Which ChatGPT plans you can use for company data
Start with the contract, because Article 28 GDPR requires one whenever a provider processes personal data on your behalf. On the personal plans there is nothing to sign. Under OpenAI's European privacy policy, OpenAI Ireland Limited is the controller for those accounts, so your company has no processor relationship with it at all. The business offerings work differently: ChatGPT Business, ChatGPT Enterprise and the API come with a data processing agreement, per OpenAI's enterprise privacy page, which links to the signing form.
| Plan | Data processing agreement | Inputs used for training by default | Storage in Europe offered |
|---|
| Free, Plus, Pro | No | Yes, can be switched off | Not listed |
| Go | No | Not stated on the help pages | Not listed |
| ChatGPT Business | Yes | No | Not listed |
| ChatGPT Enterprise | Yes | No | Yes, for new customers |
| API | Yes | No | Yes, for eligible customers |
Sources for the table, all read on 16 October 2026: the enterprise privacy page, the help article How your data is used to improve model performance and the help article on data residency for ChatGPT.
That turns the head of sales' Plus account into a problem the day his team pastes in a customer email. The data reaches OpenAI, and your company has no Article 28 contract covering it. If you want to know how much of that already goes on in your firm, the piece on shadow AI shows how to find out without starting a witch hunt.
File OpenAI's data processing addendum in the version in force since 1 January 2026, and read two clauses before you do. One fixes the counterparty: customers based in the EEA or Switzerland contract with OpenAI Ireland Ltd. The other covers sub-processors, which OpenAI keeps on a public list. You are notified of new ones and have 30 days from that notice to object.
How to use ChatGPT in a GDPR-compliant way
The contract covers OpenAI's side. These duties stay with your company, and none of them takes long. In Germany, the DSK, the joint body of the supervisory authorities, set out its view in the guidance "Künstliche Intelligenz und Datenschutz" of 6 May 2024. I cite it by its margin numbers.
A data protection officer at the table. Most firms in this size range already have one, because section 38 BDSG, Germany's Federal Data Protection Act, requires it once at least 20 people regularly work with automated processing of personal data. The same section requires one regardless of headcount if your processing needs a DPIA. The DSK wants data protection officers involved whenever decisions about AI tools are prepared or made, in margin number 31.
Company accounts, set up properly. Employers should provide the devices and accounts for work use, so that staff don't have to fall back on private ones. That is margin number 41. Margin number 43 adds that training and history settings should be chosen with data protection in mind when the account is first set up.
A short risk check, and a DPIA if it says so. OpenAI promises reasonable help with DPIAs in section 2.6 of its addendum, which tells you it expects customers to do them. Article 35(1) GDPR requires a data protection impact assessment where processing, especially with new technologies, is likely to result in a high risk to people. With AI, the DSK expects that often, in margin number 39, and points to its list of processing that always needs one. Write down the preliminary check even if it ends in no.
An entry in the record of processing. Firms under 250 staff often assume Article 30 GDPR doesn't concern them. Paragraph 5 does exempt them, but the exemption falls away once processing is more than occasional, carries a risk or touches special category data, and a tool people open every day is more than occasional as I read it. The record then lists the purpose, the categories of data and people, recipients, transfers to third countries and erasure periods. Expect this rule to move. Negotiators for Parliament and Council agreed a change on 26 June 2026 in procedure 2025/0130(COD), and Parliament's plenary vote is expected on 23 November 2026. Until the amendment is adopted and in force, the current paragraph 5 applies.
Telling people. Customers whose data goes into prompts have to be informed, says the DSK in margin number 48, so check that your privacy notice covers it. Staff are entitled under Article 13 GDPR to know the purposes and legal basis of processing their usage data, who receives it and how long it is kept. On ChatGPT Business, OpenAI says workspace admins can view, export and delete members' conversations. Say that plainly in the staff notice.
The works council, if you have one. In margin number 37 the DSK suggests a works agreement between management and works council, whether or not the tool processes personal data, and notes that introducing an AI application will in some cases trigger co-determination rights. Admin access to every member's conversations is the sort of feature a works council will ask about. What that means for your timeline is in the piece on AI and works councils.
A written rule for staff. Deleting a name often leaves a prompt identifiable, the DSK warns in margin number 49, because the context can still point to the person. Hence its advice in margin number 36 to issue documented instructions with concrete examples of allowed and forbidden uses. Ten clauses you can copy are in the AI usage policy template.
