Can you use ChatGPT with customer data?

Your team is probably already pasting customer emails, CVs and support tickets into AI chatbots. Under the GDPR, each of those prompts is processing personal data, and with most cloud AI it also means sending that data outside the EU. This guide explains what the law asks of you, where businesses usually slip up, and how an AI that runs in your own office avoids the transfer question.

Updated November 3, 2026 · 9 min read

What counts as personal data in a prompt?

Under the GDPR, personal data is any information about a person who can be identified, directly or indirectly. In everyday AI use, that includes far more than names:

• Customer emails and support tickets you ask the AI to summarise or answer
• CVs and cover letters you ask it to compare
• Invoices, contracts and order lists
• Meeting notes and call transcripts that mention clients or staff
• Photos of documents, ID cards or handwritten forms

Deleting the name often isn’t enough. If the rest of the text (a job title, a town, an order number) still points to one person, it’s still personal data. Pseudonymised data, where you swap names for codes but keep the key, also still counts.

What the GDPR asks when you use a cloud AI

When your company decides to put personal data into an AI service, you are the controller and the AI company is your processor. That brings a set of duties:

1. A lawful basis (Article 6). You need a legal reason to use the data, and the new use has to fit the purpose you collected it for. Data a customer gave you to fix their order doesn’t automatically cover any AI experiment.
2. A data processing agreement (Article 28). The GDPR requires a contract with every processor, covering what it may do with the data, security, and deletion. Business plans of the big AI services usually offer one. Free and personal accounts don’t give your company one.
3. A basis for the transfer (Chapter V). If the provider processes data outside the European Economic Area, the transfer needs a legal basis: an adequacy decision for that country (for the United States, that only covers companies certified under the EU–US Data Privacy Framework) or safeguards such as standard contractual clauses, usually with a written assessment of the risks.
4. Transparency (Articles 13 and 14). Your privacy notice should tell people that their data may be shared with an AI provider, and whether it leaves the EU.
5. Extra care for sensitive data (Article 9). Health, religion, ethnic origin, biometric data and similar categories are only allowed under narrow conditions. Feeding them to a new technology can also require a data protection impact assessment (Article 35) before you start.
6. People’s rights. Customers can ask what you hold about them, or ask you to delete it. That’s harder to answer when copies sit in an outside provider’s systems.

Where businesses slip up

Staff using personal AI accounts for work

This is the most common problem, and often nobody decided it. Someone signs up for a free chatbot with their own email and starts pasting in customer messages. There is no contract between your company and the provider, you can’t see or delete what was shared, and personal plans can use chats to improve the provider’s models unless the user switches that off. In 2023, Samsung restricted staff use of generative AI after employees pasted internal source code into ChatGPT. Customer data is the same risk, with a legal duty attached.

Thinking “EU data storage” closes the question

Some providers now let business customers keep data in Europe, usually on their higher plans. That helps, but read what it covers: where data is stored, where the model actually runs, and where support and safety copies go can all differ. If the provider is a US company, it is also subject to US law wherever its servers are.

Assuming the rules for US transfers won’t change

The EU’s top court struck down the two earlier deals for sending personal data to the US: Safe Harbor in 2015 and Privacy Shield in 2020. The current EU–US Data Privacy Framework survived its first court challenge in 2025, but it is being challenged again. If it falls, transfers that rely on it need another basis, quickly.

Forgetting that regulators already look at AI

Breaking the transfer rules can bring fines of up to €20 million or 4% of worldwide annual turnover, whichever is higher. Data protection authorities have acted on AI services before: Italy’s regulator temporarily blocked ChatGPT in 2023 and fined OpenAI €15 million in 2024.

The local option: keep the AI in your office

Open AI models are now good enough for most office work: drafting replies, summarising documents, rewriting text, answering questions about a file. And they can run on a normal computer. When the model runs on a Mac or PC that you own, in your own office:

• No AI provider sees the prompts. There is no outside processor for the conversations, and nothing is used to train someone else’s model.
• No transfer abroad. The data stays on your premises and in your country, so the Chapter V transfer rules don’t come into play for your prompts.
• Deletion is in your hands. Chats are stored on your own devices, so you can find and delete them when someone asks.

What it doesn’t change: the GDPR still applies to you. You still need a lawful basis, you still need to keep the computer secure (disk encryption, a password, limits on who can use it), and the processing still belongs in your records. A local AI removes the outside company and the border, not the rules.

How PetekMind handles this

PetekMind is built for exactly this setup. PetekMind Hub runs the AI model on one Mac or Windows PC in your office, and your team uses it from the PetekMind app on their Android phones and iPhones, over your office Wi-Fi.

• Prompts and answers stay on your hardware. Messages go from the phone to your computer and back. They are never sent to a cloud AI or a server abroad.
• You decide who uses it. Only devices you approve can connect, and each one belongs to a named person. You can pause someone, limit them to certain models, or remove a device at once.
• You can see usage per person. The hub keeps a local log of who used it and how much, on your computer.
• Nothing goes online by default. Model downloads contain no personal data. Web search is optional: when someone turns it on for a message, only the search itself goes out to the internet, so tell your team not to search for personal details.

If you’d rather set up the engine yourself, our guide to using local AI from your phone compares the options, including plain Ollama.

Cloud AI or local AI: quick comparison

Checklist before your team uses AI with customer data

• Write down which AI tools are allowed for work, and for which kinds of data.
• Ban personal AI accounts for anything that contains customer or staff data.
• For a cloud AI: use a business plan, sign the data processing agreement, check the transfer basis, make sure training is off, and update your privacy notice.
• For health or other sensitive data: talk to your data protection officer about an impact assessment first.
• For a local AI: approve devices one by one, encrypt the computer’s disk, and add it to your records.
• Tell your team what not to paste, with a few concrete examples from your own work.

Doing business in Turkey? Turkey’s data protection law has its own rules for sending data abroad. Read our KVKK guide (in Turkish).