Check the information, the account, and the permissions. A setting about model training does not settle every question about using business data.
Start with what the task actually needs
There is a difference between asking for questions about a public service page and uploading an entire customer email thread. The second may contain names, contact details, attachments, and information unrelated to the task. Start by describing what you want the tool to do and the smallest amount of information it needs.
For learning, a public or invented example is often enough. You can understand how a summary works without using a real customer’s message. Removing a name alone may not make a document anonymous: an address, unusual transaction, or combination of details can still identify someone.
Check the actual account and product
OpenAI’s Data Controls FAQ explains how to turn off model improvement in ChatGPT. Conversations can remain in your history even when they are excluded from training. Its business data commitments say that ChatGPT Business, Enterprise, and API inputs and outputs are not used for model training by default.
Those are specific policies, checked on September 10, 2026. They are not a blanket answer about every account, app, or document. Training, storage, access, and sharing are separate questions. Check the current terms and settings for the product and workspace you will actually use, including any connected service.
Use our inbox work as a concrete example
For Parker & Associates, we implemented inbox scanning that identifies potential business inquiries and presents a summary, contact information, and property details for broker review. Replies remain in the team’s email workflow.
That illustrates a specific task: helping a broker review an incoming request. It does not mean every business should connect its whole inbox to a chatbot. When considering a similar workflow, ask which messages it needs, who may access the output, and whether it can only read information or also send a reply. Our case study describes the work; it is not a security certification for another company’s setup.
Answer these questions before uploading
- Information: does the file contain customer details, employee records, passwords, private deal terms, or unnecessary attachments?
- Permission: has your business approved this use of the information and this particular tool?
- Handling: what do the provider’s terms say about training, storage, deletion, and access?
- Connections: what can linked apps read or change, and who can share the result?
- Review: who checks the output before it is used?
If you cannot answer a question yet, use an invented example while you find out. For confidential or regulated records, involve the person responsible for your business’s data requirements before choosing a setup.
A useful first exercise
Write a fictional inquiry with the same structure as the task you want to understand. For example: “A prospective tenant needs roughly 1,500 square feet for a shop and wants to discuss availability next month.” Ask for a summary and a list of missing information. Check that the tool does not invent a budget, name, or suitable property.
You have now tested an interaction without uploading a real inquiry. If you want help moving from that exercise to a business decision, see what a tailored Endcap engagement includes.