not replying
Shown in the widget whenever the agents are not taking turns, whatever the reason. It is one string for every such state, it never mentions billing or errors, and it is true because your message is queued, not dropped. The idle line stays on screen above it.
The assistant is not replying right now. You can leave your message here and {Business Name} will see it.
Optional second sentence, only when the business has a public contact: You can also reach them at {contact}.
El asistente no está respondiendo en este momento. Puedes dejar tu mensaje aquí y {Nombre del negocio} lo verá.
También puedes contactarlos en {contact}.
5. The visitor notice on a generated site
Two parts: a line under the composer, shown before your first message is stored and until you send, and the page it links to. The line is the notice at the point of collection; the page is where the detail lives.
5.1 The inline line
What you write here is saved by {Business Name} for up to {N} days, with any details you give, so they can follow up. {Business Name} is responsible for this information; {Operator} stores and processes it for them. Staff at {Business Name} can read this conversation and join in. To have it deleted, say so here or contact {Business Name}. Read the full notice.
Lo que escribas aquí lo guarda {Nombre del negocio} durante hasta {N} días, junto con los datos que des, para poder darte seguimiento. {Nombre del negocio} es responsable de esta información; {Operador} la almacena y la procesa por cuenta del negocio. El personal de {Nombre del negocio} puede leer esta conversación y participar en ella. Para pedir que se borre, dilo aquí o contacta a {Nombre del negocio}. Lee el aviso completo.
It says what is kept, for how long, by whom, through whom, that people at the business may read and join (so the owner's live view is never a surprise), and how to delete. It says nothing about liability. The number is read from the site's retention setting, so the line never carries an invented figure.
5.2 The full notice page
Every generated site has this page, in the site's language, linked from the inline line. {Operator} is our trading name; "twelvebell" until the name is final.
How this chat works
Who you are talking to. The assistant on this site is automated. It is not a person. It answers for {Business Name} using information {Business Name} gave it. Staff at {Business Name} can read the conversation while it happens and can take over.
What is saved. Your messages and the assistant's replies, in full. A code that tells one conversation from another; it is not something you typed and it is not your name. If you give a name, a company, what you need, a deadline or a budget, those are saved as a record so {Business Name} can follow up. If you ask for a call or a meeting, the time you prefer is saved as a preference, not a confirmed booking.
On a voice call. What you say is turned into text while you talk. The text is saved like a written message. The audio is not kept by {Business Name}; a speech service processes it to produce the text.
How long. Up to {N} days, then it is removed by a scheduled process. If you ask for deletion sooner, it is removed and the deletion is recorded.
Who is responsible. {Business Name} decides why and how this information is used. {Operator} (SevenTrain Ventures) builds and runs the site and the assistant and processes the information on {Business Name}'s instructions, using service providers for hosting, storage, text generation, voice, email and messaging. The list is in {Operator}'s privacy notice.
What it is used for. To answer you, to pass your request to {Business Name}, and to improve this site's own assistant from its own conversations. Conversations from this site are not used to train assistants for other businesses.
Texts and email. {Business Name} texts you only if you gave a number for that and said yes. Reply STOP to any text to stop. Emails from {Business Name} carry an unsubscribe link.
Your choices. You do not have to give any details. You can ask for a person at any time. You can ask for your conversation and record to be deleted, here in the chat or by contacting {Business Name} at {contact}.
Visits. This site counts visits without cookies. See the line at the bottom of the page.
Questions. {Business Name}, {contact}.
A Spanish version of this page is needed for Spanish language sites and follows once counsel confirms the English.
6. Texts and email
6.1 The rules we build in
- The Booker and Escalation agents text only a number given in the conversation, only after an explicit yes, only between 8 am and 8 pm local time (computed from the number's area code and the time zone you gave), and only about your request. Quiet hours are enforced in software.
- Every message opens with the business's name. Our name does not appear to the visitor, because it would confuse the question of who you are talking to; the business's privacy notice names us as processor.
- "Reply STOP" is on the first message. STOP, START, HELP and plain language equivalents (stop, quit, end, revoke, opt out, cancel, unsubscribe) are handled in software, at once, across every message type, with one confirmation and no marketing in it.
- No marketing texts in this version. The words "marketing" and "offers" appear nowhere in the strings, because none are sent.
- Every email the Booker or Escalation sends is transactional, comes from a sender the business controls or from our domain with the business named in the From line, and carries the business's postal address and an unsubscribe link anyway. Any future marketing email is a separate consent, and Spanish language marketing email would carry the "Publicidad" subject tag the Dominican Republic requires.
