AI
AI Chatbots and Customer Data: A GDPR and KVKK Privacy Checklist for Messaging Teams
An AI agent on WhatsApp or Instagram moves customer messages through several companies. This checklist shows how to map that data, pick providers, set retention and handle consent under GDPR and Türkiye's KVKK.
MonoChat Team Updated: 8 min read
On this page
To make an AI chatbot GDPR and KVKK compliant, map every system that customer messages pass through, have a lawful basis and a clear privacy notice, sign a data processing agreement with each provider, use a valid transfer mechanism when data leaves the EU or Türkiye, keep personal data out of prompts where you can, set retention periods, and tell people they are talking to an AI. Compliance comes from how you deploy the chatbot, not from the chatbot itself.
The tricky part is that one WhatsApp reply can touch four or five companies: Meta, your messaging platform, an LLM provider, a vector database and the CRM the agent looks things up in. This checklist walks through each one. It is practical guidance for messaging teams, not legal advice, so confirm the details with your data protection officer or legal advisor.
Where customer data goes when an AI agent replies
Start by drawing the path of a single message. A typical flow looks like this:
| Step | System | Data involved | Your question |
|---|---|---|---|
| 1 | Messaging channel (WhatsApp, Instagram, Messenger) | Phone number or username, message content, media | Which terms and storage rules apply? |
| 2 | Messaging platform | Conversation history, contact profile, tags, notes | Who can access it, where is it stored, for how long? |
| 3 | Knowledge base and vector database | Your content; personal data only if you uploaded it | Does any personal data sit in the index? |
| 4 | LLM provider | Prompt: instructions, retrieved text, history, latest message | Is data retained, used for training, processed where? |
| 5 | Connected tools (CRM, booking, store) | Order data, appointments, account details | What can the agent read and write? |
| 6 | Human agents | Full conversation after handover | Who sees what, and is it logged? |
Every row is a place where personal data is processed and where you need a contract, a retention rule and access control. If you cannot fill in a row, that is your first task.
The AI chatbot privacy checklist
1. Pick a lawful basis and write it down
Under GDPR (Article 6), every processing purpose needs a legal basis. Answering a customer who wrote to you usually rests on contract or legitimate interest. Marketing messages usually rest on consent. Using conversations to improve your agent is a separate purpose, and you should assess it separately rather than assume the original basis covers it.
KVKK has its own list of processing conditions. If you serve both EU and Turkish customers, document the basis under each law.
2. Tell people about the AI and about their data
Your privacy notice should say that conversations may be handled by an AI system, which providers process the data, where, for how long, and how to reach a person. GDPR Articles 13 and 14 set out what to include, and KVKK has a parallel duty to inform.
In the EU, the AI Act’s transparency obligations apply from August 2, 2026. They require that people be informed when they interact with an AI system, unless it is obvious from the context. A short first message handles both the legal point and customer expectations:
Hi, I'm the Northwind Home virtual assistant, an AI. I can help with
orders, returns and delivery. Type "agent" any time to talk to our team.
Privacy notice: northwindhome.example/privacy3. Sign a data processing agreement with every processor
Your messaging platform, LLM provider, vector database host and any tool provider that handles personal data on your behalf is a processor. GDPR Article 28 requires a written contract that limits processing to your instructions, requires confidentiality and security, and covers sub-processors, deletion and audits. Keep a list of sub-processors and check it when providers change.
4. Check transfers outside the EU or Türkiye
Many LLM providers process data in other countries. Under GDPR Chapter V, transfers outside the EEA need an adequacy decision or appropriate safeguards, such as standard contractual clauses.
KVKK’s rules on transfers abroad changed in June 2024. Transfers now rely on an adequacy decision, appropriate safeguards such as the standard contracts published by the Personal Data Protection Board, or limited exceptions. Standard contracts must be notified to the Authority within five business days of signing. Check where each provider processes data and choose regions or providers that match your obligations.
On WhatsApp, Meta’s Cloud API offers local storage: you can choose where message data is stored at rest for each business phone number. Content may still be processed in Meta data centers internationally while in use, for up to 60 minutes on Cloud API, before it is kept only in the selected region.
5. Choose the LLM provider deliberately
The model provider sees everything you put in the prompt. Before you connect one, check:
- Training use. Does the provider use API data to train models, and can you opt out?
- Retention. How long are prompts and outputs kept, including for abuse monitoring?
- Region. Where is data processed and stored, and can you choose?
- Contract. Is a data processing agreement available for your plan?
- Self-hosting. Would an open-source model on your own servers suit sensitive use cases?
Using your own provider account (bring your own key) means the contract, region settings and retention options are yours, rather than inherited from a vendor’s account.
