How to Delete What AI Remembers About You: ChatGPT, Claude, and Gemini

To delete what an AI remembers about you, you have to work three separate controls, not one. Every major assistant — ChatGPT, Claude, and Gemini — keeps persistent memory, chat history, and training use as distinct dials, and clearing any one of them does not clear the others. This guide gives the exact steps for each assistant, and explains why deleting a conversation often leaves the memory behind.

The most common mistake is assuming a single "delete" wipes everything. It does not. You can turn memory off and still have every saved fact intact. You can delete a chat and still have the assistant remember what you said in it. You can stop your data being used for training and still have your full history sitting in the account. Knowing which dial does what is the whole task.

The three dials

Before the step-by-step, hold these three controls apart, because the rest of the article depends on the distinction.

• Persistent memory. A separate layer where the assistant stores selected facts about you — your name, preferences, ongoing projects — and reuses them in future conversations. It survives across chats and is the thing people usually mean by "what the AI remembers about me."
• Chat history. The stored transcripts of your past conversations. Deleting a chat removes the transcript, but not necessarily any memory the assistant already extracted from it.
• Training use. Whether your content is used to help improve the provider's models. This is a privacy setting, not a deletion control: turning it off changes future use of your data, it does not erase memory or history.

The practical rule that follows from this: to make an assistant genuinely forget something, you usually have to act on more than one dial. The steps below make that concrete for each product.

ChatGPT: delete saved memories and chat history

ChatGPT's memory is governed by two settings under your profile. Per OpenAI's memory documentation, one is "Reference saved memories" — details you explicitly asked ChatGPT to remember — and the other is "Reference chat history." Both live under the profile icon → Settings → Personalization.

To review or remove stored items, open Settings → Personalization → Manage Memory, per the same documentation. There you can delete individual saved memories or clear all of them. You can also simply tell ChatGPT in conversation to forget something.

The deletion subtlety is the one most people miss. OpenAI is explicit that "turning saved memory off won't delete anything that's already been remembered. And deleting a chat doesn't remove saved memory from that conversation," per the memory documentation. To fully remove something, the same page instructs you to "delete both the saved memory in Settings and the chat where you originally shared it." That is two actions, in two places, for one fact.

Training is a third, independent control. Per the Data Controls FAQ, the "Improve the model for everyone" toggle lives under Settings → Data Controls, and turning off training does not delete your memories or chat history. The same FAQ notes account-level settings apply "to your entire account. It doesn't matter which device you're using," so you do not have to repeat these steps per device.

If you want a conversation that never touches any of this, ChatGPT offers Temporary Chat. Per the Temporary Chat FAQ, Temporary Chats aren't used to train models, don't get saved in history, and don't create memories; they are deleted from OpenAI's systems within 30 days, though they may be reviewed to monitor for abuse.

Claude: pause, edit, or reset memory

Claude's memory is a single toggle you can also inspect and edit. Per Anthropic's chat search and memory documentation, the memory toggle lives under Settings → Capabilities, and "View and edit memory" in Settings shows everything Claude remembers, with edits applying to your next conversation. That view is the surgical option: you can correct or remove specific items without wiping the whole store.

Turning memory off gives you two choices, and the difference matters. Per the same documentation, "Pause memory" keeps existing memory but won't use or make new memories, while "Reset memory" permanently deletes all memories, including project memories, and cannot be undone. Note that each project has its own separate memory space, so a reset reaches across projects.

For a one-off conversation that leaves no trace, use incognito chats. Per Anthropic's incognito chats documentation, incognito chats aren't saved to Claude's memory or chat history, and are marked with a ghost icon.

Gemini: Personal Context and Gemini Apps Activity

Gemini splits the same job across two surfaces. Per the Gemini Apps Privacy Hub, Gemini personalizes by referencing chats in your Gemini Apps Activity, and that referencing can be disabled. Google describes the personalization control on its August 13, 2025 blog post as "Personal context" under Settings → Personal context → "Your past chats with Gemini."

