AI and GDPR: What a Small Business Actually Needs to Know

"Fine — but where does my data go?" It is the first question every business owner asks, and it is the right one. This guide answers it with facts rather than reassurances, so you can decide for yourself. AI and GDPR is not a reason to avoid the technology; it is a reason to set it up properly.

What actually happens when a customer talks to an AI chatbot

Think of it as a journey.

The customer types a message. That text leaves your website and travels to an AI provider — a company that runs the model producing the answer. The model reads it, generates a reply, and the reply travels back. The whole trip takes a second or two.

Three things matter about that journey: who receives the text, how long they keep it, and whether they use it to train their models. Everything else in this article follows from those three.

The right question is not whether AI is allowed — it is where the data goes

Using AI in your business, and charging customers for work you do with it, is entirely legal. That is not the risk.

The risk lies in what travels, to whom, and under what agreement. A chatbot answering "what time do you open?" carries almost no personal data. A chatbot to which someone types their name, phone number and a medical question carries a great deal. Same technology, completely different obligations.

So before you worry about the law, map what will actually pass through it. In most small businesses the honest answer is: far less than they feared.

What to demand from a provider

Three criteria reduce most of the risk:

1. A written data processing agreement (DPA). Serious providers offer one as standard. If there is not one, that alone is your answer.
2. A clear statement on training. Does the provider use your inputs to improve their models, and can you switch that off? For business use it should be off.
3. Where the servers are. Europe, or another jurisdiction with appropriate safeguards in place, is one situation. A provider offering neither a DPA nor clarity about location is another — and we avoid those. We are not going to name companies here; ask any supplier these three questions and judge the answers yourself.

What we do in practice

Three rules we apply to every agent we build:

The agent gets only the data it needs for its job. Not everything available — only what the task requires.

We never take access to a whole mailbox when drafting is enough. Our own email agent reads, understands and leaves a draft reply; it does not send, and it does not need broad permissions. That single design decision removes most of the exposure.

When genuine access to customer data is required, there is a written contract, because at that point we become a data processor under GDPR and the responsibilities have to be set down in writing.

A practical checklist before you switch on a chatbot

Five things, none of them difficult:

1. Tell your customers they are talking to AI. One line at the start of the conversation.
2. Do not put sensitive data through a chatbot — health, ID numbers, card details. Set that as a rule when you build it.
3. Keep the right to deletion. Know how a conversation gets removed if someone asks.
4. Get it in writing — what is stored, for how long, and where.
5. Decide who internally can read the conversations. Usually far fewer people than have access by default.

What the AI Act says about telling people

The principle is simple: when AI is talking to a human, the human should know.

It is not a burden. It is good practice, and in our experience it increases trust rather than reducing it — people are far more forgiving of a machine that says it is a machine than of one pretending to be a person and getting caught.