AI chatbots and the GDPR: what to check before going live
Informing visitors, sensitive data, retention, trackers, processors: the GDPR and AI Act points to check before you publish an AI chatbot on a client website.
Security and data6 min read
A chatbot on a website is a free-text box open to every visitor. People type whatever they like: a question about opening hours, but sometimes also their name, address, order number, or even details about their health. As soon as visitors can write freely, you should assume personal data will be processed.
For an agency that installs assistants for its clients, this is not a side issue. The client expects a minimum of rigour from you, and a rushed launch can expose them. This article offers a checklist of points to review before publishing an AI chatbot, based on the official texts. It is written from a European perspective, under the GDPR and the EU AI Act.
This article gives practical pointers, not legal advice. For a specific case, ask your client’s data protection officer or a lawyer.
1. Clarify who does what
The GDPR distinguishes the controller, who decides the purposes and means of processing, from the processor, who processes data on the controller’s behalf. A chatbot project often involves several parties: the business publishing the site, the agency configuring the assistant, the platform hosting it and the AI model providers.
Article 28 of the GDPR requires the relationship with a processor to be governed by a contract that sets out, among other things, the subject matter, duration, nature and purpose of the processing, the type of data and each party’s obligations.
To check:
- who is the controller for visitors’ conversations;
- which contracts govern the relationship between your client, your agency and the platform;
- which processors are involved, model providers included.
At blabla.bot, the precise split of responsibilities between the platform, the agency and the end client is still being defined. Ask for the available contractual documents before you commit to a client.
2. Inform visitors
Article 13 of the GDPR lists the information to provide when data is collected: the controller’s identity, purposes and legal basis, recipients, any transfers outside the European Union, retention period and data subjects’ rights.
In practice, the site’s privacy policy must mention the chatbot, and a link to that policy must be easy to find.
The EU regulation on artificial intelligence (the AI Act) now adds to this. Its Article 50 requires AI systems meant to interact directly with people to be designed so that those people know they are dealing with an AI, at the latest at the first interaction, unless it is obvious. According to the European Commission, the AI Act’s transparency rules apply from August 2026.
To check:
- the privacy policy mentions the chatbot and its purposes;
- visitors understand, as soon as the widget opens, that they are talking to an automated assistant. The welcome message is the ideal place to say so clearly.
The blabla.bot widget also shows a note reminding visitors that the assistant can make mistakes.
3. Limit data and head off sensitive data
The data minimisation principle (GDPR Article 5) requires data to be limited to what is necessary. In a free-text chatbot, you do not control what visitors type.
France’s data protection authority, the CNIL, recommends that when sensitive data is not expected but remains possible, a warning be shown before the chatbot is used and a purge be set up, immediate or regular. Sensitive data includes health, political opinions and religious beliefs.
To check:
- the welcome message asks visitors not to share sensitive information, especially in exposed sectors (health, social services, legal);
- risky topics are in the agent’s forbidden topics;
- the contact form only asks for what is needed. In blabla.bot it collects an e-mail address and a message, with a note that the e-mail is only used to reply.
4. Set retention periods
GDPR Article 5 also requires that data not be kept longer than necessary. For chatbots, the CNIL explains that the period depends on the purpose: a shopping-help conversation can be deleted at the end of the exchange, while a complaint may justify keeping it longer.
At blabla.bot, the retention period for conversations is not set yet. For PDF files added to the knowledge base, the default is to delete the file once it has been indexed.
To check: the retention period stated in the privacy policy matches what the platform actually does.
5. Check trackers
A chatbot may set cookies or use browser storage. The CNIL distinguishes two cases: a tracker set before the visitor activates the chatbot requires prior consent; a tracker set only once the visitor activates the chatbot, and only to make it work, can be considered strictly necessary.
In blabla.bot, the chat window is only created on the visitor’s first click, and the conversation memory is kept in that window’s storage so it carries over from page to page. Have this behaviour reviewed by whoever handles the site’s compliance, against its own cookie policy.
6. Know where processing happens
GDPR Articles 44 to 46 govern transfers of data outside the European Union: an adequacy decision by the Commission, or appropriate safeguards such as standard contractual clauses.
An AI chatbot relies on a model provider. In blabla.bot the agency picks each agent’s AI option: the default option uses a GPT model, and a Mistral option is offered as the sovereign choice, to be selected explicitly. The platform’s hosting region choices are still being defined.
To check: for the chosen option, where data is processed and on what basis any transfer rests. For a client who cares about this, the AI option can be changed at any time.
7. Keep a human in important decisions
The CNIL points out that a chatbot conversation without human involvement cannot, on its own, be the basis of an important decision about a person, such as turning down a loan or a job application. GDPR Article 22 governs such automated decisions.
A website assistant informs and guides. If it is used in a process that leads to a decision, plan for genuine human involvement.
8. Document it
GDPR Article 30 requires a record of processing activities, with a limited exemption for organisations with fewer than 250 employees. Add the chatbot to your client’s record: purpose, data categories, recipients, retention period, security measures.
The checklist
| Point | Question to ask |
|---|---|
| Roles | Who is the controller, which contracts are signed? |
| Information | Does the privacy policy mention the chatbot? |
| AI transparency | Do visitors know from the start they are talking to an AI? |
| Sensitive data | Is a warning shown? |
| Retention | Is the stated period the real one? |
| Trackers | Is storage only triggered on activation? |
| Transfers | Where does the chosen AI option process data? |
| Decisions | Is a human involved in important decisions? |
| Record | Is the processing in the record of activities? |
In short
Launching a compliant AI chatbot is mostly a matter of method: clarify roles, inform visitors, limit data, set retention periods, check trackers and know where processing happens. blabla.bot does not promise automatic compliance: compliance also depends on your client’s choices and on how the assistant is configured.
Common questions about security and data are gathered in the FAQ.
Sources
- Regulation (EU) 2016/679 (GDPR), Articles 5, 13, 22, 28, 30 and 44 to 46: https://eur-lex.europa.eu/eli/reg/2016/679/oj
- CNIL, “Chatbots : les conseils de la CNIL pour respecter les droits des personnes”, 19 February 2021 (in French): https://www.cnil.fr/fr/chatbots-les-conseils-de-la-cnil-pour-respecter-les-droits-des-personnes
- Regulation (EU) 2024/1689 (AI Act), Article 50: https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- European Commission, AI regulatory framework: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
Tags
- GDPR
- AI Act
- personal data
- compliance
On the site
Read next
Security and data
Mistral and sovereign AI: when to favour it for your clients
Announced EU hosting, GDPR, buyer requirements: when to offer a client the Mistral Small option, and what you should be careful not to promise them about it.
· 5 min read
The product
What an AI chat agent for a website is (and what it is not)
A website AI agent answers visitors from the client's own content, within limits set by the agency. Here is what it does well and what it does not do.
· 5 min read
Agencies and white label
White label: what must really disappear for the end client and their visitors
Real white label goes beyond swapping a logo. Here is what visitors and end clients should never see, and where blabla.bot stands on each point today.
· 5 min read