AI and Transparency Policy

1. Where AI is used

AI assists with these platform features:

• Content and creatives: ad copy, posts, emails, images, videos, voices and avatars, based on the Customer's business information.
• Platform chat: the dashboard assistant suggests actions and answers the user's questions.
• Replies on messaging channels: suggested replies and, only if the Customer turns them on, automatic replies on WhatsApp, Instagram or Messenger.
• Voice agent: answers or makes calls on the Customer's behalf, only if the Customer creates and turns it on.
• Transcription and summaries: of calls, meetings and dictation.
• Lead scoring: a contact's score is calculated by deterministic point-based rules; AI only writes the explanation, and what a person sets manually prevails.
• Closed classifications: choosing among predefined options (for example, the type of message or the specialist who answers). These classifications do not write text or make any decision with legal effects about a person.
• Public website assistant: an AI that only chooses among human-written answers from our FAQ; if none fits, it says so and offers to connect you with a person.

2. Notice that it is an AI

People have a right to know when they are talking to an AI. The platform sets the notice based on the location of the person who receives or makes the contact, not the Customer's; if it is unknown, the strictest rule applies.

• The notice is fixed text (in Spanish, English and Portuguese), not written by the AI, so it does not depend on the model.
• In chat it is shown once per conversation, and again after 24 hours of silence.
• In every location, if someone asks whether they are talking to a person, the AI tells the truth.
• The person can always ask to speak with a person from the business.

3. What the system does on its own and what it does not

By default everything is manual: the platform proposes and a person approves (in the dashboard, via WhatsApp or by email with a signed single-use link).

• Automatic agents: exist only if the Customer creates and turns them on. They start switched off, and each has a daily cap, schedule, a log of every decision, an undo option and a global stop button.
• Flows and automations: run on their own because a person on the Customer's side set them up and activated them.
• Ads: protection rules only pause or lower budgets. Raising a budget only happens with a monthly cap turned on, below 80% spent and in steps of up to 20%.
• Always with a person's approval: activating campaigns, reallocating budget, refreshing creatives and creating forms.
• The platform chat always asks first before publishing, sending or spending.
• Calls: the voice agent only calls the countries the Customer declares, within the recipient's local calling hours, to contacts with permission and with available balance.

4. Limits of AI

• AI can make mistakes: it can produce inaccurate, incomplete or made-up information. The Customer must review what it publishes or sends.
• It does not guarantee commercial, sales or ad performance results.
• It does not give legal, medical, financial or tax advice, and does not generate political content.
• It is not used to decide on a person's credit, employment, housing, insurance, health or education. The Terms prohibit that use.
• It follows restrictions on regulated products according to the ad's location (for example, alcohol, medicines or financial services).
• Performance metrics come from the ad platform; AI does not calculate or invent them.

5. Synthetic content

• Generated images and videos carry metadata stating that the content was created with AI (IPTC "trainedAlgorithmicMedia"), and we do not strip it when processing them.
• When an ad targets the European Union or New York, a visible label saying the content is AI-generated is also added.
• No made-up testimonials or reviews: avatar templates are a "brand presenter" who speaks on behalf of the business, never an invented customer. A review is only quoted if it is real.
• No celebrities or "lookalikes" are generated, and real people are not imitated without their license.
• Cloned voices and avatars: only with the consent of the person whose voice or image it is, who can revoke it and request deletion.

6. Call and meeting recording

• Calls made through the platform's telephony are recorded, with a notice at the start of the call.
• If the person objects, the recording is discarded and not transcribed.
• In the European Union, the United Kingdom, Brazil, Colombia, Peru, Argentina, Chile and Mexico the notice includes the purpose, who the controller is (the business) and how to object.
• Meetings recorded on video-conferencing platforms are announced by that platform; ours only imports the recording from the Customer's mailbox.
• AI does not infer people's emotions during calls.

7. Automated decisions and human review

The platform does not make decisions with legal or similarly significant effects about a person. Ordering whom to contact or scoring a lead uses explainable rules. Anyone can request an explanation of the criteria and a human review (for example, under Article 20 of Brazil's LGPD or US state laws) from Privacy Rights.

8. Data and training

• We do not use Customer data to train our own models.
• AI providers are used through their paid API services and under their business terms. They are listed in the subprocessor list.
• If the Customer connects its own key for an AI provider, that processing is also governed by its contract with that provider.

9. Customer responsibility

• Review and approve content before publishing or sending it, and be responsible for it.
• Hold the consents needed to contact, record and use a person's voice or image.
• Keep the notices required by its contacts' locations turned on; it may add notices but not remove them.
• Comply with ad and messaging platform policies on AI-generated content.

10. How to report a problem

If an AI on the platform produced something harmful, misleading or that imitates a person, or did not disclose that it was an AI, write to {CORREO_LEGAL}. We review every report and, where appropriate, remove the content or fix the feature.