The checklist
1. SWITCH OFF TRAINING. Settings > Data controls > "Improve the model for everyone" > off. It doesn't make ChatGPT worse at helping you. It just stops it using your work as training data.
2. READ THE MEMORY. Settings > Personalization > Manage memories. Read every line. Delete anything about a client, a health issue or money that you wouldn't want sitting there.
3. DECIDE ON MEMORY ITSELF. In the same place you can switch off "Reference saved memories". Turning that off also turns off "Reference chat history".
4. DELETE THE CHATS YOU'D RATHER NOT KEEP. Deleting removes a chat from your view straight away. OpenAI deletes it within 30 days unless it has to keep it for legal or security reasons. Deleting a chat doesn't delete a memory saved from it, so do step 2 as well.
5. DO IT FOR EVERYONE ON THE TEAM. Each personal account has its own switch. Send this page to anyone using their own ChatGPT for work. It takes them 30 seconds.
A ChatGPT Business or Enterprise workspace is different: OpenAI doesn't use those chats for training by default. If your team uses AI for client work every day, that's usually the cleaner fix than asking everyone to remember a switch.
Using Claude too? The equivalent is Settings > Privacy, the model-improvement switch. With it on, Anthropic keeps chats for up to five years, with it off 30 days. The full list of settings is on the
AI privacy guide.
Sources, checked 29 Sep 2026: OpenAI Data Controls FAQ, Memory FAQ and business data pages. Menus move, so if a label looks different in your app, search for it in settings.