ChatGPT has privacy settings. Most people have never touched them. That's partly because they're scattered across multiple menus, and partly because the defaults are designed to benefit OpenAI, not you. This guide walks through every privacy-relevant setting in ChatGPT, explains what each one actually does, and tells you which ones are worth changing.
If you want the big picture on what ChatGPT collects and whether it's safe, start with our overview of ChatGPT's safety and privacy practices. This guide is the hands-on follow-up: less context, more settings.
The Settings That Actually Matter
ChatGPT has a lot of options buried in its Settings menu. Not all of them are privacy-relevant. Here are the ones that are, in order of impact.
1. "Improve the model for everyone"
Default: On. This is the big one. When enabled, your conversations can be used to train future versions of OpenAI's models. That means the things you type, the files you upload, and the responses you receive can all become training data.
Turning this off prevents future conversations from being used for training. It does not retroactively remove anything already used. Once data has entered the training pipeline, that toggle can't pull it back out.
One nuance that trips people up: even with this setting off, if you click the thumbs up or thumbs down button on a response, you're giving OpenAI permission to use that specific conversation for training. Feedback overrides the toggle. If you've turned off training but still rate responses, you're partially opting back in.
Also worth knowing: turning off this setting does not change how long your conversations are stored. OpenAI still retains them on their servers. This toggle only controls whether your data feeds into model improvement.
2. Temporary Chat
Default: Off. Temporary Chat creates a conversation that won't appear in your chat history and won't be used for model training. It's the closest thing to a private mode that ChatGPT offers on consumer plans.
The catch: OpenAI still stores the conversation on their servers for up to 30 days for safety monitoring before permanently deleting it. It's not instant deletion. It's scheduled deletion.
Temporary Chat is useful when you want to ask something sensitive without it showing up in your history or feeding into AI training. But understand what it doesn't do. It doesn't prevent OpenAI from having access to the conversation during that 30-day window. And it doesn't prevent longer retention if OpenAI determines they need to keep it for legal or security reasons.
Think of Temporary Chat as "don't save and don't train" rather than "private."
3. Memory Controls
Default: On. ChatGPT's Memory feature lets the AI remember information across conversations: your name, where you live, your dietary preferences, your job title, the names of your kids. It has two components.
Saved Memories are explicit facts ChatGPT stores about you. You can view, edit, and delete these in Settings. ChatGPT can create them automatically based on what you say, or you can tell it to remember something specific.
Reference Chat History lets ChatGPT pull context from your past conversations even without saving an explicit memory. When enabled, there's no storage limit on how far back it can reference.
To see what ChatGPT remembers about you right now, open a chat and type "What do you remember about me?" You can also go to Settings → Personalization → Memory to see the full list of saved memories and delete any of them individually.
If you turn off "Reference chat history," ChatGPT will delete all the information it remembered from past chats. That deletion happens on OpenAI's servers within 30 days.
A detail most people miss: memories persist independently of chats. If ChatGPT creates a memory during a conversation and you later delete that conversation, the memory survives. You have to delete it separately. If you want to fully remove something ChatGPT knows about you, delete both the memory and every chat where you mentioned it.
4. Chat History and Deletion
Default: All chats are saved indefinitely. On consumer plans (Free, Plus, Pro), your conversations stay in your account forever unless you manually delete or archive them.
When you delete a chat, it disappears from your sidebar immediately. OpenAI then removes it from their servers within 30 days.
Archiving is different from deleting. Archived chats move out of your main sidebar but remain fully stored in your account with the same retention rules as regular chats. If you want data actually removed, you need to delete, not archive.
You can also delete all chats at once: Settings → Data Controls → Delete all chats. This clears everything from your sidebar and starts the 30-day deletion clock on all of them simultaneously.
5. File Library
Files persist independently of chats. Documents, images, and other files you upload during conversations are stored in your Library and can be used across chats. Deleting a conversation does not delete the files you uploaded during it.
This catches a lot of people off guard. If you uploaded a tax return, medical document, or business plan during a chat and then deleted the chat, the file is still sitting in your Library. You need to go into the Library separately and delete it.
