What the chatbot actually collects
List the data before you write the first prompt. A booking assistant on Zalo OA sees the customer's Zalo ID and display name, the messages, and anything the customer types: a phone number, a preferred date, a symptom, an address.
The list is usually longer than expected. Customers volunteer details the bot never asked for, such as a photo of a prescription or an ID card for a hotel check-in.
- Identifiers: Zalo or Messenger ID, display name, phone, email
- Conversation text and attachments
- Booking or order details written to your systems
- Technical logs: time, channel, errors
The rules: what we know and what your lawyer decides
Vietnam regulates personal data. Decree 13 set the earlier rules on personal data protection, and the Law on Personal Data Protection (Law No. 91) now governs the area. These texts cover consent, sensitive data and transfers abroad, among other things.
We do not restate the rules here, because their details and current status matter and change. Ask your lawyer three questions: what consent we need from chat users, which of our data counts as sensitive, and whether sending messages to an AI provider abroad needs extra steps.
Not legal advice. Check the current legal texts and their application to your business with a lawyer before launch.
Consent in a chat: how it looks in practice
In a chat, consent has to be short and visible. A common pattern is a first message that says an AI assistant answers, what it records and why, with a link to your privacy policy, and a clear way to talk to a person instead.
The system should store when and how the customer agreed. A line in the conversation log is often enough technically; your lawyer decides whether it is enough legally.
Where the data goes and how long it stays
Map the path of one message: Zalo or Messenger, your server, the AI model provider, your CRM or calendar, and backups. Each step has its own storage and its own deletion rule.
Keep only what the task needs. A booking assistant needs the date and contact; it rarely needs to keep the full chat history for years. Set retention in your own systems and check the AI provider's settings for storage and training use.
- Send the model only the fields it needs for the answer
- Do not paste ID numbers or full medical notes into prompts
- Store logs in your systems, with access limited to named staff
- Agree a retention period and delete on schedule
Health, children and money: extra care
Clinics, dental clinics and spas receive health details in chat. Language centers talk to parents about children. Real estate and finance see income and ID documents. In these cases the bot should collect less and hand over sooner.
A practical rule: the bot books and answers general questions; anything that looks like a diagnosis, a child's personal details or a financial document goes to a trained person. The exact consent and storage rules for these categories are settled with your lawyer before launch.
Deletion and customer requests
Customers may ask what you hold about them or ask you to delete it. Decide who receives such requests and how they find the data across chat logs, CRM and backups.
Build a simple procedure before launch: one inbox for requests, a search by phone or Zalo ID, and a record of what was deleted and when. It is much harder to add after a year of conversations.
A short checklist before launch
Use this list with your lawyer and your developer in the same meeting. Each line needs an owner and a decision.
- List of data fields the bot collects, including what customers may volunteer
- Consent text in Vietnamese and English, and where it is recorded
- AI provider, its region and its storage and training settings
- Retention period for chat logs and CRM records
- Topics the bot must hand over: health, children, complaints, payments
- Procedure for access and deletion requests
- Privacy policy updated to mention the chatbot