An IT lead at a regional law firm has thirty people on ChatGPT Business. It works, and nobody has asked for more. Then the data protection officer sends two questions for the record of processing: how long do conversations stay in the workspace, and can the firm retrieve them if a client complains or there is a breach?
At a university across town, a quote for ChatGPT Edu is waiting for sign-off. The question there runs the other way: what does the institution get for the extra cost?
Neither question is about training, where the three business plans are identical. Both are about administration, retention and storage, and that is where Business, Enterprise and Edu go separate ways.
#How Business, Enterprise and Edu differ, in short
OpenAI's business data page (checked 17 November 2026) states that data from ChatGPT Business, Enterprise and Edu is not used to train its models by default. It is encrypted at rest and in transit, and it falls under OpenAI's business terms. If your question is whether a business account beats a personal one, that is answered in ChatGPT Business vs Plus.
Between the business plans, four things change: how you buy it, who can get to the conversations, how long they are kept and where they are stored. Business is built for a team that sets up its own workspace. Enterprise is for an organisation that needs logging, retention and storage written into the arrangement. OpenAI describes Edu as a variant of Enterprise for educational institutions (checked 17 November 2026), with the same admin controls.
#The three plans side by side
The features per plan come from the business data page, the seat limits from the ChatGPT Business FAQ, and sharing and export from the Help Center article on data in Business, all three checked 17 November 2026. OpenAI lists a Standard Business seat at 20 US dollars per user per month billed annually, or 25 dollars billed monthly, and notes that prices can vary by country. Enterprise and Edu pricing depends on the agreement.
#Who can get to the conversations
In Business, every member has their own chat history. OpenAI's Help Center says other members do not see it automatically. The usage analytics admins get show how much the workspace is used and give no access to what was written. A workspace owner can still export a member's data from the Members page, where the option is available. Members cannot export their own Business data.
In Enterprise and Edu, central retrieval is a standard feature. The Compliance API (checked 17 November 2026) passes logs and metadata from the workspace to eDiscovery, DLP and SIEM tools, Microsoft Purview among them. Only a workspace owner can grant access to conversation messages. Messages from locked chats are included: the PIN protects a chat inside ChatGPT, not against retrieval through the Compliance API. The logs are held there for 30 days, so anyone who needs them longer has to pull them into their own systems.
For the law firm's DPO, that is the first real difference. On Business, the firm can export one person's data after an incident. On Enterprise, it can capture conversations continuously in the tools security already runs. Whichever you choose, tell staff beforehand who can read their work chats, when and why. How that works in other tools as well, from Gemini to Copilot, is covered in Can your employer see your ChatGPT?.
#How long chats are kept
Under the GDPR you must be able to justify how long you keep personal data, and the ICO's guidance on storage limitation (checked 17 November 2026) expects a policy with standard retention periods wherever possible. For an AI tool, that means knowing how long a chat with a client's name in it stays put.
In Business, a chat stays until the member deletes it. After deletion, OpenAI schedules permanent removal within 30 days (checked 17 November 2026), unless security or legal obligations require it to be kept longer. OpenAI does not list a workspace-wide retention setting for Business. Retention therefore depends on what each person tidies up.
In Enterprise and Edu, the organisation decides how long data is kept. OpenAI's enterprise privacy page (checked 17 November 2026) lists that control for Enterprise, Edu and Healthcare only. Saved files there follow the workspace policy too. That makes it possible to write one retention period into the record of processing and have it actually applied.
#Where the data is stored
OpenAI offers data residency (checked 17 November 2026) to new Enterprise and Edu workspaces. The regions include Europe, meaning the EEA plus Switzerland, and the United Kingdom as a separate option. Eligible customers can also keep model inference in Europe. Business is not on the list.
The region is set when the workspace is created. If you move from Business to Enterprise, the Business FAQ says OpenAI can upgrade the existing workspace and keep its data. Ask in that conversation whether a converted workspace can use regional storage, because OpenAI describes residency for new workspaces only.
Storing data in Europe does not make OpenAI a European company. What that means for jurisdiction is set out in European AI chatbots.
#When Business stops being enough
Business covers a team that wants ChatGPT without personal accounts, with single sign-on and no training on its data. Enterprise or Edu becomes the answer when one of these is a yes:
- Does the organisation need to capture conversations for eDiscovery, a complaint or an investigation? That needs the Compliance API.
- Do you want one retention period for the whole workspace? On Business, it depends on what each user deletes.
- Does the data have to be stored in Europe or the UK? Then you need a new Enterprise or Edu workspace with data residency.
- Should accounts come from your own directory, and close automatically when someone leaves? That is SCIM, in Enterprise and Edu only.
- Do you want to turn off shared links for everyone? Only Enterprise allows that.
- Do you need more than 200 seats? Since 24 August 2026, that is the limit for a new Business workspace.
For a university or college, Edu is the natural form. The questions are the same, except that students are part of the picture. More on AI in education is on AI governance for education.
#Where every plan stops: the prompt itself
Each of the three plans governs what happens to a conversation after it is sent: no training, a retention period, storage in a region. None of them looks at what goes into the text box. A client file with a passport number in it is just as sensitive on Enterprise as on Business, only better logged.
There is also the question of whether people work in the business workspace at all. Someone who also has a personal Plus account can just as easily open that one in the same browser. The organisation's workspace only helps if the conversation happens there.
The choice between Business, Enterprise and Edu belongs in the decision about ChatGPT, alongside what may and may not go into a prompt. The BeeSensible AI register records that decision: every tool starts on No decision yet, and if you choose Limited use you write the condition yourself, for example "only in the company workspace, no client data". Employees see that condition in the notice when they open ChatGPT, and on the business variant of a tool the notice can point out that the account is not recognised as a business account. What then goes into the text box is highlighted before the message is sent, whichever plan the organisation runs on.