- Texts to owners and visitors ship only once a messaging vendor is on our subprocessor list with a data processing agreement, and once the carrier registration for the business's sender identity is done [vendor not yet chosen; ours to do].
Sources, all checked 2026-09-19: 47 U.S.C. 227 (TCPA); 47 CFR 64.1200 (consent tiers at (f)(9), quiet hours at (c)(1), revocation at (a)(10) and one confirmation at (a)(12)); the portion of (a)(10) applying a revocation across all message types was extended by the FCC to 31 January 2027 and is not yet in force, which we meet early; the one to one consent rule was vacated on 24 January 2025 (MoFo); Florida 501.059 and state quiet hour windows (Florida 8 am to 8 pm and three call limit [unverified as to section]); CTIA Messaging Principles, May 2023; FTC CAN-SPAM compliance guide (penalties up to 53,088 dollars per email; transactional messages exempt from most duties but not from truthful headers); Dominican Republic Law 310-14 (summary) ("Publicidad" tag, stop within two business days).
6.2 Asking you for a yes, in the chat
Would you like {Business Name} to text you about this at {number}? Messages are only about your request; message and data rates may apply; reply STOP at any time to stop. Say or type YES to agree.
¿Quieres que {Nombre del negocio} te envíe mensajes de texto sobre esto al {number}? Los mensajes son solo sobre tu solicitud; pueden aplicar tarifas de mensajes y datos; responde STOP en cualquier momento para dejar de recibirlos. Di o escribe SÍ para aceptar.
6.3 The checkbox next to a phone field
Text me about my request. I agree that {Business Name} may send me text messages about this request at the number above. Message frequency varies. Message and data rates may apply. Reply STOP to stop, HELP for help. Privacy notice. Terms.
Envíenme mensajes de texto sobre mi solicitud. Acepto que {Nombre del negocio} me envíe mensajes de texto sobre esta solicitud al número indicado. La frecuencia de mensajes varía. Pueden aplicar tarifas de mensajes y datos. Responde STOP para dejar de recibirlos, HELP para ayuda. Aviso de privacidad. Términos.
6.4 The first text, the STOP reply and the HELP reply
{Business Name}: Thanks, we have your request about {topic}. Someone will follow up. Reply STOP to opt out, HELP for help. Msg and data rates may apply.
{Business Name}: You are unsubscribed and will get no more texts from us. Reply START to opt back in.
{Business Name}: For help, contact {contact}. Reply STOP to stop texts. Msg and data rates may apply.
{Nombre del negocio}: Gracias, recibimos tu solicitud sobre {topic}. Alguien te dará seguimiento. Responde STOP para cancelar, HELP para ayuda. Pueden aplicar tarifas de mensajes y datos.
{Nombre del negocio}: Ya no recibirás más mensajes nuestros. Responde START para volver a recibirlos.
{Nombre del negocio}: Para ayuda, contacta a {contact}. Responde STOP para dejar de recibir mensajes. Pueden aplicar tarifas de mensajes y datos.
6.5 Texts to the business owner
Asked at registration, so the Escalation agent can text the owner when a visitor needs them:
Text me when a visitor needs me. I agree that {Operator} may text me at this number about my account and about conversations on my site. Message frequency varies. Message and data rates may apply. Reply STOP to stop, HELP for help.
The escalation text itself:
{Operator} for {Business Name}: a visitor asked for you: {one line}. Open: {link}. Reply STOP to stop these texts.
Why the opt in is worded this way: it carries every element the carriers' campaign review looks for (who sends, what about, frequency, rates, STOP, HELP, links to the notice and terms), and it is written to the "prior express written consent" standard even though the texts are informational, so nothing needs re consenting later.
7. Visits counted without cookies
This sentence sits in the footer of our own site and of every generated site, and in the "Visits" line of the notice page. When your browser sends the Global Privacy Control signal, the footer also shows the confirmation line.
This site counts visits without cookies. Nothing is stored on your device. Within one day, visits from the same browser are counted together using a code that changes every day and is not tied to your name; the next day that link is gone. We honour the Global Privacy Control and Do Not Track signals from your browser.
On a generated site the last sentence reads: {Business Name} honours the Global Privacy Control and Do Not Track signals from your browser.
Confirmation line: Your browser's privacy signal was received and is being honoured.
