How to Stop ChatGPT, Claude, Gemini and Copilot Training on Your Data (and Check It Stayed Off)

Most people who ask how to stop an AI chatbot "training on my data" are really asking three different questions at once. Is my chat used to train the next model? Does the assistant remember things about me? How long does the company keep my chats? Each app has a separate control for each, and switching off one does nothing to the other two.

On 19 October 2026 I opened the settings in my own ChatGPT and Claude accounts to check the current labels. For Gemini, Microsoft Copilot and GitHub Copilot I worked from the companies' help pages, linked at the bottom. Menus get renamed every few months, so if a label below doesn't match yours, look for the closest wording in the same menu.

The short answer

None of these switches is retroactive. If your chats were already used in a training run, switching off now doesn't pull them back out.

Related reading: How to Put a Hard Spending Cap on AI Agents (OpenAI, Claude, Bedrock, OpenRouter) Before an Overnight Run Costs You Thousands · Custom GPTs Retire on December 11, 2026: How to Migrate to ChatGPT Plugins (and What Happens to Gemini Gems)

Training, memory and retention are three different things

• Training means your conversation can be used to build future models. This is what the switches in the table control.
• Memory (or "personalization") means the assistant stores facts about you and uses them in later chats. Turning off training doesn't touch memory, and the other way round.
• Retention is how long the company keeps a copy on its servers. Even with training off and memory off, every company here keeps chats for a while for safety and legal reasons.

ChatGPT

1. On the web, open your profile menu → Settings → Data controls and turn off Improve the model for everyone. The direct link is chatgpt.com/settings/data-controls. In the iOS and Android apps it's under your profile → Data controls.
2. Signed in, the choice follows your account to every device. Signed out, it's saved only in that browser, so a logged-out chat on a new laptop starts with training on.
3. On a personal plan the same switch covers Codex tasks.
4. For a belt-and-braces opt-out, OpenAI's privacy portal (privacy.openai.com) has a "do not train on my content" request. OpenAI says it also applies to Sora. Either route is enough on its own, but the portal request is a record that doesn't depend on a toggle.

Two catches. If you press thumbs up or down, OpenAI says the whole conversation linked to that feedback may be used for training, even with the switch off. And Temporary Chat is the only way to keep a single chat out of training, history and memory without changing your settings. OpenAI may still keep a copy for up to 30 days for abuse monitoring.

"I deleted the chat, so why does ChatGPT still remember?"

Saved memories are stored separately from chats. OpenAI's Memory FAQ says deleting a chat doesn't delete a memory created from it. To remove something completely, delete it in Settings → Personalization → ChatGPT memory → saved memories → Manage, and delete the original chat too. Switching memory off stops new memories but doesn't erase the old ones. Deleted chats leave your account straight away and are scheduled for permanent deletion within 30 days. The exceptions are data already de-identified and data OpenAI must keep for security or legal reasons. OpenAI's help pages also note that "recent legal developments" may affect retention for some services.

Claude

1. Open Settings → Privacy (claude.ai/settings/data-privacy-controls) and turn off Help improve our AI models. The description under it reads "Allow the use of your chats and coding sessions to train and improve Anthropic AI models."
2. On Free, Pro and Max, the switch covers Claude Code as well.
3. Anthropic's privacy center says what happens when it's off. New chats and coding sessions aren't used for future training. Data already in a training run, or in a model that's already trained, stays there.

How long Anthropic keeps things:

• With the setting on, de-identified chats can be kept for up to 5 years for training. This only covers chats started or resumed after you switched it on.
• Deleted chats leave your history straight away and are removed from back-end storage within 30 days.
• Thumbs-up/down feedback is kept for 5 years.
• Chats flagged by safety systems are kept longer, whatever your setting: up to 2 years for the content and 7 years for safety scores.

Incognito chats aren't used for training and aren't saved to history or memory, but they're still kept for 30 days by default. Business products (Claude for Work, the API) are under separate commercial terms and aren't covered by this switch.

Gemini

1. Go to myactivity.google.com/product/gemini and turn off Keep Activity.
2. Google's help page, updated 13 October 2026, explains what changes. Future chats won't be used to train its AI models unless you send feedback, and they won't show in your activity.
3. The fine print:
   • Chats are still saved for 72 hours.
   • Google says it may still use them, "including with help from human reviewers", to respond and keep people safe.
   • Chats already selected for human review are kept separately from your account for up to three years. Deleting your activity doesn't remove them.
4. If you leave Keep Activity on, the default auto-delete is 18 months. You can change it to 3 or 36 months.
5. Temporary Chats are kept for 72 hours and aren't used to improve Google AI with help from human reviewers.

Microsoft Copilot (the consumer app)

1. On copilot.com, open your profile icon → your name → Privacy. In the Windows or Mac app it's Settings → Privacy. On mobile it's Account → Privacy.
2. Turn off Training on conversation activity and Training on voice conversations. They're separate switches.
3. Microsoft's support page says the opt-out covers future conversations only. It doesn't stop your chats being used for general product improvement, advertising, digital safety, security or compliance.
4. Memory is a separate menu: profile → Memory → Personalization and memory. "Delete all Memory" clears what Copilot has learned but not your conversation history, which you delete separately.

Work and school accounts using Microsoft 365 Copilot fall under Microsoft's Enterprise Data Protection instead.

GitHub Copilot

Since 24 April 2026, GitHub uses Copilot interaction data from Free, Pro and Pro+ accounts to train its models unless you opt out. That covers prompts, suggestions and code context. Business and Enterprise plans aren't affected. The setting is in your Copilot settings at github.com/settings/copilot/features, under privacy. Reports differ on whether earlier "product improvement" opt-outs carried over, so check it rather than assume.

"The toggle turned itself back on"

A Hacker News thread in September, "OpenAI keeps re-enabling the 'allow training' setting", got hundreds of upvotes. One commenter said the same had happened to them on Claude. I couldn't confirm a cause. Commenters suggested three likely triggers:

• accepting new terms or an offer;
• using a new device while signed out;
• a new app version resetting preferences.

Whatever the cause, the fix is the same: check the switch now and then, and for ChatGPT, file the privacy-portal request too, since it doesn't depend on a toggle.

A two-minute monthly check

1. ChatGPT: Data controls → Improve the model for everyone is off. Glance at saved memories.
2. Claude: Privacy → Help improve our AI models is off.
3. Gemini: Keep Activity is off, or auto-delete is set to the period you want.
4. Copilot: both training switches are off.
5. GitHub: the Copilot model-training setting is off.
6. After any new terms pop-up, new device or big app update, run the list again.
7. For anything you'd never want stored, use Temporary or Incognito chat, and remember that even those are kept for 30 days (72 hours on Gemini).

Last checked: 19 October 2026. These settings move around often. If a menu has changed, tell me and I'll update this page.

Sources

• OpenAI: Data controls in ChatGPT
• OpenAI: Temporary Chat FAQ
• OpenAI: Chat and file retention in ChatGPT
• OpenAI: Memory FAQ
• OpenAI: How your data is used to improve model performance
• Anthropic: How do I change my model improvement privacy settings? (updated 3 Aug 2026)
• Anthropic: How long do you store my data?
• Anthropic: Use incognito chats
• Google: Gemini Apps Privacy Hub (updated 13 Oct 2026)
• Microsoft: Microsoft Copilot privacy controls
• InfoQ: GitHub Copilot to use interaction data for training (April 2026)
• Hacker News: "OpenAI keeps re-enabling the 'allow training' setting"