Every major AI assistant now has some form of memory. They remember your name, your preferences, the context you’ve shared about your life, your work, and your habits. That memory makes them more useful. It also creates a record of your personal information that exists on someone else’s servers, subject to their retention policies and security practices.
Knowing how to delete that memory — and understanding the limits of what deletion actually achieves — is now a practical skill for anyone who uses AI assistants regularly.
Why AI Memory Is Different from a Conversation
Earlier AI chatbots treated each conversation as fresh. When the session ended, nothing carried over. Modern AI assistants work differently.
ChatGPT’s memory feature stores facts about you explicitly — things you tell it to remember and things it decides to save on its own. Gemini builds context from your Google account activity and conversation history. Microsoft Copilot integrates with your Microsoft account and organization data. Claude uses project-level context that persists across conversations.
The practical result is that these systems accumulate a profile: your name, profession, family situation, health concerns, financial context, creative projects, and anything else you’ve discussed or that was inferred from the conversation. Some of this is stored in memory systems you can see and delete. Some is embedded in conversation history. Some may have been used in model training before you could act on it.
These are different things, and they require different approaches.
The Right to Erasure
Under GDPR, users have the right to request deletion of personal data. Under CCPA, California residents have the right to request deletion. These legal frameworks apply to the personal information stored in AI systems just as they apply to other services.
In practice, all major AI platforms provide some mechanisms for deletion in compliance with these frameworks. The challenge is that the mechanisms are sometimes incomplete, not always easy to find, and in some cases leave data in technical systems even after you’ve exercised deletion rights.
As of March this year, 32 European data protection authorities launched a coordinated investigation specifically into whether companies can actually prove they delete personal data when users ask. This investigation was triggered in part by concerns about AI systems — because AI creates data in more places and in less predictable ways than traditional software.
The European Data Protection Board named the right to erasure as its primary enforcement priority for this year. That’s a signal that deletion compliance is not yet reliable across the industry.
How to Delete Your Memory from Major AI Platforms
ChatGPT (OpenAI)
ChatGPT’s memory system has two components: explicit memory (facts the model stored based on your conversations) and conversation history (the actual chat logs).
To manage and delete explicit memory:
- Open ChatGPT and click your profile icon
- Go to Settings → Personalization → Memory
- Click “Manage memory” to see what the model has stored
- Delete individual entries or click to clear all memory
To delete conversation history:
- Settings → Data Controls
- Under “Chat history and training,” you can disable future saving
- “Delete all chats” removes your conversation history
Note that disabling chat history also disables the memory feature. OpenAI’s privacy policy describes a retention period for deleted data during which it may remain in technical systems for up to 30 days before full removal.
For full data export and deletion: OpenAI’s privacy page allows you to export all your data or submit a deletion request for your account. Account deletion is permanent and removes stored data, though there is a processing window.
If you want to prevent your conversations from being used for model training going forward, the Data Controls setting has an option to opt out. Note: this applies to future conversations. Past conversations that were used in training are not recoverable.
Google Gemini
Gemini’s memory draws from Google account activity, conversation history, and Google Workspace data for users with workspace integrations.
To manage Gemini activity:
- Go to myactivity.google.com
- Filter by “Gemini” in the product list
- Delete individual activities or choose “Delete activity by” to remove by time range
To turn off activity saving:
- Open Gemini settings in the app or at gemini.google.com
- Navigate to “Gemini Apps Activity”
- Toggle off to stop saving new activity
Google’s auto-delete settings allow you to set automatic deletion at 3, 18, or 36 months. Setting this to the shortest period limits accumulation.
For Gemini Extensions (workspace integrations): If you’ve connected Gemini to Gmail, Drive, or other Google services, review and revoke those connections at myaccount.google.com → Security → Third-party apps with account access.
Microsoft Copilot
Copilot for personal use and Copilot for Microsoft 365 (enterprise) have different data practices. These instructions apply to personal use.
To clear Copilot conversation history:
- Open Copilot at copilot.microsoft.com
- Click the settings icon
- Select “Privacy settings”
- Use “Clear conversation history” to remove chat logs
To manage your Microsoft privacy dashboard:
- Go to account.microsoft.com/privacy
- Review and delete activity data associated with Copilot
- Under “AI product improvements,” you can opt out of contributing your interactions to model improvement
For enterprise users: Your organization’s Microsoft 365 admin controls data retention, and deletion through personal settings may not remove data from organizational systems. Contact your IT department or review your organization’s Copilot data governance policy.
