It depends on the account your team uses. On individual plans (Free, Plus, Pro), conversations can be used to train OpenAI's models unless you turn that off, and your company has no contract with OpenAI covering the data. On ChatGPT Business, Enterprise, and the API, your data is not used for training by default and a Data Processing Addendum (DPA) is available. On every plan, some information should never be pasted in.

The ChatGPT privacy facts below come from OpenAI's help center and policy pages, checked on October 23, 2026. These pages change often, so recheck them before you write your own policy.

What happens to what you paste, plan by plan

Topic Free, Plus, Pro Business Enterprise API
Model training On by default, can be turned off Off by default Off by default Off by default (since March 1, 2023)
Retention Kept until deleted, then removed within 30 days Deleted chats removed within 30 days Retention period set by workspace admins Abuse monitoring logs kept up to 30 days
Shared links Anyone with the link can view Workspace members only Workspace members only, admins can turn sharing off Not applicable
Privacy framework Covered by OpenAI's own privacy policy DPA available DPA available, European data residency available DPA available, zero data retention on approval

Model training

On individual plans, OpenAI says it may use your content to train its models. To opt out, turn off Improve the model for everyone under Settings > Data controls, or select "Do not train on my content" in OpenAI's Privacy Portal. The opt-out applies to new conversations.

One detail many people miss: even after opting out, giving a thumbs up or thumbs down on a response lets OpenAI use the entire conversation for training. On Business, Enterprise, and Edu, that feedback does not permit training (OpenAI on how your data is used).

Retention

A deleted chat leaves your sidebar right away, and OpenAI schedules it for permanent deletion within 30 days, unless it was already de-identified or OpenAI must keep it longer for security or legal reasons (OpenAI on chat retention). That exception is real: in the lawsuit brought by The New York Times, a US court order required OpenAI to preserve consumer conversations, deleted ones included, until September 26, 2025.

Temporary Chat stays out of your history, does not create memories, and is not used for training. OpenAI may still keep a copy for up to 30 days for safety purposes.

On Enterprise, workspace admins control how long data is retained (OpenAI enterprise privacy).

Memory

With Memory on, ChatGPT can draw on past chats, saved memories, files, and connected apps such as email. Two things matter for a business. Deleting a chat does not necessarily delete a saved memory created from it, so you also need to remove it under Settings > Personalization > Memory. And a client named in one conversation can come back in an unrelated answer weeks later. In Business and Enterprise workspaces, Memory can also be governed by organization-level settings.

Shared links

A link created from a personal account can be opened by anyone who has it. Links have no expiration date and can be forwarded. OpenAI says shared pages are not intended for search engine indexing, but that this does not make a link private (OpenAI on shared links). On Business and Enterprise, a link only opens for members of the same workspace. Only Enterprise lets admins turn off sharing for the whole organization.

Who can read your chats

On individual plans, a limited number of OpenAI staff and trusted service providers can access content to investigate abuse or security incidents, handle support requests, deal with legal matters, or improve models (unless you opted out). Content is stored in the US and other countries. OpenAI's own advice is not to enter sensitive information you would not want reviewed or used.

What never to paste into ChatGPT

On a personal account, none of the following. On a business plan, only if your policy allows it and a DPA is signed:

  • personal data about customers, leads, or employees: contact details, resumes, payroll, medical leave, performance reviews;
  • health data and other special categories (political opinions, union membership, racial or ethnic origin);
  • usernames, passwords, API keys, and banking credentials;
  • contracts, quotes, and documents covered by a nondisclosure agreement;
  • unpublished financials, acquisition plans, negotiated prices;
  • proprietary source code and trade secrets;
  • privileged communications with your lawyer or accountant.

For many tasks, anonymizing is enough: "Client A," "a warehouse employee," "an amount of X dollars." Rewriting an email or reviewing the structure of a contract does not require real names.

