Grok's Privacy Settings: A Real Walkthrough

The Beginner's Guide mentions that Grok has privacy controls. It doesn't walk through what each one actually does, where to find it, or which one you need for a specific situation, like a conversation you don't want feeding into Memory, or a chat you want gone entirely. This article is that walkthrough.

The first thing worth untangling is that "Grok's privacy settings" isn't one settings page. Depending on whether you're using grok.com, the Grok app, or Ask Grok inside X, you're looking at a different surface with a different set of controls, and they don't all sync automatically.

Private Chat mode: what it actually turns off

Private Chat is the setting most people reach for first, and it's worth being precise about what it does and doesn't do. Turning it on for a conversation means that chat won't be used to update your persistent Memory, and it won't show up in your regular chat history the way a normal conversation does.

You start one from the private option in the app, and the chat screen describes how that conversation is handled, including any limited retention for safety purposes. Read that screen rather than assuming. What it does not do: it doesn't make the request vanish on xAI's servers, since that short safety retention still applies, and it doesn't retroactively scrub anything Memory already picked up from an earlier, non-private conversation. If you've discussed a topic in a normal chat last week and want to keep this week's follow-up private, Private Chat mode on this new conversation stops this conversation from adding to Memory, but whatever Memory already has from last week is unaffected until you go remove it separately.

Separate from Private Chat, Grok's Data controls settings include a toggle, labeled along the lines of "Improve the Model", that governs whether your conversations can be used to train xAI's models. Labels and locations shift between app versions, so check yours rather than trusting a screenshot from months ago. Turning it off applies going forward and does not pull anything already used in earlier training back out.

Referenced chats: seeing and removing what Memory kept

Because Memory persists across separate conversations by default, at some point you'll want to check what it's actually drawing on when it answers something. The Referenced chats control is where that lives. On grok.com, it lets you see which earlier conversations a given answer pulled from, and remove specific ones from future consideration without deleting the conversation itself.

This matters most when Grok says something in an answer that seems to come from nowhere, referencing a preference or fact you don't remember stating clearly, or stating one you no longer hold. Checking Referenced chats tells you exactly which past conversation it's leaning on, so you can decide whether to leave it, correct it, or remove it.

Deleting conversation history

Separate from Memory management, you can delete individual conversations or clear history outright from your account settings. Deleting a conversation removes it from your visible history. If that conversation had already contributed something to Memory before you deleted it, deleting the conversation itself doesn't necessarily undo what Memory already extracted from it, which is why Referenced chats and conversation deletion are two different tools solving two different problems. One controls what shows up in your history. The other controls what's actively shaping new answers.

If your goal is "make sure nothing from this conversation influences anything going forward," check Referenced chats for anything it fed into Memory, in addition to deleting the conversation itself.

X versus grok.com: two different privacy surfaces

This is the part that trips people up most. If you use Ask Grok on X, that interaction runs through X's own product surface, with X's own account settings governing things like who can see your posts and how your activity on the platform is used. It is not the same settings panel as grok.com or the standalone Grok app, even though it's the same underlying model answering you.

Concretely, that means turning on Private Chat mode inside the grok.com app has no effect on how Ask Grok behaves on X, and adjusting your X privacy and data settings doesn't touch your Memory settings on grok.com. If you use both, you're managing two separate configurations, not one setting that covers everything.

One week, four situations

Abstract descriptions of controls are easy to nod along to and forget. Here is a single invented week for Priya, a freelance designer who uses Grok on grok.com for client work and Ask Grok on X for keeping up with her industry. Each situation calls for a different control.

Monday: a health question. Priya wants to ask about a persistent shoulder pain and does not want it turning up as context in a client brief three weeks later. She turns Private Chat on before typing anything. The point is the timing: the switch protects the conversation it is on from the start, and does nothing for a conversation that already happened.

Wednesday: a strange assumption. Grok drafts a project outline and assumes she prefers formal, corporate phrasing. She left that behind two years ago. She asks what it is basing that on, opens Referenced chats, and finds one old conversation from a former client's project. She removes it from future consideration. She keeps the conversation itself, because the client work in it is still useful to her.

Thursday: something she wants gone. She pasted a draft contract into a chat last month and no longer wants it in her history. She deletes the conversation. Then, because deleting the conversation is not the same as removing what Memory learned from it, she checks Referenced chats for anything that conversation fed.

Friday: the other surface. She tightens things on grok.com, then remembers she also asks Grok questions on X. Those settings live in X's own account settings, so she reviews them separately.

1. 1

   Private Chat, switched on first: Stops this one conversation from updating Memory or appearing in regular history. It has to be on before the sensitive part, not after.

2. 2

   Referenced chats: Shows which earlier conversations shaped a specific answer, so a surprising assumption can be traced instead of guessed at.

3. 3

   Remove from future answers: Takes one conversation out of consideration without deleting it, useful when the chat is worth keeping but its influence is out of date.

4. 4

   Delete conversation: Removes it from your history. It does not, on its own, undo anything Memory already extracted, which is why Priya rechecks Referenced chats afterward.

Why each control matters, in one line

A reasonable default setup

For most people, a sensible baseline looks like this: leave Memory on for genuinely useful ongoing context, like your writing preferences or recurring project details, since that's the feature actually earning its keep day to day. Reach for Private Chat mode specifically for one-off conversations that touch something sensitive you don't want shaping future answers, a legal question, a health question, a draft you don't want referenced later. Check Referenced chats occasionally, not just when something looks wrong, so you have an accurate sense of what's actually accumulating. And if you use Ask Grok on X regularly, review your X account's own privacy settings on its own schedule rather than assuming your grok.com choices carry over.

None of this requires obsessive management. It requires knowing which control does which job, since guessing tends to leave the actual gap unaddressed while you tighten a setting that was never the problem.

If you're still getting oriented with Grok generally, the Complete Beginner's Guide to Grok covers the basics these settings sit on top of.