How to Stop AI Chatbots From Training on Your Data: ChatGPT, Gemini, Claude, Copilot and Meta AI

Every chat you have with an AI assistant is data. Depending on your settings, that data might be kept for months or years, reviewed to improve the service, or fed into training the next version of the model.

The good news is that most of the big assistants now give you real controls. The bad news is that each company names them differently, hides them in different menus, and renames them every few months. This guide walks through where the switches live in September 2026, what they actually do, and where the limits are.

Training, retention and memory are three different things

Before you start flipping toggles, it helps to know what you are controlling. Most assistants separate three ideas that people tend to lump together.

Model training is whether your conversations can be used to improve future models. Retention is how long the company keeps your chats on its servers, whether or not they train on them. Memory or personalization is whether the assistant remembers details about you across chats to tailor its answers.

Turning off one does not turn off the others. You can opt out of training and still have a full chat history and a long memory, or keep training on while using a disposable chat for one sensitive question.

Key takeaway: opting out of training usually only affects future chats. It does not pull your data out of models that were already trained, so the earlier you change these settings, the better.

Where each setting lives at a glance

Menu wording changes often and can differ between web, desktop and mobile apps. Treat the labels below as a guide, and look for similar wording if your version reads slightly differently.

ChatGPT: turn off "Improve the model for everyone"

OpenAI's main switch is a single account-wide toggle. Once it is off, new conversations still appear in your history but are not used to train its models, and the setting follows you across devices.

1. Click or tap your profile icon.
2. Open Settings, then Data Controls.
3. Turn off Improve the model for everyone.
4. If you use Codex on a personal plan, note that the same toggle also covers those tasks.

OpenAI also offers a privacy portal where you can select "Do not train on my content," and that choice is reflected in the same account setting. Signed-out users get a simpler version of the toggle under the help menu's settings.

Use Temporary Chat for one-off sensitive questions

To start a Temporary Chat, open a new chat and select Temporary near the top of the window. These chats do not appear in your history, do not create memories and are not used for training.

They are not instantly erased, though. OpenAI says Temporary Chats are kept for up to 30 days for safety and abuse monitoring, then deleted.

Review what ChatGPT remembers

Memory lives under Settings > Personalization > Memory (some versions show it simply as Settings > Memory). From there you can turn off saved memories, stop ChatGPT from referencing past chat history, or delete individual memories.

Worth knowing: if the training toggle is on, OpenAI says saved memories can be used to improve its models too. Business, Enterprise and Edu accounts are not used for training by default.

Google Gemini: turn off Keep Activity

Google renamed "Gemini Apps Activity" to Keep Activity, and you will still see both names depending on the screen. For adults it is on by default. When it is on, Google stores your chats in your account and says a sample of your future uploads can be used to improve its services.

On the web:

1. Go to gemini.google.com.
2. Click Settings & help, then Activity.
3. Near the top, click On, then choose Turn off or Turn off and delete activity.

On Android or iPhone, open the Gemini app, tap Menu or your profile picture, choose Gemini Apps Activity, then tap On and pick one of the same two options.

Know the trade-offs and the 72-hour window

Turning off Keep Activity has side effects. Google says many connected apps, including Workspace apps such as Gmail and Drive, Google Maps, Google Photos and YouTube, stop working inside Gemini while it is off. Chats started with Gems also will not be saved.

Even with the setting off, Google keeps conversations with your account for up to 72 hours to run the service and process feedback. That activity does not show up in your history.

Prefer to keep history? Shorten it

If you want your history but not forever, use auto-delete instead. By default Gemini deletes activity older than 18 months, and you can change that to 3 or 36 months, or turn auto-delete off, from the same Activity page.

For single sensitive questions, use Temporary Chat. Those chats skip your history, are not used for personalization or training, and are kept for up to 72 hours. To stop Gemini from drawing on your older conversations, look under Settings > Personal context > Your past chats with Gemini.

Using a work or school account? Your Google Workspace administrator controls these settings, and you may not be able to change them.

Anthropic Claude: switch off the model-improvement setting

Since late 2025, Anthropic has asked consumer users (Free, Pro and Max plans) to choose whether their chats and coding sessions can be used to train Claude. If you accepted the prompt without changing the toggle, you may be opted in.

1. Click your name in the sidebar to open the menu.
2. Select Settings.
3. Open Privacy.
4. Find the model-improvement toggle and switch it off. Anthropic's help pages currently label it Help Improve our AI models, while some app versions show Help improve Claude.

Why this one matters for retention

This toggle changes more than training. Anthropic says that if you allow training, it may keep de-identified data from new or resumed chats for up to five years. If you opt out, the standard retention period is 30 days.

