Does ChatGPT Train on Your Data? The Product Boundary
ChatGPT data use depends on the product and workspace. OpenAI may use content from individual services for model training unless the user opts out. ChatGPT Business, Enterprise, Edu, Healthcare, Teachers, and the API are excluded from training by default. Training settings, retention, memory, and access controls remain separate decisions. This guide maps those boundaries and shows what an enterprise must verify before employees send company data.
Does ChatGPT train on your data? OpenAI's answer changes with the product boundary. Content from individual ChatGPT services may be used for model training. A user can opt out, and the choice applies to new conversations. OpenAI excludes content from ChatGPT Business and Enterprise from training by default. The same default applies to ChatGPT Edu and Healthcare. It also covers ChatGPT Teachers and the API Platform. I would reject a company policy that says only "ChatGPT is approved." The policy needs to name the workspace and account type. It also needs the data class and approved route.
TL;DR
- Individual ChatGPT content may be used for training unless the user turns off Improve the model for everyone.
- Temporary Chat excludes the conversation from training and history, while OpenAI may retain it for up to 30 days for safety.
- Managed business workspaces and the API are excluded from training by default; an explicit data-sharing choice can change that rule.
- Training, retention, memory, access, and third-party app transfers require separate review decisions.
Does ChatGPT train on your data in an individual account
OpenAI's model-performance data policy says content from services for individuals, including ChatGPT, may be used to train its models. The policy also gives the user an opt-out. After the user opts out, new conversations are excluded from training.
The setting lives under Settings, then Data controls, as Improve the model for everyone. OpenAI's Data Controls FAQ says the account-level choice follows a signed-in user across devices. Signed-out use stores the choice in that browser. Picture the practical control: a small toggle in a personal account menu governs the treatment of every new conversation under that account. An enterprise administrator needs an organization-level control instead of trusting that personal toggle.
Feedback creates a separate branch. OpenAI says a conversation associated with thumbs-up or thumbs-down feedback may be used for training even after the user has opted out. A company policy that permits individual accounts needs to address that feedback path explicitly.
Temporary Chat changes training and retention
Temporary Chat gives an individual user a narrower handling mode. OpenAI says these conversations stay out of chat history, avoid creating or updating memories, and remain excluded from model improvement. The company may retain a Temporary Chat for up to 30 days for safety purposes.
That 30-day statement exposes a common procurement error. Training and retention answer different questions. A conversation can be excluded from training while still being retained for a stated operational purpose. Deleting or archiving a regular chat also has its own effect. The Data Controls FAQ says turning off model improvement leaves saved chats in history, and archiving a chat leaves the training preference unchanged.
I would put model training and service retention in separate boxes on the review sheet. Memory and human access need their own boxes too. Combining them under a single "privacy" checkbox hides the exact treatment a security reviewer needs to approve.
Business workspaces use a different default
OpenAI's Enterprise Privacy page states that it excludes business data from model training by default. The commitment covers inputs and outputs from ChatGPT Business, Enterprise, Healthcare, Edu, Teachers, and the API Platform. An organization can explicitly opt in to share data through a feedback mechanism, which creates an exception to the default.
The same page says the commitment applies to GPTs used inside managed workspaces. It also says data accessed through workspace apps stays out of training by default. Workspace administrators control which apps are enabled, and users authenticate to connected applications. Those controls establish the OpenAI-side rule. The enterprise still owns the decision to connect a particular source and the permissions exposed through that source.
This product distinction belongs in the same approval record as the controls described in ChatGPT Enterprise controls. A screenshot showing an Enterprise badge is useful evidence. The durable record should also identify the tenant, administrator, retention configuration, enabled apps, and date of review.
The API separates training from stored application state
OpenAI's API data controls documentation says API data has been excluded from model training since March 1, 2023, unless a customer explicitly opts in to share it. The documentation then divides stored data into abuse-monitoring logs and application state.
By default, abuse-monitoring logs may include prompts and responses plus derived metadata, with retention for up to 30 days. Eligible customers can request Modified Abuse Monitoring or Zero Data Retention, subject to OpenAI approval and additional requirements. Some API capabilities still store application state because the feature needs that state to perform its job. Zero Data Retention also changes endpoint behavior, including treatment of the store parameter for Responses and Chat Completions.
A shared API key proves access to an OpenAI project. It says little about the natural person or agent behind a specific request. The request-level ownership gap remains after the provider's training exclusion, as the OpenAI enterprise security boundary explains.
An enterprise approval needs five named facts
A usable approval record should identify the exact OpenAI service and the account or workspace that will send data. It should record the training default and any opt-in exception. The retention configuration belongs beside the approved data classes and enabled apps. The record also needs an accountable owner.
The control should also distinguish direct ChatGPT use from an internal application calling the API. A personal ChatGPT tab, a managed Business workspace, and an API project can show similar model output while applying different administrative controls. Route-level policy prevents the product label from becoming a substitute for evidence.
Finally, review the data before transmission. A no-training commitment governs OpenAI's use after receipt. The enterprise separately decides if an employee may send a customer list, source file, legal memo, or incident transcript. That decision belongs at the point where the HTTPS request carries content toward the provider. Preventing data leaks to ChatGPT starts at that inspection point.
DeepInspect
DeepInspect sits at the HTTP AI request boundary between authenticated users or agents and approved LLM endpoints. It can bind a verified identity to each routed prompt, classify the content, enforce destination and data rules, and write a tamper-evident record before the request reaches the model.
Provider-side training attestation remains with OpenAI and the enterprise's vendor-assurance process. The DeepInspect record can show that a specific employee sent an approved data class to an approved OpenAI route under the policy version in force at that moment. Requests aimed at an unapproved consumer endpoint can be denied, while permitted API traffic follows the enterprise's named route and policy.
Book a technical deep dive at deepinspect.ai.
Frequently asked questions
- Does turning off training delete existing ChatGPT conversations?
No. OpenAI's Data Controls FAQ says turning off Improve the model for everyone affects the use of new conversations for training. Saved, archived, and project chats remain in their respective locations. Deletion and export use separate controls. A review should record the training preference and the retention action independently.
- Does ChatGPT Enterprise train on company data?
OpenAI says ChatGPT Enterprise business data is excluded from model training by default. The company can explicitly opt in to share selected data through feedback mechanisms. Administrators should verify the managed workspace, record any opt-in path, and review enabled GPTs and apps. An employee's statement that they used "enterprise ChatGPT" supplies too little evidence.
- Is Temporary Chat the same as Zero Data Retention?
No. OpenAI describes Temporary Chat as a ChatGPT mode that avoids history, memory updates, and training, with possible retention for up to 30 days for safety. Zero Data Retention is an API control for eligible approved customers and changes storage behavior for supported endpoints. They belong to different product surfaces.
- Can a gateway prove that OpenAI kept data out of training?
A gateway can prove which routed request left the enterprise, which identity sent it, the destination, the data classification, the policy version, and the enforcement outcome. Provider-side training happens after receipt and remains under OpenAI's controls and contractual commitments. Verification of that provider process belongs in vendor assurance and contract review.