How to stop ChatGPT, Claude and Gemini from training on your chats

Every major AI chat has a switch that stops your conversations being used to train future models. Here's where each one is, how long your chats are kept either way, and why the switch alone isn't enough.

Updated October 16, 2026

ChatGPT

1. Open Settings → Data controls (on the app: sidebar → your profile → Settings → Data controls).
2. Turn off "Improve the model for everyone".

For a one-off sensitive chat, use Temporary Chat. It isn't saved to your history, doesn't update memory and isn't used for training, but OpenAI may still keep it for up to 30 days for safety reasons.

Claude

1. Open Settings → Privacy (claude.ai/settings/data-privacy-controls).
2. Turn off the model improvement setting ("Help improve Claude").

Since Anthropic's September 2025 update, chats from people who allow training can be kept for up to five years; with it off, the standard is about 30 days after you delete a chat. Incognito chats are never used for training, even if the setting is on. Chats flagged by safety systems can still be reviewed.

Gemini

1. Go to myactivity.google.com/product/gemini (or Gemini → Settings → Activity).
2. Turn off Keep Activity (called "Gemini Apps Activity" in older versions).

With it on, Google keeps your chats for 18 months by default and human reviewers may read a sample of them. Reviewed chats are kept for up to three years, even if you delete your activity. With it off, chats are still held for up to 72 hours. Google's own help page says not to enter confidential information you wouldn't want a reviewer to see.

What these switches don't do

• They don't stop the company receiving your text. Whatever you paste still travels to their servers and sits there for days to weeks.
• They don't cover safety review. Flagged chats can be read by staff regardless of your setting.
• They reset when you switch. A new account, a work laptop, a different app or a colleague's login all start with their own defaults.
• They can change. Defaults and retention periods have changed several times in two years.

The only way to be sure a name, account number or SIN isn't stored is to never send it. That's what Hushtag does: it swaps private details for hashtags like #PERSON_1 before you hit send, then puts the real details back in the answer, on your screen only.

Do both: turn the training switch off and hush what you paste.

Sources: OpenAI, Data controls · Anthropic, Who can view my conversations · Google, Gemini Apps Privacy Hub. Settings names checked October 2026; companies change them often.