In April 2023, Samsung engineers pasted proprietary source code into ChatGPT. Three separate times over 20 days. Confidential chip designs. Internal meeting transcripts. All of it landed on OpenAI's servers, where it became training data.
Most people have no idea what happens to the stuff they type into that chat window. It feels like a private conversation. It's closer to a group email you can't unsend.
Every major AI platform trains on your conversations by default. But every one of them also lets you turn it off. Here's what their policies actually say, and exactly how to opt out.
ChatGPT
OpenAI's policy is straightforward. From their data usage page:
"When you use our services for individuals such as ChatGPT, Codex and Sora, we may use your content to train our models."
Training is on by default for free and Plus users. Business, Enterprise and API users are excluded.
There's a catch most people miss: even if you opt out, clicking thumbs-up or thumbs-down on any response re-opts you in for that entire conversation. OpenAI's own help center says "the entire conversation associated with that feedback may be used to train our models."
How to turn it off:
- Open chatgpt.com and log in
- Click your profile picture (bottom-left on desktop, top-right on mobile)
- Go to Settings → Data Controls
- Toggle off "Improve the model for everyone"
Your chat history stays. Your conversations just stop becoming training data. For anything especially sensitive, use Temporary Chat (top-right icon), which is never trained on and deletes automatically.
Google Gemini
Google is the only platform that explicitly confirms human reviewers read your conversations. From their Gemini Apps Privacy Hub:
"A subset of chats are reviewed by human reviewers to help improve Google services, including Gemini models, other generative AI models that power Gemini Apps, and technologies that help reduce unintended activations."
Reviewed conversations are retained for up to three years. As of September 2025, Google expanded the scope so file uploads, photos and screenshots are also included by default. The setting was renamed from "Gemini Apps Activity" to "Keep Activity."
How to turn it off:
- Go to myactivity.google.com/product/gemini
- You'll land on the Keep Activity page
- Click "Turn off"
- To clear past data too, choose "Turn off and delete activity"
Claude
This one stings. Anthropic spent years marketing itself as the privacy-first alternative. Then in August 2025, they updated their consumer terms and flipped the default:
"We're now giving users the choice to allow their data to be used to improve Claude."
"Choice" is generous. When existing users were prompted to accept the new terms, the toggle labeled "You can help improve Claude" was pre-checked to ON. If you clicked Accept without noticing that small toggle, you were opted in. Allow training and data retention jumps from 30 days to five years. Their policy spells this out:
"Chat and coding session data we may use for improving our models includes the entire related conversation, along with any content, custom styles or conversation preferences."
How to turn it off:
- Open claude.ai/settings
- Click Privacy in the left sidebar
- Find "Help improve Claude" and toggle it off
Retention drops from five years back to 30 days. Incognito mode conversations are never used for training regardless of your settings.
Microsoft Copilot
Copilot is the most honest about how little their opt-out actually does. From their privacy controls page:
"This setting will not exclude your conversations from being used for other general product or system improvements nor from use for advertising, digital safety, security and compliance purposes."
Read that again. You can opt out of model training, but Microsoft explicitly reserves the right to keep using your conversations for product improvements and advertising. They also confirm that "some Copilot conversations are subject to both automated and human review."
How to turn it off (for what it's worth):
- Go to copilot.microsoft.com
- Click your profile icon → Privacy → Personalization
- Toggle off "Model training"
- On mobile: Menu → Profile → Account → Privacy
Four platforms, about two minutes each. The settings are real. The opt-outs work (mostly). But the defaults are all set the same way, and none of them are set in your favor.