Files uploaded to custom GPTs or Projects are retained until the GPT or Project is deleted. After deletion, the file is removed within 30 days.
6. Atlas Browser Settings
Default: Off for training. If you use ChatGPT's Atlas browser, it can save "browser memories" based on your web browsing activity. These are separate from chat memories and have their own toggle.
OpenAI applies filters designed to keep out personally identifiable information (government IDs, bank account numbers, credentials) and blocks summaries on certain sensitive websites. Web content is deleted right after being summarized, and the filtered summaries are deleted within 7 days.
Browser memories can be managed separately under Settings → Personalization → Reference browser memories. If you use Atlas, review this periodically to see what it's picked up from your browsing.
7. Shared Links
When you share a ChatGPT conversation with a link, anyone with that URL can see the full conversation. Shared links don't expire automatically.
If you've ever shared a chat link that contained personal information, it's worth reviewing your shared links and revoking any you no longer need. The conversation itself isn't deleted when you revoke a link, but the public URL stops working.
8. Data Export
You can request a full export of everything OpenAI has stored about you: conversations, account info, and associated data. The export arrives as an email with a download link. Use this to see exactly what's on file before you delete anything.
Running an export before a major cleanup is good practice. It gives you a local copy of anything you might want to reference later, and it lets you verify what OpenAI actually has stored.
The Privacy Settings Cheat Sheet
If you're short on time, here's the five-minute version:
1. Turn off "Improve the model for everyone" (stops training on your data)
2. Use Temporary Chat for anything sensitive (prevents history and training)
3. Check your memories ("What do you remember about me?") and delete what you don't want stored
4. Clear your File Library of any sensitive uploads
5. Review and revoke shared links you no longer need
What These Settings Don't Do
Even with every setting optimized, there are limits to what you can control in ChatGPT.
Nothing deletes instantly. Every form of deletion in ChatGPT has a 30-day window where your data still exists on OpenAI's servers for safety monitoring. This applies to deleted chats, expired Temporary Chats, deleted memories, deleted files, and even full account deletion.
You can't know exactly what was already used for training. If the training toggle was on for any period of time, conversations from that period may have entered the training pipeline. There's no way to see which specific conversations were used or to remove them from trained models.
OpenAI employees can review conversations. Authorized staff and, in some cases, third-party contractors can access conversations for safety review, abuse monitoring, and model improvement. This is true regardless of your privacy settings.
Ads use conversational context. On the free tier, ChatGPT now shows contextual ads that can be influenced by your current conversation topic. This happens within OpenAI's system and doesn't share data with advertisers, but it means your conversations are being analyzed commercially.
Enterprise and Business Plans
If you use ChatGPT through a work account (Enterprise, Business, Edu, or Healthcare plans), the picture is different. Business data is not used for training by default. Workspace admins control retention policies and access. Enterprise plans offer SOC 2 compliance, SAML SSO, and data residency options.
The consumer privacy settings described in this guide don't apply to Enterprise accounts. If you have both a personal and work ChatGPT account, their privacy settings are managed independently.
The Alternative Approach
All of these settings exist because ChatGPT's defaults favor data collection. You have to actively change multiple settings across multiple menus to get a reasonable privacy posture. And even then, your data sits on someone else's servers for 30 days after deletion.
Some AI tools take the opposite approach by making privacy the default instead of an opt-in afterthought. Ask Safely auto-deletes conversations after 8 hours, never uses your data for training, encrypts everything with AES-256, and gives you a transparent memory system where you see and control exactly what the AI knows about you. No settings to hunt for. No toggles to remember. Privacy is just how it works.
For a detailed comparison of privacy-focused AI alternatives, see our guide to the best private AI assistants in 2026.
Stay in Control
ChatGPT is a powerful tool. For many people, it's worth using even with these privacy trade-offs. But using it well means understanding what you're sharing and making deliberate choices about your data.
Review your settings once, change the ones that matter, and build a habit of using Temporary Chat when you're asking about anything personal or sensitive. That's the most practical thing you can do right now.