Grok on X and Grok.com have different settings. If you use both, check both before assuming your conversations won’t be used for model training.
Turning off training also does not remove an already shared chat link or clear your history. Decide which of those outcomes you want before changing the settings.

Overview
Start where you actually use Grok
If you open Grok from the X sidebar or app, start in X privacy settings. The control applies to specified content and interactions associated with using Grok through X.
If you open Grok.com or the standalone mobile app, use its own data controls.
Even if you chose to sign in with X, confirm the preference inside the standalone service.
Note the two ways you access Grok, then open each one. For instance, you might talk to Grok in X and paste an attachment into Grok.com.
Treat the brand name as a clue, not as evidence that one switch manages every account and session. Verify the actual setting where you use the product.
Training, personalization and visibility are three different things
A training choice concerns whether information can be used to build or refine models. A personalization choice concerns how a service uses information to tailor the experience you receive.
Who can see your chats is a separate question. Someone with a shared chat link might be able to open it, regardless of what your model improvement switch says.
Similarly, a post you made in public on X is public until its visibility changes. Switching off training is not a way to make a public post private.
Pick the control that matches the problem
- Use the training switch to control the use of covered information for model development.
- Turn off personalization separately if you also want the service to stop using covered information to tailor the Grok experience on X.
- Delete the conversation if your goal is to remove it from your own history and request the service’s documented deletion process.
- Revoke a public share link separately when you no longer want it to provide access.
- Avoid feeding sensitive details into the tool merely to test a privacy setting.
Turn Off Training for Grok on X
Find Grok and third-party settings within X
Open Settings and privacy, then Privacy and safety. Under data sharing and personalization, check Grok & Third-party Collaborators.
Look for the choice about using your public data and your Grok interactions, inputs, and results for model training and fine-tuning.
Turn that choice off if you want to opt out.
X documents the wording and scope on its Grok support page. If the menu has moved in your app, compare the current setting with that page before toggling it.
When you return to the screen, confirm the option stayed off. If you maintain more than one X account, check the account you actually use with Grok.
Check training separately from personalization
The same section can contain a different control for personalizing your experience with Grok. Turn that off too only if you want to change that separate use.
Leaving personalization on does not mean you forgot to turn off training. Conversely, turning off personalization is not the same action as opting out of model training.
X distinguishes training from the operation of Grok-powered features such as recommendations. It says the training choice does not prevent a deployed model from learning through ordinary use.
Avoid reading a single switch as a promise that X won’t process any information from a tool you still use. Read what that switch specifically covers.
Pay attention to how you use Grok in X Chat
X directs Grok Bot users in X Chat to the same account privacy area. Check the setting there before looking in a standalone account.
Turn off the training option in the X account that uses the bot. X says you can continue using the bot with that option off.
This does not make the content of a conversation yours to disclose. If it includes somebody else’s details or work information, consider whether you should share that content at all.
Public posts and feedback still deserve separate care
Protecting X posts changes who can generally see them going forward. It does not recall a screenshot somebody already made or remove copies that were already shared.
X also says feedback through thumbs-up or thumbs-down can allow the associated conversation and related information to be used for training or improvement even if you opted out.
If your reason for turning training off is that a conversation contains sensitive material, do not rate it as a test. Closing the conversation is a better choice.
Turn Off Training in the Standalone Grok App or Website
Find Improve the model inside the standalone product
On Grok.com, open Settings, then Data. In the mobile app, open Settings, then Data Controls. On either, turn off Improve the model.
The provider’s consumer FAQ documents separate browser and mobile paths for this setting. Check the option on the account you use for those conversations.
The FAQ says the choice covers new conversations after opting out. It also warns that feedback you voluntarily submit may still be used for model training.
If you use the standalone product without signing in, check the provider’s regional rules. The FAQ says a training opt-out is unavailable to unsigned users in some regions.
The question is not whether the standalone app opens with your X login. It is whether the actual account and data controls you are using show the preference you want.
If you have a second personal account or switch between profiles, make sure you did not change a setting in the wrong place. Verify again from your normal entry point.
Private Chat is not just a training toggle
The standalone provider says Private Chat conversations do not appear in history and are removed from its systems within 30 days, subject to documented exceptions.
Check the exceptions in the standalone privacy policy for situations in which information may be retained longer, including for legal, compliance, or safety reasons.
A chat that is not listed on your screen is therefore not evidence of immediate deletion.
Don’t use Private Chat as a reason to paste a credential or another person’s documents.
The standalone FAQ says Private Chat content isn’t used for model training where the mode is available. That is a separate claim from the time the service may retain it.
If you need a record of a sensitive conversation, avoid copying it into several places while trying to manage it. Decide where your own legitimate copy should live, if anywhere.
Find and revoke links that already exist
The standalone FAQ says you can manage public share links through the Grok.com share-link list. It describes a Remove option alongside individual links.
Use that route if you previously gave other people a link and do not want it to remain accessible. A training opt-out does not perform that action for you.
Revoking a link cannot make somebody forget what they read. If they copied the conversation, the provider’s link settings do not remove that independent copy.
