Skip to content
Floskool

Chapter 1 · The checklist

The 8 questions to ask your provider before choosing

Free lesson 9 minGroup Flopia

Follow this course in your space: your progress is saved, and the community answers your questions.

Follow for free

A good part of your compliance depends on a tool you didn't build. The AI disclosure, where the data is hosted, how long it's kept, how it's deleted: the chatbot provider handles all of that. Your job is to ask the right questions before you choose, and to keep the answers in writing.

Here are the eight questions, what a good answer looks like, and the red flags.

1. Where is the data hosted?

Why: your customers' conversations are stored somewhere. Hosting in the European Union makes things much simpler under the GDPR.

Good answer: a specific country, the name of the hosting company, written in a document.

Red flag: "in the cloud," "our servers are secure," with no country.

2. Are conversations used to train an AI model?

Why: if your customers' messages are used to improve a model, they leave the relationship between you and your customer.

Good answer: "no," written in the terms or the contract, or a clear option to turn it off.

Red flag: a vague answer, or an option buried in a menu that you have to remember to uncheck.

3. How long are conversations kept?

Why: the CNIL advises keeping data only as long as needed, and telling people that period. You can only disclose a period if the tool sticks to it.

Good answer: a specific period, ideally one you can set yourself, with automatic deletion.

Red flag: "indefinitely," or "as long as your account is active."

4. Which subprocessors are involved?

Why: an AI chatbot often relies on other companies (hosting provider, AI model developer, email service). You need to be able to name them in your privacy policy.

Good answer: a list of names, with each one's country.

Red flag: "our technical partners," with no names.

5. Can I delete a conversation on request?

Why: a person may ask you to delete their conversations. And when someone pastes their credit card into the window, you need to be able to delete that conversation.

Good answer: yes, from your dashboard, conversation by conversation, with a stated time frame for actual deletion.

Red flag: "email support," with no time frame.

6. Is the AI disclosure built in from the first message?

Why: Article 50 of the AI Act, applicable since August 2, 2026, requires that people know they're talking to an AI, at the latest at the first interaction. The obligation falls mainly on the provider: it should be able to tell you how it complies.

Good answer: the disclosure is shown by default, and stays in place even if you customize the welcome message.

Red flag: "you can add it yourself if you want," or a human name and avatar offered by default.

7. How does the handoff to a human work?

Why: a chatbot with no way out to a real person frustrates people, and the CNIL points out that a conversation without human involvement cannot on its own be the basis for an important decision (lesson 5).

Good answer: a clear handoff (email, notification, you taking over the conversation), with the conversation passed along so the customer doesn't have to repeat everything.

Red flag: no handoff, or a handoff to an inbox nobody reads in the evening or on weekends, with no message warning the customer about the wait.

8. Do I have access to the conversation log?

Why: to fix the chatbot's mistakes, respond to a customer's access request, or understand a complaint, you need to be able to reread conversations, for the retention period you chose.

Good answer: a log you can browse and export, deleted automatically at the end of the period.

Red flag: no access, or a log that's kept forever.

The message to send, ready to copy

Hello,

Before choosing your tool for my website, I need written answers to
the following questions:

1. In which country and with which hosting company are conversations stored?
2. Are my visitors' messages used to train an AI model?
   If so, how do I turn that off?
3. How long are conversations kept? Can I choose that period, with
   automatic deletion?
4. Which subprocessors are involved (hosting, AI model, others), and in
   which countries?
5. Can I delete a specific conversation, and how quickly is it
   actually erased?
6. Is the notice saying it's an AI shown from the first message, in
   line with Article 50 of the AI Act?
7. How does the handoff to a human work?
8. Do I have access to the conversation log, and can I export it?

Could you also tell me where to find your privacy policy and, if
applicable, your data processing agreement?

Thank you,
[First and last name, business]

The comparison table

If you're torn between several tools, note their answers side by side:

QuestionTool ATool B
1. Hosting
2. Training
3. Retention
4. Subprocessors
5. Deletion
6. AI disclosure
7. Human handoff
8. Log

A single vague answer on questions 2, 3 or 6 is enough to rule a tool out, whatever its other strengths.

Your turn

  1. Send the message to your current provider, or to the two or three you're comparing.
  2. File the written answers in a "Chatbot and GDPR" folder.
  3. Carry them over to the privacy policy paragraph from lesson 2: hosting, subprocessors, training, retention, cookie.