Actually Works · Episode 36
03 / 38Every private thing you've ever typed into ChatGPT trains its next model — and that's the default.
That's turned on by default for Free, Plus and Pro accounts — and it's not just chat history. The files you upload and the answers you get back become training data too, unless you turn one setting off.
The exact clicks
- Open ChatGPT on the web or the app, and open Settings (click your name/profile in the bottom-left on web, or the profile icon in the app).
- Go to the 'Data Controls' section of Settings.
- Find the toggle labeled 'Improve the model for everyone' — if it's on (the default for Free/Plus/Pro accounts), your conversations, uploaded files and the answers you receive are being used to train future models.
- Switch that toggle off.
- Understand what this does and doesn't do: every conversation you have AFTER turning it off stays out of training completely. It does NOT retroactively remove anything already sent before you turned it off — once a conversation entered the training pipeline, this toggle can't pull it back out.
- (Optional, for a single sensitive conversation) Instead of relying on the toggle, start a 'Temporary Chat' (the toggle/icon near the message box, or a dedicated button depending on your app version) — it never enters your chat history and is never used for training, automatically, every time, with nothing to remember to turn off.
- Confirm the change stuck: open a new regular chat afterward and check Settings → Data Controls again — the toggle should still read off.
What changes
- Verified live (16.9.2026) against OpenAI's own Help Center articles: 'How your data is used to improve model performance' and 'What if I want to keep my history on but disable model training?' — this is current, documented behavior, not assumed from an older version of ChatGPT.
- Default state confirmed as ON for Free/Plus/Pro accounts, and OFF by default for Temporary Chats and Business plans, per the same source.
- This is a plain-language, personal-privacy topic — a deliberate exception to this channel's usual demand-report-driven topic selection, made explicitly per the 16.9.2026 reach-first content decision (see channel/content-memory.md).
What it will not do
- Turning the toggle off is not retroactive — anything sent before you turned it off may already be part of a training run and cannot be pulled back out by this setting.
- Temporary Chat still stores the conversation on OpenAI's servers for up to 30 days for safety monitoring before permanent deletion — it skips history and training, not all server-side storage.
- This episode covers OpenAI's ChatGPT specifically; other AI tools (Claude, Gemini, etc.) have their own separate data-training settings, not covered here.
Get the next one
One AI setup a week, straight to your inbox.