ChatGPT Data Privacy: GDPR, Training, Business
· DigifyNow
Everything you type into ChatGPT goes to OpenAI's servers, where it is processed and stored as chat history. On personal accounts, OpenAI may use that content for training by default unless you opt out. Once personal data is involved, the GDPR requires companies to have a legal basis, a data processing agreement and a review of the transfer to the US. So personal data belongs only in Business or API access under a contract, never in a personal account.
What happens to the data you enter into ChatGPT?
ChatGPT doesn't run on your computer. Every prompt, along with any files you upload, is sent to OpenAI's servers. The model processes it there and sends the answer back. Your history stays saved in the account until you delete it. For data privacy, that leaves three questions that matter: who gets to see the data, what it is used for, and where it is processed?
The sticking point is training. Google's AI overview for this search sums up the current state like this: on personal accounts, OpenAI uses prompts and chat histories by default to train future models. Once a model has been trained, individual pieces of data can't be pulled back out of it. The same overview names the second point too: processing takes place on servers in the US. For companies, that means the GDPR rules on transfers to third countries apply.
If you don't want your data going to someone else's server at all, there is only one way: run the model yourself. How that works and where the limits are is covered in our article on running AI locally, including hardware and limits.

Personal account, Business or API: how do they differ on privacy?
Depending on how you access it, "ChatGPT" comes with very different contracts behind it. For data privacy, that matters more than the model running underneath. The table lines up the same properties for each type of access and is based on OpenAI's documentation. The rules on how API data is handled are in the developer documentation at https://developers.openai.com. The provider changes contracts and default settings from time to time, so check again before you sign up.
| Access | Training on your inputs | Data processing agreement (DPA) | Central control by the company |
|---|---|---|---|
| Personal account (free or paid individual plan) | Yes by default, user can opt out | Not available as a company contract | None: each employee sets it up themselves |
| ChatGPT Business or Enterprise | No by default | Offered by OpenAI to business customers | Yes: user management in the workspace |
| API (connected to your own software) | No by default | Offered by OpenAI to business customers | Yes: your own application decides which data gets sent at all |
What a Business workspace costs and which GDPR questions are still open afterwards is covered in our article on ChatGPT Business pricing and choosing a plan. For how another provider handles the same questions, see the comparison Claude vs ChatGPT.
ChatGPT data privacy at work: what the GDPR requires
Using ChatGPT at work isn't prohibited. But as soon as personal data ends up in a prompt, such as names, email addresses, customer numbers or a résumé, the GDPR obligations kick in. At that point the company is responsible, not the individual employee. These points need to be settled before you roll it out:
- Legal basis: for what purpose is the data processed, and what is that based on, for example performance of a contract or legitimate interest?
- Data processing agreement with the provider: as long as there isn't one, personal data has no place in the tool.
- Transfer to the US: on what basis does the data go there, for example an adequacy decision by the European Commission or standard contractual clauses?
- Special categories of data such as health information, trade union membership or religious beliefs: the GDPR sets much tighter limits for these.
- Documentation: add an entry to your record of processing activities and update the privacy notices for customers and employees.
- Data protection impact assessment if the processing is likely to involve a high risk, for example when applicants or employees are being evaluated.
- A written usage policy: which access is allowed, which data may go in, and who to ask when in doubt.
The fact that a tool offers a DPA doesn't make your use of it lawful. That depends on what data your people actually enter and for what purpose. The case-by-case assessment is a job for your data protection officer or legal counsel.

What data should you never put into ChatGPT?
For personal accounts and any other access without a DPA, the rule is: anything that identifies a person or is a trade secret stays out. Typical cases from everyday office work:
- Customer emails pasted in full, with name, signature and phone number
- Résumés, employment references and payroll lists
- Sick notes or health-related notes, which the GDPR specially protects
- Contracts with terms, quotes with customer pricing, unpublished figures
- Login credentials, API keys and internal links with a token in the URL
Placeholders like "[Customer A]" instead of the real name help, but only up to a point. If the context still reveals who it's about, for instance because location, job role and incident are mentioned together, the text is only pseudonymized. Under the GDPR, pseudonymized data still counts as personal data. A text is only anonymous once nobody can work out who it's about, even with additional knowledge. The name is rarely the only identifier: a customer number, a case reference or an unusual job title can give a person away too.
Turning off training: what the setting does and doesn't do
In personal ChatGPT accounts, you can opt out in the settings so your chats aren't used for model training. What the option is called and where you find it depends on the app, the version and the interface. There is also a chat mode whose conversations don't show up in your history. Both apply only to future prompts and only to the account where you set them.
Here's what the setting doesn't change: your prompts still go to OpenAI's servers and are processed there. So the transfer to the US remains, and the toggle doesn't replace a DPA either. On top of that, nobody in the company can verify whether several employees have actually opted out in their personal accounts. For companies, the toggle is therefore not a solution, at most a stopgap for individuals. If personal data is processed at work, you need access that the company controls centrally.