Claude (Anthropic)
Claude’s Projects feature allows persistent context to carry across conversations. This is distinct from conversation history.
To manage Project context:
- Open the Project in the Claude interface
- Review the instructions and documents stored in the Project
- Delete or modify the Project to remove personal context you’ve added
To delete conversation history:
- In the conversation list, hover over a conversation
- Use the delete option to remove individual conversations
- Anthropic’s settings do not offer a single “delete all conversations” button as of this writing — conversations must be deleted individually or through an account deletion request
For full data deletion: Anthropic’s privacy policy provides a mechanism to request deletion of your personal data through their privacy request form. This covers conversation history and any personal data associated with your account.
Third-Party AI Memory Apps
Several standalone AI memory applications — tools designed specifically to remember information about you across different AI platforms — have their own data practices to consider.
These tools often require broad permissions to do their job: access to your conversations, your documents, your calendar, or your communications. Reviewing and revoking those permissions, and requesting data deletion through the app’s privacy settings, applies the same principles as the major platforms above.
The risk with smaller AI memory apps is that their deletion practices and security postures are less well-established than the major platforms. If you’ve used any such app, review its privacy policy specifically to understand what it retains and where.
What Deletion Does and Doesn’t Accomplish
Deleting memory and conversation history from your account removes the data from the user-facing systems. This is meaningful and worth doing.
It does not, in most cases, undo any use of your data in model training that occurred while you were using the service. If your conversations were used to improve a model before you deleted them — and before you opted out of training contributions — the model has learned from that data. Deleting the conversation records doesn’t un-train the model.
This limitation is inherent to how large language models are built. Once a model is trained on a dataset, removing specific examples from that dataset doesn’t change the model’s weights. The practical effect of your data on a trained model is very difficult to isolate and reverse.
Some AI providers are explicit about this. Others are not. Reading the relevant section of each provider’s privacy policy — specifically the part about model training — tells you what you’ve agreed to and what deletion actually covers.
The Ongoing Data Minimization Practice
Deletion is reactive — you’re removing data that already accumulated. The proactive version is minimizing what you share in the first place.
Don’t use AI assistants for highly sensitive processing. Tax documents, medical records, legal correspondence, and identity documents sent to AI assistants become part of the conversation record and potentially the training data. Use AI for lower-sensitivity tasks and handle sensitive documents in dedicated secure systems.
Review what’s in AI memory periodically. Both ChatGPT and Gemini allow you to see what they’ve stored. Making a habit of reviewing and pruning that memory every few months prevents excessive accumulation.
Turn off training contributions where available. All major platforms provide opt-out mechanisms for using your data in model training. These should be configured when you set up an account, not after years of use.
Understand that “private” browsing doesn’t help. Incognito or private mode prevents browser-level tracking but doesn’t change how AI platforms store conversation data server-side.
Storing Personal Information That Doesn’t Need to Be in AI Systems
Not everything you want to remember needs to be in an AI memory system. Personal documents, photos, journals, and records that you want to access across time belong in storage you control — not in AI systems that build persistent profiles from them.
Daftei is designed for this kind of private personal storage. It doesn’t build a memory profile from your content, never trains third-party AI on your files, and doesn’t sell data. Files are encrypted with AES-256 at rest and protected in transit with TLS 1.3. The platform is GDPR and CCPA compliant.
If you decide to close your account, a 30-day grace window is followed by permanent, irreversible erasure of your data. That’s a deletion guarantee that AI memory platforms — where training use is often retained even after account deletion — can’t make in the same terms.
Taking Stock of What You’ve Already Shared
The most useful thing you can do today is audit your AI tool usage. Which AI assistants have memory enabled? What do they actually have stored? When did you set up your accounts, and have you changed any privacy settings since?
The steps above give you the practical paths to deletion. The broader discipline is treating AI memory as a real form of data storage — subject to the same scrutiny as any service that holds your personal information — rather than as a transient experience that disappears when you close the chat window.
It doesn’t disappear. Now you know how to make it.