GDPR and AI customer service agents: what the regulation actually requires of a small business, the questions to ask a vendor, the settings and wording you control, and the parts people get wrong
GDPR is the question European customers ask first and the one most vendors answer vaguely. The regulation does not forbid using an AI to answer messages. It requires you to know what personal data is processed, by whom, where, for how long, and to be able to explain it and delete it. Those are answerable questions.
Quick answer
Using an AI agent to answer customer messages is permitted under GDPR when you have a lawful basis (usually legitimate interests or contract for responding to enquiries), tell customers that an assistant is involved and where their data goes (a line in your privacy notice and the agent's first message), keep only the data you need for as long as you need it, have a data processing agreement with the vendor that names its sub-processors (including the AI provider) and covers international transfers, and can respond to access and erasure requests. Before signing up, ask a vendor the eight questions below. This is general guidance, not legal advice; a privacy professional should review your setup.
Lawful basis and purpose
Replying to a message a customer sent you is processing their data to respond to their request, which most businesses rely on legitimate interests or steps prior to a contract for. Using the chat history to improve your FAQ is a compatible purpose. Using it for marketing is a separate purpose needing its own basis and, on WhatsApp, explicit opt-in; see consent rules for businesses. Write the purpose down in your records of processing.
Transparency
- Your privacy notice names the categories of data (contact details, message content), the purposes, the processors (the agent vendor and, through it, the AI provider), the retention period and the customer's rights.
- The agent discloses itself: "I'm {name}'s assistant" in the first reply. Disclosure rules by country are in AI disclosure laws and best practice.
- If a decision with legal or similar effect were made solely by the agent (it should not be; the agent answers questions and books), Article 22 would apply. Keep decisions with people.
Minimisation and retention
- Collect only what the service needs; the intake rules in intake forms on WhatsApp keep sensitive categories off the chat.
- Set a retention period for conversations and leads, and delete on schedule. ReplyKit lets you delete conversations and contacts, and deletes account data on closure.
- Special category data (health, religion, etc.) needs an explicit basis. A clinic's agent should collect the reason for a visit in general terms and send a secure form for history.
- Do not paste chat exports into knowledge bases; see training with past conversations.
Processors, sub-processors and transfers
The vendor is your processor and needs a data processing agreement with you. The AI model provider is a sub-processor; the vendor must name it and its location, and if data leaves the EU or UK, state the transfer mechanism (standard contractual clauses or an adequacy decision). WhatsApp itself is Meta, with its own terms. Ask where message data is stored and whether the AI provider retains or trains on it. ReplyKit's privacy policy names its providers and states that conversation content is not used to train models.
Rights requests
A customer can ask what you hold and ask you to delete it. You need a way to find their conversation and lead by phone number, export it, and delete it. The inbox search and delete functions cover this; the export is the conversation view and the lead's details. Answer within a month.
Eight questions for any vendor
- Will you sign a data processing agreement?
- Who are your sub-processors, including the AI model provider, and where are they?
- Where is message data stored, and what is the transfer mechanism if outside the EU or UK?
- Does the AI provider retain or train on our conversation content?
- What is the retention period, and can we delete conversations and contacts ourselves?
- How do we respond to an access or erasure request?
- Is the agent disclosed to customers by default?
- What security measures protect the data? The security side is in the security checklist.
Vague answers to any of these are a reason to look elsewhere.
Frequently asked questions
Is it legal under GDPR to use an AI agent for customer messages?
Yes, with a lawful basis, transparency, minimisation, a processing agreement with the vendor, and the ability to answer rights requests.
Do I have to tell customers they are talking to an AI?
Transparency requires it in practice, and some jurisdictions require it explicitly. Disclose in the first reply.
Does the AI provider train on my customers' messages?
Ask the vendor. ReplyKit's providers do not use conversation content to train models, as stated in its privacy policy.
Can I delete a customer's data if they ask?
You must be able to. ReplyKit lets you delete conversations and contacts from the dashboard.
What about health or other sensitive data?
Keep it off the chat; collect it through a secure form with its own basis.
Is this legal advice?
No. It is general guidance; a privacy professional should review your specific setup.
Clear answers to the GDPR questions
Disclosed by default, deletable by you, providers named, no training on your conversations. 7-day free trial, no card.
Sources and further reading
- ReplyKit Privacy Policy
- WhatsApp Business Messaging Policy · opt-in, opt-out and prohibited-use rules every business must follow
- OWASP Top 10 for Large Language Model Applications · prompt injection is listed as the first risk
This article is general information about data protection, not legal advice. Consult a qualified professional about your own obligations.