How to Turn Off Training Use Before You Share Company Materials With AI
ChatGPT, Claude and Gemini each apply different rules on using your conversations for training, depending on your account type and settings. Here is how to change the settings on a personal account, and what to check separately on a company one.
Before you paste company materials or personal data into an AI chatbot, check two things: whether the account you are using is a personal or a work account, and whether the setting that lets your conversations be used for AI training is switched on. Turning training off does not immediately delete your conversations. If the material is work related, it is best to confirm what your company allows before you type anything in.
Training and retention are two different things
“Training use” means your conversations are used to make the AI model better. That is a separate matter from whether a record of your conversations is kept. It is the difference between keeping a notebook and using what is written in it as classroom material.
With a personal ChatGPT account you can keep your chat history while switching off training use only. With Gemini, even if you turn Keep Activity off, later conversations may still be stored for up to 72 hours for purposes such as providing and protecting the service. So when you change settings, look at “is this used for training?” and “how long is this stored?” as two separate questions.
The setting paths and handling rules described here are current as of September 2026. Even within the same service, the policies that apply to consumer and enterprise versions differ. Once you confirm which account and workspace you are in, it also becomes clear which policy you need to read.
Where to turn it off in ChatGPT
For OpenAI’s personal ChatGPT accounts, Free, Plus and Pro, the setting that uses your conversations for training is on by default in a personal workspace. That applies to paid personal accounts too. Do not assume that paying means you are excluded from training.
On the web, here is how to change it.
- Open Settings from your profile.
- Go to Data controls.
- Turn off “Improve the model for everyone”.
New conversations you start after switching it off are not used for model training. You do not have to turn off chat history as well. You can leave the ability to look back at earlier conversations intact and stop only the training use of new chats.
What does the Claude setting cover?
For Anthropic’s personal Claude accounts, Free, Pro and Max, conversations and coding work for which the user has allowed model improvement may be used for training. The same setting covers not just text conversations but coding work as well.
Here is how to change it.
- Open Settings from the account menu.
- Go to Privacy.
- Turn off “Help Improve our AI models”.
New conversations and coding work after you change the setting will not be used for future model training. However, it does not apply retroactively to training already under way or to models that have already finished training. That is why turning the setting off should not be understood as a way to undo past training.
Different rules may apply to conversations flagged for safety review and to feedback that users submit. Switching off the training setting does not eliminate all of these exceptions.
What changes when you turn Gemini’s setting off
For Google’s personal Gemini accounts, when Keep Activity is on, your conversations and the content you share may be used to improve the service and to train AI models. Some of this is reviewed by humans.
In Gemini on the web, here is how to change it.
- Open Settings & help.
- Go to Activity.
- Select “On” at the top, then switch it to “Turn off”.
Once Keep Activity is off, later conversations are not used to train AI models unless you submit feedback. Storage for purposes such as providing and protecting the service can still continue for up to 72 hours. Some connected apps also become unavailable, so it is worth checking in advance whether the connections you normally rely on will be affected.
The thing to remember with Gemini is that a single activity setting changes both training use and connected features. You also need to check whether you have submitted feedback.
Is it all the same on a company account?
Enterprise accounts come with different default policies from personal accounts. But signing in with a company email address is not by itself a reason to relax. You need to check which product you are using and how your organization has configured it.
| Enterprise service | Training rules and what to check |
|---|---|
| ChatGPT Business, Enterprise | Inputs and outputs are not used for model training by default. |
| Claude for Work, including Claude Team and Enterprise | Inputs and outputs are not used for training by default. If the organization grants separate permission, or if a user submits feedback or reports a bug, that content may be used for training. |
| Gemini in enterprise Google Workspace | In versions where protections apply, content is not used to train generative AI models outside the organization, and is not reviewed by humans, without permission. |
With Gemini in particular, which Workspace edition you use matters. If you use the Gemini app with a Workspace account that is not covered by those protections, the consumer privacy rules may apply instead. For work and school accounts, activity settings are managed by the administrator. Hunting for the setting yourself is not enough; you need to ask your administrator which policy applies.
What to check before you type
Before entering company materials or personal data, run through the following.
- Confirm whether the account and workspace you are using are personal or enterprise.
- If it is a personal account, review that service’s training use setting.
- Check when a setting change takes effect and whether there are exceptions related to feedback.
- For company material, check which services and what kind of input your organization permits.
- Leave out any personal data the question does not require.
The training use setting is a way to control the extent to which your conversations feed model improvement. The important thing is not to assume that flipping one switch settles every aspect of how your data is handled. Next time you revisit these policies, it will be easier to find what you need if you separate the questions: training use, retention period, feedback exceptions, and which enterprise products are actually covered by protections.