Data location and transfers outside the EEA
If EU storage is a hard requirement, ChatGPT Business won't meet it, because OpenAI's data residency article doesn't list it. Storage at rest in Europe, meaning the EEA plus Switzerland, is open to eligible API customers and to new ChatGPT Enterprise and Edu customers. Even then some data can sit outside the region: logins, billing details, workspace metadata, and whatever passes through web search or connected apps. Enterprise and Edu customers who also want the model computation in Europe can add inference residency.
For the United States there is an adequacy decision, Implementing Decision (EU) 2023/1795 on the EU-US Data Privacy Framework. Per the European Commission's overview of adequacy decisions, it covers only US organisations taking part in the framework, and participants are searchable on the Data Privacy Framework list. OpenAI's addendum keeps both routes open. Section 4.1 says OpenAI Ireland passes EEA data to affiliates or third parties outside the EEA on the basis of standard contractual clauses or an adequacy decision under Article 45 GDPR. Your record of processing names the third country and the transfer basis.
Once you connect ChatGPT to your file share, folder permissions decide what it can read, and residency doesn't extend to connected apps. Treat that connection as a project of its own. The piece on an AI assistant for company documents covers the groundwork. If you're rolling out Microsoft 365 Copilot instead, the same permissions problem sits in SharePoint, OneDrive and Teams, and the Microsoft 365 Copilot rollout checklist walks through it.
Settings to check on day one
The hole in the personal opt-out. On the personal plans, rating an answer with thumbs up or down lets OpenAI use the whole conversation for training, even after the user has opted out. The opt-out itself is the "Improve the model for everyone" switch under Settings, Data controls, or "Do not train on my content" in OpenAI's Privacy Portal. Without it, OpenAI may train on those conversations. Temporary chats aren't used for training, though OpenAI's data controls FAQ says they may be kept for up to 30 days. None of these switches supplies the missing contract, which is why personal plans stay off the list for company data.
Training on business plans. Here the default works for you: Business and Enterprise content stays out of training, thumbs ratings included. If you'd rather staff didn't send ratings to OpenAI at all, an admin can turn response feedback off for the whole workspace.
Retention. Choose a period that matches what people use the tool for and write it into the record of processing. On Business, workspace admins set it. OpenAI's own deletion limit sits underneath: deleted and unsaved conversations leave its systems within 30 days, it says, unless the law requires longer or OpenAI considers it reasonably necessary to protect its services.
Output about real people. Have a person check anything ChatGPT writes about a named customer or employee before it goes into a file or an email. The EU data protection authorities' ChatGPT taskforce report of 23 May 2024 explains why. The model can produce biased or made-up output, and users tend to take it as accurate anyway.
Written out, the whole list fits on a few pages. Settle the plan first, because it fixes the contract and the training default before anyone has typed a word. If you plan to put your own chatbot on your website instead, use the chatbot GDPR test.
Which of these points is holding up your rollout? Send it to me. The legal sign-off stays with your data protection officer whatever you decide. The setup work is what a caadoo subscription covers: I configure the workspace's training, retention and feedback settings, draft the entry for your record of processing and introduce the tool to your staff.
Frequently asked questions
Is ChatGPT Plus GDPR compliant for company use?
Not for company data, in my view, because OpenAI offers no data processing agreement for Plus. On the personal plans OpenAI acts as a controller under its own privacy policy, and conversations may be used for training unless the user opts out. Use ChatGPT Business or Enterprise for work involving personal data, and note in your records which plan you chose and why.
Does OpenAI sign a data processing agreement?
Yes, for ChatGPT Business, ChatGPT Enterprise and the API, according to OpenAI's enterprise privacy page, which links to the form. For customers in the EEA or Switzerland the contracting party is OpenAI Ireland Ltd. The current version took effect on 1 January 2026.
Where does ChatGPT store data?
Without a residency option, possibly outside the EEA. Section 4.1 of OpenAI's data processing addendum lets data leave the EEA under standard contractual clauses or an EU adequacy decision. Storage in the EEA and Switzerland is available to new Enterprise and Edu customers and eligible API customers, and OpenAI's data residency article doesn't list ChatGPT Business among them.
Do I need a DPIA to use ChatGPT?
Often, according to Germany's DSK, which expects AI use to carry a high risk in many cases. Article 35 GDPR requires a DPIA where processing is likely to result in a high risk to people. Do a documented preliminary check first. If it says yes, section 38 BDSG also requires a data protection officer regardless of headcount.