ChatGPT has added more privacy controls over the past two years than at any point in its history. Training opt-outs, temporary chats, persistent memory toggles, data export, and deletion requests are all now available in some form.
The problem is not that these controls do not exist. The problem is that most users do not know where to find them, what they actually do, or — critically — what they leave unprotected.
This guide covers every privacy setting available in ChatGPT as of 2026, explains precisely what each one does and does not do, and identifies the structural gap that no OpenAI privacy setting can close.
---
ChatGPT Training Data Opt-Out
The most consequential privacy setting in ChatGPT is the training data opt-out. By default, OpenAI uses conversations from free and Plus users to train and improve its models. This means that when you describe your medical symptoms, paste in your company's internal strategy document, or work through a personal legal situation, that content can — by default — become training material.
How to turn it off
Navigate to Settings > Data Controls and toggle off "Improve the model for everyone." This disables the use of your future conversations for model training.
What this does:
What this does not do:
Enterprise and API users are excluded from training by default. If you are using a personal ChatGPT account — free or paid — this toggle needs to be set manually.
The OpenAI API difference
If you or your organisation accesses ChatGPT via the API rather than the consumer product, OpenAI's policy is that API inputs and outputs are not used for training by default. This is a meaningful structural difference, but it requires building your own integration rather than using the standard ChatGPT interface.
---
Temporary Chats: What They Actually Do
Temporary chats were introduced as a session-level privacy mode. When you start a temporary chat, ChatGPT will not save the conversation to your history, and the chat will not be used to improve OpenAI's models.
Where to find it
Select "Temporary chat" from the model selector dropdown at the top of the ChatGPT interface, or toggle it on from the sidebar.
What temporary chats do:
What temporary chats do not do:
Temporary chat is useful for one-off sensitive queries where you do not want a permanent record in your account history. It is not the same as a private or encrypted session. Your ISP, network proxy, or employer (if using a corporate network) can still observe that you are sending requests to OpenAI's servers, even if the content itself is HTTPS-encrypted in transit.
---
ChatGPT Memory Controls
ChatGPT's persistent memory feature allows the model to remember facts about you across conversations. It can store your name, profession, preferences, ongoing projects, and any other detail you — or ChatGPT itself — decides is worth retaining.
This is genuinely useful. It is also a centralised cloud store of personal information associated directly with your OpenAI account.
Managing memory
Go to Settings > Personalization > Memory to:
When memory is enabled, ChatGPT automatically decides what to save. You can also explicitly ask it to "remember" or "forget" specific things during a conversation.
What the memory setting controls
Turning off memory stops ChatGPT from storing new memories and prevents existing memories from influencing future conversations. It does not delete memories that have already been saved — you need to do that manually via "Manage memories" or "Clear all memories."
The structural issue with cloud-stored memory
The core limitation of ChatGPT's memory system from a privacy standpoint is architectural: your personal context is stored on OpenAI's servers, associated with your account. It is subject to OpenAI's data retention policies, potential data breaches, government requests, and future policy changes.
If you delete your OpenAI account, OpenAI's documentation states that it will delete your personal information within 30 days, with some exceptions for information it is legally required to retain. The stored memories go with the account — but the timing and completeness of deletion are outside your control.
Custom instructions — the system prompt you can set to shape ChatGPT's behaviour — are similarly stored server-side and persist until you manually clear them.
---
Data Export and Deletion
OpenAI provides two mechanisms for acting on your stored data: export and deletion.
Exporting your data
Go to Settings > Data Controls > Export data. OpenAI will email you a download link (typically within a few hours) containing:
This export is useful for auditing what is stored, migrating conversation history to another tool, or retaining a local copy before account deletion.
Deleting your account and data
Go to Settings > Data Controls > Delete account. This initiates account deletion and requests removal of associated personal data. Per OpenAI's privacy policy, data is typically deleted within 30 days, subject to legal retention obligations.
There is no granular "delete all my conversations but keep my account" option in the standard interface. You can delete individual conversations from your chat history, but bulk deletion requires either manually removing each conversation or deleting the account entirely.
Data deletion requests without account deletion
OpenAI's privacy policy includes a data subject rights process. You can submit a request to privacy@openai.com to request deletion of specific data without closing your account. Response timelines vary, and the scope of what can be deleted is constrained by OpenAI's legal obligations.
---
What ChatGPT Privacy Settings Don't Protect
Understanding the limits of these controls is as important as knowing how to use them.
Your data is still processed in the cloud. Every message you send to ChatGPT — regardless of your privacy settings — travels to OpenAI's servers for processing. The opt-out and temporary chat controls determine what happens to that data after it is processed, not whether it leaves your device.
Sensitive context accumulates invisibly. Even if you never explicitly share sensitive information, the pattern of your queries over time reveals a great deal. The medical questions you ask, the legal situations you describe, the business decisions you work through — these create a profile that exists entirely in the cloud.
Your context is tied to a corporate account. Everything you tell ChatGPT is associated with your OpenAI account. That account is subject to account takeover risks, subpoenas, regulatory investigations, and OpenAI's own policy evolution. The fact that you have opted out of training does not mean your data has zero exposure.
There is no client-side processing. ChatGPT does not offer a mode in which your messages are processed locally before being sent to the model. All inference happens on OpenAI's infrastructure.
Memory controls require active management. Unlike a system that never stores data in the first place, ChatGPT's memory system requires you to proactively manage what it retains. Most users do not audit their stored memories regularly.
None of this is a criticism of OpenAI's privacy controls specifically — these are structural properties of cloud-based AI assistants as a category.
---
Filling the Gap: Local Shadow Protection
The gap that ChatGPT's privacy settings cannot close is the gap between the data leaving your device and the AI processing it in the cloud.
There is an alternative architectural approach: instead of sending your real personal context to the AI, you keep the real data on your device and send a shadow value — a privacy-preserving substitute — in its place. The AI gets enough context to be useful. Your real name, your client's name, your actual salary figure, your medical details — none of it ever leaves your device.
This is what MyYaad does. Your vault of personal context lives locally, encrypted on your device. When you use an AI assistant through the MyYaad extension, the shadow engine substitutes realistic but fake values for sensitive entries before the request is sent. The AI works with the shadow. You see the real answer translated back.
The difference is structural. ChatGPT's training opt-out is a policy control — it changes what OpenAI does with your data after they have it. Local shadow protection is an architectural control — it changes whether they receive your real data in the first place.
Both matter. But only one of them protects you before the data leaves your device.
If you want to keep using ChatGPT — with all the privacy settings configured as tightly as possible — and add a layer of protection that operates before your data reaches OpenAI's servers, download MyYaad and pair it with your browser.
---
Quick Reference: ChatGPT Privacy Settings
| Setting | Location | What It Does | What It Misses | |---|---|---|---| | Training opt-out | Settings > Data Controls | Stops future chats from being used for training | Data still processed in cloud; no retroactive removal | | Temporary chat | Model selector dropdown | Excludes session from history and training | Content still transmitted to and retained by OpenAI (up to 30 days) | | Memory off | Settings > Personalization | Stops saving new memories | Existing memories remain until manually deleted | | Data export | Settings > Data Controls | Download all stored data | Point-in-time snapshot only; does not delete data | | Account deletion | Settings > Data Controls | Requests full data removal within 30 days | Subject to legal retention exceptions |
Using all of these settings together represents the most privacy-protective configuration available within ChatGPT. It is still a cloud service. Your messages are still processed remotely. The settings determine what OpenAI retains — not what they receive.
That distinction is where local shadow protection begins.