Can I put customer data into ChatGPT? GDPR compliance in practice

A customer writes in to complain about a late order. Whoever answers copies the email into ChatGPT on their phone and asks for a polite reply. The email carries the customer's name, phone number, delivery address and order number, and within seconds that information sits on another company's servers. Something like this probably already happens in your company, nearly always with good intentions and without anyone having decided it should.

In plain terms, this is what the General Data Protection Regulation (GDPR) means for that habit. This is general information and does not replace legal advice; when one of our projects needs legal support, we work with lawyers.

What happens to the data when someone pastes it into ChatGPT?

Under the GDPR, personal data is any information relating to an identified or identifiable person, such as a name, a customer number or an email address. Processing covers almost anything done with that data, and the regulation's definition includes, amongst other operations, use and disclosure by transmission. Pasting a customer's email into an AI tool is therefore processing of personal data, and when it happens at work it is usually your company that answers for it, as the controller.

From there, what becomes of the data depends mostly on the kind of account.

Personal account or company account: what changes

For personal ChatGPT accounts (Free, Plus and Pro), OpenAI's help centre says it may use the content of conversations to train its models. The user can turn this off in the account settings, though the decision is left to each person and the company has no way to control it. The same page warns that even with the option off, giving feedback on a response (thumbs up or thumbs down) allows the whole conversation to be used for training. These accounts also involve no contract between your company and OpenAI.

For OpenAI's business offerings, the published terms are these:

• By default, data from ChatGPT Business, ChatGPT Enterprise and the API is not used to train the models.
• OpenAI can sign a Data Processing Addendum (DPA) with customers of ChatGPT Business, ChatGPT Enterprise and the API.
• In ChatGPT Business, admins decide how long data is kept, and deleted conversations leave OpenAI's systems within 30 days, with the exceptions the page lists. On the API, OpenAI may keep inputs and outputs for up to 30 days to provide the service and detect abuse, also with exceptions.
• New ChatGPT Enterprise workspaces can store their content in Europe, and eligible API customers can have data processed in Europe for certain services.

In Microsoft 365, Microsoft states that prompts, responses and data Copilot accesses through Microsoft Graph are not used to train foundation models. Its data residency commitments are set out in Microsoft's Product Terms and Data Protection Addendum. For EU customers, Copilot runs within the EU Data Boundary; models from Anthropic, which an admin can switch on, are currently outside it. Copilot only shows each person content they already have access to, so the company's sharing permissions need to be set correctly.

In Google Workspace, Google says that, in qualifying editions, content used with Gemini is not reviewed by people or used to train generative AI models outside the company's domain without permission. Prompts typed by users count as customer data under Google's Cloud Data Processing Addendum.

We checked the vendors' pages on 6 November 2026 (the links are in the sources), and the safeguards they describe apply to business accounts only.

What the GDPR asks of you, in plain terms

For a company whose staff paste customer data into an AI tool, six points in the regulation matter most.

Article 6 requires a valid reason to process the data: processing is lawful only if at least one of the conditions the article lists applies, among them the person's consent, the performance of a contract with them, and the company's legitimate interests where these are not overridden by the person's rights. So it is worth asking whether the AI task fits the reason the company holds that data in the first place.

On the amount of data, Article 5(1)(c) allows only what is needed: data must be adequate, relevant and limited to what is necessary for the purpose. To improve the wording of a reply to a complaint, the AI rarely needs the customer's name or address.

Article 28 deals with whoever processes data on the company's behalf, which the regulation calls a processor. The company may only use processors that provide sufficient guarantees, and that relationship has to be governed by a contract, which is the job of the DPA the business plans offer.

Under Article 32, the company and the provider must put in place technical and organisational measures that ensure security appropriate to the risk, and the text gives pseudonymisation and encryption as examples.

Transfers outside the European Economic Area fall under Chapter V: if the data is processed in a third country, the transfer has to meet the conditions of that chapter, for instance an adequacy decision by the European Commission or appropriate safeguards such as the standard data protection clauses the Commission has adopted. Ask the provider where data is processed and look for the answer in the DPA.

Where processing, in particular using new technologies, is likely to result in a high risk to people's rights and freedoms, Article 35 requires the company to assess the impact before it starts. Whether your case falls within this is a question for a lawyer.

In Portugal, the authority that oversees compliance with the GDPR is the CNPD (Comissão Nacional de Proteção de Dados).

Four rules for your staff

1. Put a short rule in writing. One page is enough, saying what may and may not be pasted into an AI tool, with examples from your own staff's work, such as what to do with a customer email or with health and salary data.
2. Remove names and identifiers wherever the task allows. Before pasting, replace the name with 'the customer' and delete the tax number, phone number, address and order number; for many writing tasks, what is left is all the AI needs.
3. Give your staff business accounts, with the provider's DPA signed, and ask that work with customer data happens only there. If the company already pays for Microsoft 365 or Google Workspace, check what those accounts already allow before buying another tool.
4. Know which tools are in use. Ask your staff, without making it feel like an inquiry, which AI tools they use and for what. The answer is the starting point for the written rule and for deciding which accounts to pay for.

How we work on our projects

On Ontia Labs projects, client data stays in the company's own accounts or on its own server with open models, and the technology page explains both options. We sign a GDPR data-processing agreement before any build starts, and a person reviews what the tool produces before anything is sent.

In the workshop, you can apply these four rules to your team's actual tasks, with the people who do them.

After launch, through monthly support, we keep the tools running and answer your staff's questions, including those that come up about what may be pasted into an AI tool.