AI Usage Policy
In plain English
- Your AI answers from your own content - never mixed with other businesses' content, and never used to train AI models.
- Customers can always reach a human; the AI identifies itself and escalates anything sensitive.
- Keep your knowledge base current - outdated prices or hours mean outdated answers.
- You review AI-generated marketing content before it's published.
- Don't use the AI to deceive, impersonate, or send unsolicited marketing.
Summary for convenience; the full policy below governs.
1. Your content & our model providers
Your AI agent answers from the content you give it (menus, FAQs, price lists, policies, documents, your website). It is not trained on that content: it looks up the relevant parts when a customer asks. Your knowledge base and your customers' conversations are never combined with other businesses', and never used to train our own or anyone else's models. We use OpenAI (replies, speech and reading documents) and Google's Gemini API (images) through their business APIs, whose terms say they don't train their models on this data.
2. Transparency & identification
AI involvement is always identifiable: the dashboard labels each conversation as AI- or human-handled, and customers can ask for a human at any time. Our default templates introduce the agent as an AI assistant, in line with AI-transparency requirements including the EU AI Act's disclosure obligations.
3. Human escalation & high-risk topics
Every AI response passes a confidence check; low-confidence answers, requests for a human, and configured keywords route to your human queue. The AI never makes legally significant automated decisions about people. It must not be relied on for medical, legal, financial or other regulated advice - configure escalation so those enquiries reach a qualified human. For clinics: the AI may handle booking and general info, but symptom triage and urgent cases must route to clinical staff.
4. AI-generated marketing content
Flyers and captions are generated with Google Gemini and OpenAI models, and the platform enforces an approval step - you must review before publishing. Don't generate content depicting real, identifiable people without consent, content that infringes trademarks or copyright, or content that breaches the advertising policies of the channel where it's posted.
5. Accuracy & your responsibilities
AI quality depends on your knowledge base - keep it current. Review the knowledge-gap reports in Analytics to see where the AI lacked information. You remain responsible for your business's communications, including those sent by your AI agent.
6. Prohibited uses & contact
You may not use InfrAgent AI to generate or send illegal, deceptive or harmful content; impersonate people or businesses; surveil or profile unrelated to customer service; send unsolicited bulk marketing; mislead customers about whether they're talking to an AI; or violate Meta, Google or Twilio platform terms. Breaches may lead to suspension. Questions: support@infragentai.com.
© 2026 INFRAGENT AI LIMITED · Registered in England & Wales, company number 17291063. All rights reserved.