AI Disclosure

The business's AI assistant can:

• Answer questions about the business's services, prices, opening hours and location — drawing only on the FAQ and price list the business itself has entered
• Check which appointment slots are free and suggest them
• Book, reschedule or cancel an appointment, and update a contact's details, where the business has enabled those actions
• Hand a conversation over to business staff, and close one that is finished
• Tag and summarise conversations so staff can pick up context quickly

Every action it takes is recorded in the business's audit log with its inputs and results, and any booking it makes is re-validated inside the writing transaction so it cannot double-book a slot that filled while it was replying.

• No medical diagnosis or treatment recommendation
• No advice about medication, dosage or side effects
• No interpretation of symptoms, test results or images
• No access to clinical records — Medivis holds none, so there is nothing clinical for the AI to read or write
• No action outside the tools the business has switched on. It cannot message a contact who has opted out, and it cannot reach another business's data

Several layers sit between a contact's message and an AI reply.

Emergency blocker (deterministic)

Before any model is called, the message is checked against a list of emergency phrases in English and Malay — chest pain, difficulty breathing, severe bleeding, loss of consciousness, seizure, stroke signs, and expressions of suicidal intent. A match short-circuits the AI entirely: the contact is told to call 999 or go to the nearest hospital, and the conversation goes to business staff.

This gate is deliberately narrow. It covers content where a wrong reply costs more than a lost booking, and nothing else — ordinary symptom words are what contacts message a business about, and blocking those would leave real questions unanswered. How the assistant responds to an everyday health question is governed by its instructions, which forbid diagnosis and medication advice and require it to hand over instead.

Business-configured keywords

Each business can add its own blocking terms — drug interactions, overdose, severe allergy — and a match there also stops the AI and brings in staff.

Identifier scrubbing

Before message text is sent to the AI provider, identity card numbers, phone numbers and email addresses are detected and replaced with placeholders. Only scrubbed text leaves our servers.

Handover to a human

The conversation is passed to business staff when:

• The emergency blocker or a business keyword fires
• The contact asks to speak to a person, complains, or asks for a refund
• A medical question is pressed, is urgent, or clearly wants a clinical answer — the assistant is instructed to hand over rather than attempt one
• Business staff take the conversation over themselves

Handover pauses the AI on that conversation until staff release it.

Disclaimer on every reply

Where the business has configured one, its disclaimer is appended to every AI-generated message the contact receives.

• Ask for a human: a contact can ask to speak to a person at any point and the conversation is handed to business staff
• Staff override: business staff can disable the AI for an individual conversation, or for the branch entirely
• Transparency: the business's configured disclaimer identifies AI replies as such
• Opt out of messaging: replying STOP or BERHENTI stops marketing messages immediately

• Provider: OpenAI API (United States)
• A Data Processing Agreement (DPA) is in place with OpenAI
• Identifiers are scrubbed from message text before transmission
• Contact data is not used to train OpenAI models

It also only knows what the business has told it. If the FAQ or price list is out of date, so is the answer — keeping them current is the business's responsibility. And WhatsApp is not an emergency channel: a message may go unread for hours. For a medical emergency, call 999.