Safety and Privacy: What to Set Up in 5 Minutes
Safety and Privacy: What to Set Up in 5 Minutes — ChatGPT From Scratch: How to Use AI Every Day. Training, storage, and deletion are separate; The five-minute settings check; Shared links, ratings, and workplace rules; What to keep out and how to anonymize; Fictional example: a useful request without the personal file. Practical assignment: Set up privacy…
Demo edition. The materials are undergoing editorial review.
Topics covered
- Training, storage, and deletion are separate
- The five-minute settings check
- Shared links, ratings, and workplace rules
- What to keep out and how to anonymize
- Fictional example: a useful request without the personal file
Training, storage, and deletion are separate
When you write in ChatGPT, the service processes your message on its servers. Three separate mechanisms then matter. Model improvement means eligible content may help improve future models. Storage determines whether a conversation remains in your history. Deletion is a separate action affecting a chat or account.
As of 16 November 2026, personal ChatGPT Free, Plus, and Pro workspaces let you turn off Improve the model for everyone under Data controls. Doing so does not hide or delete saved chats. Deletion also cannot “unsend” text that has already been submitted. The strongest rule therefore applies before you press the button: do not enter a secret the service does not need.
Common mistake: “I turned off training, so my earlier data is gone.” Training controls, history, memory, shared links, and deletion have separate controls.
The five-minute settings check
Menu labels can change with the version, plan, and workspace, so look for the function instead of memorizing a route. In the current ChatGPT interface, start with Settings → Data controls. Decide whether you want Improve the model for everyone enabled. Then locate Temporary Chat, Memory or Personalization, and Shared links.
While a chat remains temporary, it does not appear in history, create or update memories, or improve OpenAI’s models; OpenAI says it may be retained for up to 30 days for safety purposes. In the current version, a temporary chat can use existing personalization if you allow that when starting it, and a saved temporary chat becomes a regular chat. “Temporary” therefore does not mean “never processed.”
Review memory separately: turning memory off does not erase history, and deleting a chat alone may not remove a separately stored memory. To remove a detail fully, check memory, chats, files, and connected apps where it may remain.
Shared links, ratings, and workplace rules
A conversation link grants access to selected content. For a personal account, anyone with the link can open and forward it; the shared material may include messages, images, or uploaded files. Review the entire preview before sending. In the current interface, the list is under Settings → Data controls → Shared links → Manage. Deleting a link stops future access through that URL, but it does not erase a copy that a recipient has already saved in their account.
Reddit discussions reveal a practical problem: some users report creating a link without realizing it or being unable to find link management in a mobile app. These are user reports, not proof of a system-wide fault, but they suggest a useful check—open the shared-link list on the web.
Rating an answer with thumbs up or down is another separate action. OpenAI’s documentation says the associated conversation may be used for training even after a general opt-out. Do not rate a chat containing personal details. Workplace and workspace rules also apply; a personal setting does not authorize uploading confidential employer material.
What to keep out and how to anonymize
Settings reduce particular risks; they do not make secret material necessary for a request. Do not paste passwords, one-time login codes, full card or account numbers, identity-document scans, identifiable health information, another person’s contact or financial data, or confidential employer material outside an approved workflow.
Anonymization keeps the shape of the task while removing identifiers. Replace a name with “Client A,” an address with “Address Y,” and an exact amount with “Amount X” or a rounded figure when scale matters. Remove policy, account, and phone numbers; retain a date only if timing affects the answer. Covering text with a black rectangle may be inadequate if an accessible text layer remains underneath. A safer method is to create a separate cleaned copy, then test search and text selection before uploading it.
For medical, legal, or financial questions, treat an anonymized AI answer as a draft to verify, not a personalized professional conclusion.
Fictional example: a useful request without the personal file
Fictional teaching scenario. Nora works at a clinic and wants to explain an estimate clearly to a patient. The original document includes a name, birth date, phone number, treatment history, policy number, and exact amounts. Even a temporary chat does not make all those fields necessary.
Nora creates a cleaned copy and writes: “Patient A, about 70, has been offered two fillings and a crown. The total is Amount X, insurance covers Amount Y, and Amount Z remains. Explain this in plain language in 4–5 sentences. Do not add medical conclusions.” She keeps the treatment type, approximate age, and relationship between the amounts because they support the explanation, while removing identifiers.
After receiving the answer, Nora moves the wording into an internal document and restores verified real details only there. She does not ask the model to diagnose the patient, and she checks the final text against clinic policy. A common mistake would be replacing only the name while leaving the phone, policy number, and exact address, which may still identify the person.
Practical assignment
Set up privacy and create a cleaned request
Continue to learningCourse outline
- Your First Conversation: Save 15 Minutes With AI Today
- What to Ask: 10 Everyday Tasks and a Simple Request Formula
- Safety and Privacy: What to Set Up in 5 Minutes
- Photos, Files and Voice: AI That Sees and Hears
- When AI Gets It Wrong: A Confident Tone Isn't the Truth
- Memory, Instructions and Projects: Give ChatGPT Useful Context
- AI at Home: Money, Food, Trips and Paperwork
- AI at Work and in Study: The Rules and Being Honest
- Your 10-Task Kit and What Comes Next