If you don’t remember sharing a particular chat, check the list before assuming. A link created for one colleague may have been sent to others after you shared it.
What to Do if You Already Pasted Something Sensitive Into Grok
- Identify where you sent it: the standalone app, Grok.com, or Grok through X. Note the date and topic without copying the sensitive content into another service.
- Change the training preference where you sent it and check the other service separately if you use both. Avoid submitting feedback on the chat as a test.
- Check the conversation history and delete the relevant entry where supported. If you shared a public link, review the link list and revoke that link as well.
- If you shared a password, token, or similar secret, replace it through the service that issued it. Do not wait to see whether deleting the chat is enough.
- If the information concerns work or somebody else, follow the relevant reporting process. For a privacy request about Grok itself, use the official provider route in its consumer documentation.
Match your response to the kind of information
An API token can be revoked and replaced. A private address, medical detail, or family document can’t simply be rotated, so focus on limiting further disclosure.
If your employer provided the document, check who handles unintended sharing. They can tell you whether the conversation, recipient, or link needs to be documented.
Do not send the same sensitive text in full when asking how to delete it. A reference to the account, timing, and type of content may be enough to start.
If you later realize you posted the information publicly on X as well, address both copies. Deleting a Grok conversation will not erase a separate public post.
Test the setting without creating new exposure
After changing things, reopen the relevant settings and read the current switch state. Do not paste a second copy of the information just to check how Grok responds.
If you regularly move between a personal X account and a Grok account used for work, label them privately. That helps you catch changes made in the wrong session.
Check the history and share links independently. A clean history view does not tell you that an older public link was revoked or that training choices were changed.
What to expect from a later deletion
X and standalone Grok publish different instructions for controlling chat history. Follow the instructions for the product you used rather than assuming its retention policy matches the other.
The provider’s documentation describes conditions and retention limits, not a tool for seeing whether one past prompt was used to train a model. Avoid inferring that from a settings screen.
If you need a formal answer about data handling, follow the provider’s privacy process. Describe the result you want, such as removal of a conversation or revocation of a link.
Reduce exposure before the next conversation
If you use Grok to polish a document, remove customer names, direct contact details, and internal identifiers before copying text.
Keeping unnecessary information out is stronger than changing a setting afterward.
A sample paragraph with invented details can be enough for help with tone or structure.
Do not give it a real invoice or an unredacted email when a short excerpt suffices.
If there is a workspace policy about AI tools, check it before uploading anything. A personal privacy preference does not give you permission to include somebody else’s information.
If you ask about a real case, remove enough detail that a reader cannot easily infer who it concerns.
Changing just one name may leave the rest of the context recognizable.
Distinguish a preference that applies to a consumer account from instructions for an enterprise account. The standalone consumer FAQ expressly addresses ordinary users of the website and mobile app.
When you switch products, use the policy and controls for that exact product. Do not carry a claim about one tool’s retention into a conversation hosted somewhere else.
If you shared a transcript directly instead of generating a link, record where it went. A chat-link control can disable the link, but cannot reach a file you already sent.
Frequently Asked Questions
Does turning off Grok training in X also cover Grok.com?
Check the setting in both products if you use them. Standalone Grok documents its own Improve the model setting, separate from the X instructions.
Using an X sign-in does not eliminate the need to verify the standalone preference. Check the account that actually holds the conversation you’re concerned about.
Can I still use Grok after I opt out?
X says you can still use Grok on X after turning off training. The standalone controls likewise describe a choice about model improvement, not a command to close your account.
You can review other settings separately if your concern is personalization or conversation retention. Don’t assume training is the only data-related option you’ll need.
Does this remove public posts from X?
The training switch does not change whether a post is public. Review its visibility, or remove the post if that’s what you intend.
Private settings won’t retrieve copies other people saved earlier. Be cautious about treating a current visibility setting as an answer about past exposure.
Can I use a thumbs-down button after opting out?
X says voluntarily rating an answer may allow the associated conversation to be used for training or improvement despite opting out. Consider that before rating a sensitive conversation.
If all you want is to manage the data, use the privacy or history controls instead. Do not send the content again as part of a feedback test.
Does Private Chat mean the data is never retained?
The standalone service says Private Chat keeps the conversation out of normal history and that removal is generally within 30 days.
That window still has important exceptions described in its privacy policy.
That is not a guarantee of instant erasure. Treat it as a mode with a specific privacy policy, not as a replacement for deciding what should never be uploaded.
Will deleting a conversation remove it from past training?
Neither the history controls nor the training preference gives you a way to verify whether a particular past conversation was used to train a model.
Use them for the choices they actually provide. If you also shared the conversation, revoke the link and remember that any saved copy is a separate issue.
The Bottom Line
Check Grok’s training switch on X and inside the standalone product if you use both. Then handle personalization, old chats, and shared links as separate decisions.
If a conversation already contains a secret, the fastest useful action is often to replace that secret where it was issued. Reviewing Grok settings is not a substitute for that.
Reviews & discussion
0 commentsShare a firsthand experience, or ask a question.
Community reviews
grok.com
Moderated reader experiences, not an editorial verdict.
Start the conversation
Have a question or something useful to share? Add the first comment.