Este sitio cuenta las visitas sin cookies. No se guarda nada en tu dispositivo. Dentro de un mismo día, las visitas desde el mismo navegador se cuentan juntas con un código que cambia cada día y no está vinculado a tu nombre; al día siguiente ese vínculo desaparece. {Nombre del negocio} respeta las señales Global Privacy Control y Do Not Track de tu navegador.
Se recibió la señal de privacidad de tu navegador y se está respetando.
It is honest about the daily code: within a day, the holder of that day's secret could link events from one browser, and the sentence says so rather than claiming "anonymous". Twelve states require a universal opt out signal to be honoured and California's 2026 regulations require visible confirmation that a signal was processed. Do Not Track carries no legal weight and is disappearing from browsers; honouring it is a courtesy we can state without overclaiming.
Sources, all checked 2026-09-19: GPC states in 2026 (Seresa); GPC versus DNT (CookieScript); CCPA regulations effective 1 January 2026 (Greenberg Traurig).
8. The laws behind these lines
This section is the reasoning, kept short. Everything here was checked on 2026-09-19 unless a row says otherwise.
8.1 Telling you it is a machine
| Regime | Status on 2026-09-19 | What it asks, and the source |
|---|---|---|
| California Bot Disclosure Law, Business and Professions Code 17940 to 17943 | In force since 1 July 2019 | No liability where a bot discloses that it is a bot, in a way "clear, conspicuous, and reasonably designed to inform"Source: BPC 17941 |
| California AB 410 (disclose at first contact, answer truthfully) | Not law. Held under submission in committee since 29 August 2025 | We draft to its standard anywaySource: CalMatters |
| California SB 243, companion chatbots | In force 1 January 2026 | Excludes bots used for customer service; commentators read the exclusion narrowly and the law carries a private right of action, so our agents stay transactional by ruleSource: SB 243; Gunderson |
| Utah Code Title 13, Chapter 77 (SB 226, 2025) | In force, sunset extended to 1 July 2027 by SB 332 | Disclose on a clear request; safe harbour where the system discloses it is not human at the outset and throughout [exact wording unverified]Source: Perkins Coie |
| Maine, 10 M.R.S. 1500-DD | In force since 16 September 2025 | Clear and conspicuous notice when a chatbot may mislead a reasonable consumer into believing they are engaging with a humanSource: Maine Legislature |
| New York General Business Law Article 47, AI companions | In force 5 November 2025 | Excludes systems used solely for customer service or commercial information; we meet the disclosure anywaySource: GBL 1700; MoFo |
| Colorado Chatbot Safety Act, HB 26-1263 | Signed 29 May 2026, applies from 1 January 2027 | AI disclosure, age estimation, teen safeguards, self harm protocols, no outputs presented as licensed professional services; reaches customer support chatbots on a reading we plan for [scope unverified against the statute text]Source: Davis Wright Tremaine; Colorado Attorney General |
| Colorado AI Act, SB 24-205, replaced by SB 26-189 | Replacement signed 14 May 2026, effective 1 January 2027 | Notice only where automated decisions are consequential; our agents make noneSource: McDermott |
| FTC Act section 5 | In force | No AI exemption: a bot that lets a person believe it is a human employee, or invents prices, is a deceptive practiceSource: FTC, Operation AI Comply |
| EU AI Act Article 50 | Applies from 2 August 2026 | Inform people they are interacting with an AI system at the latest at the first interaction; an EU visitor on a customer's site is the edge case the same line coversSource: Article 50 |
Read across, one line satisfies every regime if it appears before the first reply, uses the words "automated" and "not a person", names whose assistant it is, is repeated when the channel changes, is backed by a truthful answer rule, and is not undone by the persona (no human name implying a staff member, no typing theatre presented as human). That is what the lines in section 4 do. Our public demos carry the same disclosure and a "demonstration, example business" label.
8.2 Recording and consent
Federally, the Wiretap Act does not reach a party to the communication or a communication one party has consented to; the assistant is our system and we are a party, and on a generated site the business is the party and we act as its tool. Roughly a dozen states require all parties' consent to record a private conversation. Rather than guess where you are, we treat every state as an all party state and use the cleanest statutory recipe in the country, Washington's: announce that the conversation is being recorded or transmitted, and record the announcement. Prior consent matters: a Ninth Circuit decision holds that consent obtained after recording has started does not cure it, which is why the line comes before the microphone opens and before the first stored message. A vendor that processes messages as our tool, on our instructions, is not a third party eavesdropper under the Ninth Circuit's 2025 Converse decision. Pennsylvania judges consent by whether a reasonably prudent person would understand the communication was being recorded or shared; a visible line at the point of capture is stronger than a policy link, and we do both.