An 8-rule AI policy for a small team

France's data protection authority, the CNIL, recommends governing generative AI with an internal policy that lists allowed and prohibited uses, and banning certain confidential or personal data from it (CNIL Q&A on generative AI, in French). The advice works anywhere. For a small team, one page is enough:

  1. Company content goes into company accounts only. No personal accounts for internal data, even with training off: your company has no contract with OpenAI on those.
  2. A posted "never paste" list. The categories above, adapted to your line of work.
  3. Anonymize by default. Swap names, addresses, and amounts for neutral placeholders before pasting.
  4. Temporary Chat for one-off questions. It leaves nothing in history or memory.
  5. No shared links outside the team. Copy the useful text instead of the conversation, and delete links you no longer need.
  6. A quarterly memory review. Everyone deletes what concerns clients or coworkers.
  7. Approved connectors only. Email, calendar, or file storage get connected only with the owner's sign-off.
  8. One named owner. That person keeps the DPA, records the tool in your data inventory, and receives incident reports. If someone pastes data by mistake: delete the chat, delete any shared link, tell the owner.

GDPR: who is responsible for what

If you handle personal data about people in the EU or the UK, the GDPR (or UK GDPR) applies, wherever your company is based.

Controller and processor

When an employee processes data about your customers or coworkers in ChatGPT, your company decides on that processing, which makes it the controller. The question is what role OpenAI plays.

  • Individual plans: for users in the European Economic Area and Switzerland, OpenAI Ireland Limited is the controller of the data described in its EU privacy policy, including conversation content. Your company has no contract governing what OpenAI does with it.
  • Business, Enterprise, API: OpenAI acts as a processor on the customer's behalf, under a DPA.

The DPA

OpenAI signs a DPA for ChatGPT Business, ChatGPT Enterprise, and the API through an online form. It covers processing instructions, confidentiality, the list of subprocessors, and returning or deleting data at the end of the contract. Article 28 of the GDPR requires this kind of contract whenever a processor handles personal data for you. Keep it with your records of processing.

If you work with health data in the US, note that OpenAI states it can sign Business Associate Agreements for HIPAA purposes on its API Platform. Check the scope with OpenAI before relying on it.

Transfers outside the EU

OpenAI processes data outside the European Economic Area, including in the US. These transfers rely on European Commission adequacy decisions or on Standard Contractual Clauses. European data residency, which keeps stored content in the EEA and Switzerland, is limited to new Enterprise and Edu customers and eligible API customers. Even then, some data, such as metadata or content sent to external integrations, can be processed elsewhere.

One point no contract changes: OpenAI is a US company, subject to US law. For most uses that is not a blocker. It becomes one when your clients contractually require European vendors.

Alternatives

Move to an OpenAI business plan. If your team already uses ChatGPT, this is the simplest fix: ChatGPT Business covers a small company, with no training by default, workspace-only links, and a DPA. Enterprise adds European data residency and the option to disable sharing.

Pick a European provider. Mistral, a French company, offers its own assistant with a team plan. We compared both, including pricing and privacy settings, in Mistral vs ChatGPT for small businesses.

Use a workspace that lets you choose the models. In Omeon, the AI assistant works from your company's documents and gives access to models from Anthropic, Google, Mistral, OpenAI, and xAI. Admins choose which providers the organization allows and can restrict it to Mistral's European models. Omeon is hosted in Europe, with its database in France. As stated in our privacy policy, our contracts with AI providers prohibit them from using your data to train their models. A usage policy is still worth having, whatever the tool.

FAQ

Is ChatGPT GDPR compliant?

No tool makes a company compliant on its own. OpenAI provides a DPA and transfer safeguards for its business plans. On individual plans, there is no contract between your company and OpenAI, so avoid them for any personal data about customers or employees.

Does turning off training protect my data?

Only partly. It stops new conversations from being used to train models, but chats stay stored until you delete them, then for up to 30 days. A thumbs up or down still allows training on that conversation.

Is ChatGPT Business enough for a small company?

For most small teams, yes: no training by default, workspace-only shared links, and a DPA. European data residency and the ability to disable sharing for the whole organization are Enterprise features.

What if an employee pasted client data into a personal account?

Delete the chat and any shared links, remove related memories, and log the incident. If the data is sensitive or covers many people, check with your data protection officer or a legal advisor whether it counts as a personal data breach you must report to your regulator.