6. Minimize what goes into prompts and tools
Data minimization (GDPR Article 5) is the most effective privacy control for AI agents.
- Do not upload customer records, ID documents or internal notes to the knowledge base.
- Send the model only the history it needs, not the full contact profile.
- Give tools the narrowest access possible: an order-status tool needs an order number, not the whole customer database.
- Mask or avoid fields like national ID numbers, card numbers and passwords.
- Instruct the agent never to ask for sensitive data it does not need.
Agent rule: Never ask for card numbers, passwords or ID numbers in chat.
If a customer shares one, do not repeat it, and offer the secure payment
link or a call with our team instead.If your agent connects to business systems through protocols such as MCP, the same rule applies to every connected system. See MCP for customer messaging for the risks.
7. Treat health and other special data with extra care
Health, religion, biometric data and similar categories receive extra protection under GDPR Article 9 and KVKK. Clinics, pharmacies and insurers should limit what the AI handles, avoid medical advice by AI, and route clinical questions to qualified staff. Consider a data protection impact assessment (GDPR Article 35) before launching an AI agent that processes such data at scale.
8. Set retention for every store
Storage limitation (GDPR Article 5) means you keep personal data only as long as needed. Set and document a period for:
- conversation history in your messaging platform,
- logs at the LLM provider,
- exported transcripts and analytics,
- any personal data in a vector database or knowledge base.
Then make sure deletion actually happens, including in backups where possible.
9. Be ready for access and deletion requests
Customers can ask what you hold about them and ask you to delete it (GDPR Articles 15 and 17, with similar rights under KVKK). Know how to find a person’s conversations across every system in your data map, and how to delete them in each one.
10. Avoid decisions made only by AI
GDPR Article 22 restricts decisions based solely on automated processing that significantly affect people. Let the AI answer, collect and route, and keep approvals, refusals, credit decisions and exceptions with people. A reliable AI-to-human handoff is both a service and a compliance feature.
11. Secure access and keep logs
Limit who can read conversations, use role-based access, require multi-factor authentication, and keep audit logs of who did what. GDPR Article 32 requires security appropriate to the risk, and Article 33 sets a 72-hour deadline for notifying the supervisory authority of a personal data breach, where the breach poses a risk to people.
12. Follow the channel’s own rules
Meta’s WhatsApp Business Messaging Policy has its own requirements on opt-in, data use and AI. Our guide to the WhatsApp AI chatbot policy for 2026 covers what Meta allows, and WhatsApp opt-in covers consent for proactive messages.
Quick reference: GDPR and KVKK side by side
| Topic | GDPR | KVKK (Law No. 6698) |
|---|---|---|
| Inform users | Articles 13 and 14 | Duty to inform |
| Processor contracts | Article 28 | Controller jointly responsible for processor security (Article 12) |
| Transfers abroad | Chapter V: adequacy, standard contractual clauses | Adequacy, standard contracts notified to the Authority, limited exceptions |
| Special categories | Article 9 | Special categories of personal data |
| Retention | Storage limitation, Article 5 | Deletion, destruction or anonymization when purpose ends |
| Data subject rights | Access, erasure and others | Similar rights, including access and deletion |
How MonoChat helps you keep control of AI and data
MonoChat is built so the sensitive parts of the stack can stay under your control. Here is what the product provides, as described on our security and compliance page:
- Your own LLM keys. With custom LLMs you connect your own OpenAI, Anthropic, Google, Azure, AWS or other provider account, so you keep your own pricing and data agreements, and you decide which assistant uses which model.
- Self-hosted models. You can connect models running on your own servers through Ollama, or your own model API, when prompts should stay on your infrastructure.
- Your own vector database. VectorDB integration keeps knowledge-base vectors in the vector database you choose, with your own storage and retention policies, in the region or cloud you choose.
- Scoped knowledge. The AI knowledge base separates content into scopes with their own access control, so each agent retrieves only what it should.
- Controlled AI actions. AI function tools run with validation, permissions, allowlisted endpoints and logs, so an assistant can only do what you allow.
- Access control and logs. Role-based access, workspace isolation, audit logs and encrypted, workspace-scoped provider credentials.
- Official channel APIs. WhatsApp connects through the official WhatsApp Business Platform (Cloud API), and MonoChat is an official Meta Business Partner.
- Certifications and policy. MonoChat holds ISO 27001 and ISO 9001 certifications, and its privacy policy is prepared under KVKK (Law No. 6698). Requests to access, correct or delete personal data are answered within 30 days.
- On-premise option. The Enterprise plan offers on-prem deployment for stricter requirements.
You can evaluate the product on the free Basic plan (one agent, one channel) at app.monochat.ai, and read the full privacy policy before your review.
Sources
Put this into practice with MonoChat