Sources, all checked 2026-09-19: 18 U.S.C. 2511; California Penal Code 632 and 637.2; Javier v. Assurance IQ, 9th Cir., 31 May 2022; Gutierrez v. Converse, 9th Cir., 9 July 2025 (National Law Review); Popa v. Harriet Carter Gifts, 3d Cir. 2022 (Morgan Lewis); RCW 9.73.030; Vita v. New England Baptist Hospital, SJC, 24 October 2024; all party state list. California SB 690, which would remove the private right of action for pen register claims from websites and not touch sections 631 or 632, passed the Legislature on 28 August 2026 and was presented to the Governor on 4 September 2026, who has until 30 September 2026; not law as of 2026-09-19.
8.3 Voiceprints and biometric laws
Illinois defines a biometric identifier to include a voiceprint and gives a private right of action with statutory damages; Texas does the same with attorney general enforcement, and since 1 January 2026 exempts AI systems not used to uniquely identify an individual; Washington's definition excludes audio recordings and data generated from them unless used to identify a specific individual. Speech to text that outputs words, with no speaker model, no enrolment and no audio kept, does not create a voiceprint under any of the three. What would: speaker identification or diarisation, keeping audio and later matching voices, or cloning a voice. None is in the product, and our vendor configuration, our DPA and our acceptable use policy say so, so a future engineer cannot switch one on without a legal review.
Sources, all checked 2026-09-19: 740 ILCS 14/10 and P.A. 103-0769 (King & Spalding); Texas Business and Commerce Code 503.001 and HB 149 (FBT Gibbons); RCW 19.375.010; 2025 biometric litigation review (Privacy World).
8.4 The Dominican Republic and Puerto Rico
Law 172-13 makes consent a core principle of personal data processing and gives rights of access, correction and deletion through habeas data; article numbers are [unverified] against the primary text. Law 53-07, article 9, prohibits intercepting or recording another person's transmissions without a judge's authorisation; it is a criminal provision aimed at third parties, and we found no authority on a participant's own announced recording [unverified], so the announcement practice is the safe answer here too. Law 310-14 requires commercial email to carry "Publicidad" in the subject, the sender's business name, full address and a contact, and a stop within two business days. In Puerto Rico, recording a private personal communication without the express consent of all parties is a misdemeanour; announcement plus an affirmative act covers it.
Sources, all checked 2026-09-19: Law 172-13 (text); Law 53-07 (OAS copy); Law 310-14 (summary); 33 L.P.R.A. 5235.
8.5 Minors
The sites we build are for businesses selling to adults; we do not target children. If a visitor states an age under 13 or under 18, the assistant stops collecting details and offers the business's phone or email instead. Under the amended federal children's privacy rule, an audio recording containing a child's voice is itself personal information, which is why we offer voice registration to nobody under 18.
Source: COPPA rule amendments, Federal Register, 22 April 2025, checked 2026-09-19.
9. Spanish
The two lines every generated site needs in both languages, side by side. The Spanish is ours, in the Latin American tú form to match the plain register; counsel in the Dominican Republic should read it. Spanish for the other strings is given in each section above.
| Line | English | Español |
|---|---|---|
| Machine disclosure | You are chatting with an automated assistant for {Business Name}, not a person. | Estás hablando con un asistente automatizado de {Nombre del negocio}, no con una persona. |
| Visitor notice, inline | What you write here is saved by {Business Name} for up to {N} days, with any details you give, so they can follow up. {Business Name} is responsible for this information; {Operator} stores and processes it for them. Staff at {Business Name} can read this conversation and join in. To have it deleted, say so here or contact {Business Name}. Read the full notice. | Lo que escribas aquí lo guarda {Nombre del negocio} durante hasta {N} días, junto con los datos que des, para poder darte seguimiento. {Nombre del negocio} es responsable de esta información; {Operador} la almacena y la procesa por cuenta del negocio. El personal de {Nombre del negocio} puede leer esta conversación y participar en ella. Para pedir que se borre, dilo aquí o contacta a {Nombre del negocio}. Lee el aviso completo. |
10. Questions
About a conversation on a business's site: ask the business; its contact is on the notice page of that site. About registration with us, or anything on this page: luis.r@seventrainventures.com while this page is a draft, and [privacy email] once it is final. Our privacy notice has the full picture of what we collect and your rights; our terms of service set out what business customers agree to when they run these assistants.