Deleted chats are not used for future training, and turning the setting off stops your stored chats from being used in future training runs. Data already in a training run or a finished model stays there. Conversations flagged by safety systems may still be used for trust and safety work.

Use incognito chats

Start a new chat outside of a Project and click the ghost icon in the upper right. Incognito chats are not saved to your history or Claude's memory and are not used for training, even if the model-improvement setting is on. Anthropic still retains them for up to 30 days, and Team or Enterprise plans follow the organization's retention policy.

Microsoft Copilot: two training toggles, not one

Consumer Copilot splits training into text and voice, so you need to turn off both. The path depends on where you use it:

1. On copilot.com: select your profile icon, then your profile name, then Privacy.
2. In the Windows or macOS app: select your profile icon, then Settings > Privacy.
3. In the mobile app: open the menu, tap your profile icon, then Account > Privacy.
4. Turn off Training on conversation activity and Training on voice conversations.

Microsoft is clear about the limits. The opt-out applies to future conversations and does not stop your data being used for general product improvement, advertising, safety, security or compliance as described in its privacy statement.

Personalization and memory are a separate control. You can opt out of training and keep personalization on, or turn off both if you do not want Copilot remembering details about you. Work and school accounts using Microsoft 365 Copilot are covered by enterprise data protections instead, and Microsoft says it does not train on that content.

Meta AI: fewer switches, more cleanup

Meta is the outlier. There is no simple, universal toggle to stop your Meta AI chats from being used to improve its AI, and Meta's policies allow it to use those interactions in other ways too, such as personalization. What you can do depends heavily on where you live.

To clear your history in the Meta AI app:

1. Tap Menu in the top left.
2. Tap Settings in the bottom left.
3. Under App settings, tap Data & privacy.
4. Tap Manage your information, then Delete all chats and media, then Delete all.

Deleting removes chats from your history, but it does not undo any use that has already happened.

If you are in the EU or UK, Meta's Privacy Center offers a formal "right to object" request covering the use of your public Facebook and Instagram content for AI training. Elsewhere, the main form is narrower and applies when your personal information shows up in a Meta AI response. Setting your social accounts to private also limits what public content is available.

Where it is offered, Incognito chat in the Meta AI app uses Meta's Private Processing technology. Meta says messages in these sessions are not stored once you exit. You can find related options under Settings > Data & privacy > Incognito chat.

General privacy hygiene for any assistant

Settings only go so far. The habits below protect you regardless of which assistant you use or how its menus change next month.

1. Never paste secrets. Passwords, API keys, recovery codes, full card numbers and government ID numbers do not belong in any chatbot, no matter what the toggle says.
2. Redact before you paste. Replace names, addresses and account numbers with placeholders when asking for help with a document or email.
3. Keep work and personal separate. Use your employer's approved business plan for work material. Those plans usually exclude training by default and are governed by admin retention rules, while a personal account is not.
4. Delete what you do not need. Clear old conversations every few months, and use auto-delete where it exists. Deleted chats are generally removed from company servers on a schedule rather than instantly.
5. Audit connected apps. Assistants that can read your email, files or calendar see far more than what you type. Disconnect integrations you do not use.
6. Recheck after major updates. Policy changes often arrive with a pop-up and a pre-set default. Revisit your privacy page when an assistant announces new terms.

FAQ

Does opting out delete data that was already used for training?

No. Every company covered here says opt-outs apply to future use. Data that was already part of a training run or a finished model generally stays there, which is why changing these settings early matters.

Are temporary or incognito chats completely private?

Not completely. ChatGPT Temporary Chats and Claude incognito chats are kept for up to 30 days, and Gemini Temporary Chats for up to 72 hours, mainly for safety and service reasons. They are excluded from training and your visible history, which makes them a good choice for sensitive one-off questions.

If I turn off training, will the assistant work worse?

Generally no. Training opt-outs do not reduce the quality of answers you get today. The main exception is Gemini, where turning off Keep Activity also disables many connected apps inside the assistant.

Do these settings apply to my work account?

Usually not in the same way. Business and enterprise plans from OpenAI, Google, Anthropic and Microsoft typically exclude customer content from training by default, and retention is controlled by your organization's administrator rather than by you.

The bottom line

Spend ten minutes going through each assistant you use. Turn off the training toggle, decide how long you want history kept, check what memory has stored about you, and reach for a temporary or incognito chat when a question is sensitive. None of these switches is a perfect privacy shield, but together with a strict no-secrets habit, they meaningfully reduce how much of your life ends up in someone else's dataset.