Deletion happens in a different place. Per the Privacy Hub, Gemini Apps Activity (at myactivity.google.com/product/gemini) is where you review and delete activity, change auto-delete, and control whether your data is used to improve Google AI. The same page notes the auto-delete default is 18 months, with options of 3 or 36 months or keeping activity until you delete it manually. Deleting activity there is the closest equivalent to clearing chat history.

Two Gemini specifics are worth carrying into your checklist. Saved Info and Instructions remain until you delete them, per the Privacy Hub — so disabling personalization does not remove them. And for a no-trace session, Temporary Chats "don't appear in your activity and aren't used to personalize or train," per the August 13, 2025 blog post; the Privacy Hub adds that they are retained for 72 hours and are not used for training unless you submit feedback.

Before you wipe: export first

A reset is irreversible, and that cuts both ways. If any of what the assistant remembers is useful — context on a long project, preferences you would rather not re-enter — export it before you delete it. A plain-text copy costs nothing to keep and saves you rebuilding from scratch.

Some assistants make this explicit. Per Anthropic's import and export documentation, Claude's memory can be exported as plain text from settings, which is a sensible step to take before a reset. Where a dedicated memory export is not offered, you can still preserve the substance by copying the relevant conversation text yourself before clearing it.

Separately from these in-product controls, data-protection regimes such as the EU's GDPR give people rights to access and delete the personal data a company holds, and providers publish account-level data controls to support this. The mechanics differ by provider and by where you live; the practical takeaway here is narrower — export anything you want to keep before you pull the trigger, because deletion is meant to be permanent.

A clean-slate checklist

To genuinely reset what an assistant knows about you, run every relevant dial. Deleting only one leaves the others in place.

ChatGPT:

• Settings → Personalization → Manage Memory: delete saved memories (individually or all).
• Delete the original chats where you shared anything you want fully gone — memory and chat are separate deletions.
• Settings → Data Controls → "Improve the model for everyone": turn off if you don't want future content used for training.
• For a no-trace exchange going forward, use Temporary Chat.

Claude:

• Settings → View and edit memory: remove specific items, or use Reset memory to delete everything permanently (this spans all projects and cannot be undone).
• Prefer Pause memory if you want to keep existing memory but stop new memories forming.
• Export your memory as plain text first if any of it is worth keeping.
• For a no-trace exchange going forward, use an incognito chat.

Gemini:

• Gemini Apps Activity (myactivity.google.com/product/gemini): review and delete activity; set auto-delete to the shortest option you want.
• Delete Saved Info and Instructions separately — they remain until you remove them.
• Settings → Personal context: turn off "Your past chats with Gemini" to stop personalization.
• In Gemini Apps Activity, control whether your data is used to improve Google AI.

FAQ

If I delete a conversation, does the AI forget what I said in it? Not necessarily. In ChatGPT, deleting a chat does not remove any saved memory that came from it — OpenAI's documentation says you must delete both the saved memory and the original chat. Treat the transcript and the memory as two separate things.

Does turning off training delete what the AI already remembers? No. Per ChatGPT's Data Controls FAQ, turning off "Improve the model for everyone" does not delete your memories or chat history. Training use and deletion are different dials — you have to act on both if you want both.

What is the fastest way to reset everything an assistant remembers? Use the built-in reset where one exists. Claude offers "Reset memory," which per Anthropic's documentation permanently deletes all memories including project memories and cannot be undone. For ChatGPT and Gemini, clear saved memories and delete history in their respective settings, since there is no single one-click wipe of both.

How do I have a conversation that leaves nothing behind? Each assistant provides a no-trace mode. ChatGPT has Temporary Chat, which per its FAQ creates no memories and isn't saved in history. Claude has incognito chats, which per its documentation aren't saved to memory or history. Gemini has Temporary Chats, which per Google's blog post don't appear in activity and aren't used to personalize or train.