How reliable is the information ChatGPT gives you?
ChatGPT generates text by calculating the most likely continuation, piece by piece. It doesn't look anything up in a verified database while doing so. That produces fluently worded answers that are factually wrong, complete with made-up sources, legal sections or figures. With web search or attached documents this happens less often, but the problem doesn't go away entirely. How system prompts, knowledge retrieval and hallucinations fit together is explained in our article on how an AI chatbot works.
For data privacy, this becomes important as soon as people are involved. One of the GDPR's principles is that personal data must be accurate. If ChatGPT summarizes an application incorrectly or attributes something to a customer that isn't true, and that summary ends up in a file, the company is processing inaccurate data. In practice, the rule is: a human checks any output about people against the original before it is stored or passed on.
What the EU AI Act adds when you use ChatGPT
Alongside the GDPR, the European AI Act applies, now in a version adjusted by an amending regulation. For companies using ChatGPT, two points matter most. First, deployers of AI systems take measures to support the AI literacy of their staff. They are not required to guarantee a particular level of literacy for anyone. Training on data rules and on the limits of the answers is one such measure, for example AI coaching for employees. Second, transparency obligations for AI-generated content now apply. They are aimed mainly at providers of systems that generate synthetic text, images, audio or video, meaning OpenAI rather than the company using ChatGPT. If you publish texts created with ChatGPT, you should still have someone check which labeling obligations apply in your case. The AI Act doesn't replace the GDPR: a prompt containing customer data has to meet both sets of rules.

Why company ChatGPT policies fail
- A blanket ban with no alternative: people then use their personal account on their phone. The data still flows, only now nobody notices.
- A policy without training: a PDF on the intranet doesn't answer whether a customer email with the name crossed out is already safe. Only examples from your own day-to-day work settle that.
- Business access purchased, personal accounts still allowed: then you have two data flows, and only one of them is governed.
- Overlooking connected data sources: once ChatGPT is linked to a drive, a mailbox or a CRM, the tool reads more than just the prompt. Those connections therefore need to be approved.
- No point of contact: if nobody decides the borderline cases, everyone decides for themselves.
If instead of the general chat you need an assistant with clearly defined data sources and permissions, you'll find an overview in our article on introducing an AI assistant in your company.
When outside help is worth it
Outside help pays off when ChatGPT is supposed to work with real company data, such as customer inquiries, documents or a CRM. The same goes when several departments need a shared policy and training. If only a few individuals draft texts without personal data, a short internal rule is usually enough. For a conversation, you should know which tasks the AI is meant to take on, what data comes up along the way, which accounts are already in use and who is responsible for data protection on your side. DigifyNow introduces AI assistants and automation and trains teams. The legal assessment stays with your data protection officer. How the free assessment of your AI project works is explained on the contact page, and you'll find our services at a glance on the services page.

Frequently asked questions
What is data privacy, in a nutshell?
Data privacy governs who may process information about an identifiable person, for what purpose and for how long. In the EU, this is mainly done through the GDPR. It requires a legal basis, purpose limitation, data minimization and accurate data. It isn't the same as IT security: IT security protects data from unauthorized access, while data privacy determines whether the processing is allowed in the first place.
How do I stop ChatGPT from using my data for training?
In personal accounts, you can opt out in the settings so your chats aren't used for model training. Exactly where the option sits depends on the app version. There is also a chat mode without history. Your data is still processed on OpenAI's servers, though. According to OpenAI, Business, Enterprise and API access don't train on your inputs by default.
Can I enter customer data into ChatGPT?
Not into a personal account, because there is no data processing agreement there and training is switched on by default. In a Business or API account with a DPA it can be permissible, provided the legal basis, the transfer to the US and the information obligations are settled. For health data and other special categories, the GDPR sets much tighter limits than for a name or an address.
What changes does the new data protection law bring from 2026?
For AI tools like ChatGPT, the changes come from the European AI Act, not from a new GDPR. It now applies in an amended version. The transparency obligations for AI-generated content already apply and mainly affect providers like OpenAI. Companies using AI support the AI literacy of their staff. The stricter rules for high-risk systems only kick in later and cover only certain use cases.