Try this: Log in to ChatGPT or Meta AI and type, “roast me based on my chat history.” If it knows you well enough to skewer you, that’s one sign it’s time to lock down your artificial intelligence privacy.
Most AI companies keep a file on everything you say to their bots. Your words, pictures, clicks and ideas help make their AI smarter. They use it to personalize bot responses so you stick around longer. And your chats fuel their other businesses, too: This week, Meta is starting to use what you share with its bot to target you with ads on Instagram and Facebook.
Give me 15 minutes, and I can help you protect yourself. I dug through the controls for the most popular AI tools and picked the worst default privacy settings you should change. The links and photos below will take you directly to what to change for OpenAI’s ChatGPT, Anthropic’s Claude, Microsoft’s Copilot, Google’s Gemini and the Meta AI.
Why should you care if an AI company has a file on you?
- You might not want ads following you around based on your more intimate conversations.
- When AI companies train their products on your data, we don’t know how well they edit out names, faces, addresses and other personal information — or how often they leak back out in AI answers.
- AI agents, which complete tasks on your behalf, can fall into traps or go rogue and do real damage with your credit card, passwords or personal information.
- Lawyers and governments can request your chats as evidence or leads.
Bots are also now building so-called memory files based on your chats to tailor their responses. But they get things wrong — my ChatGPT memory claimed I owned a spin studio. They can remember things you’d rather they forget, like a private medical concern. And bots pretending to be a friend that knows you can manipulate people and even fuel mental illness.
I assembled the settings below in consultation with three authorities on AI privacy: Jennifer King at the Stanford Institute for Human-Centered AI, Jacob Hoffman-Andrews at the Electronic Frontier Foundation and Ben Winters with the Consumer Federation of America. All of the bots have privacy issues. The basic rule of thumb: Don’t let a chatbot record anything you wouldn’t want to see on social media.
(Tech companies like to move around settings, so if you see one is off, send me an email.)
Protecting your privacy in AI tools comes with some trade-offs. The biggest: The less a chatbot knows about you, the less personally useful its responses might be. But I’d argue bots are not friends — and I can recommend ways to get the most out of AI without the surveillance.
If you do one thing
There’s one move that can help protect you even if you change no other settings: use “temporary chats.”
These are kind of like “private” mode in a web browser. Temporary chats stop the conversation from being saved to your history, adding to a bot’s memories of you or being used to train AI in the future.
Use temporary chats for anything even a little bit sensitive, including medical, financial and relationship questions. The downside: Your chat won’t be saved, if you want to reference it or continue the conversation later.
Here’s what the buttons look like:
Unfortunately, Meta AI has no temporary chat button, but you can log out of the app or website and ask questions in a way that won’t be logged to your account. The same goes for the personal version of Copilot.
Now let’s change a few settings that will help protect all your chats.
ChatGPT
By default, the most-used chatbot keeps a copy of each of your conversations and builds its own list of memories about you. ChatGPT’s maker OpenAI also gives itself permission to use what you type to train future versions of ChatGPT. Ads based on your chats likely aren’t far-off, though the company says it would “take a thoughtful approach” to them. (The Washington Post has a content partnership with OpenAI.)
What you should do now:
- Delete past chats. Go to Settings → Data controls and tap “Delete all.” Unfortunately, you can only delete them after the fact, and if you remember. ChatGPT offers no way to automatically delete chats on a schedule.
- Turn off memory. Go to Settings → Personalization and toggle off “Reference saved memories.” A more privacy-preserving way to personalize ChatGPT is to go to the same menu and type in some “Custom instructions” for how you want the bot to respond. Mine reads: “ask for clarification before answering if a prompt is vague, and cite sources when you do.”
- Stop ChatGPT from training on you, because you get nothing out of it and it won’t impact the quality of your responses. Go to Settings → Data controls and toggle “Improve the model” to off.
Claude
Anthropic’s bot, popular with coders and power users, keeps a copy of all your conversations, though it asks permission before using your data to train its AI.
What you should do now:
- Delete past chats. Scroll to the bottom of your list of recent chats and tap All chats. There you can bulk select bunches at a time and delete them. Claude offers no way to automatically delete chats on a schedule, or instruct it not to keep any history at all.
- Stop Claude from training on you. Go to Settings → Privacy and toggle “Help improve Claude” to off.
- If you have a paid account, turn off memory. Go to Settings → Capabilities and toggle “Generate memory from chat history” to off.
Gemini
By default, Google keeps your Gemini chats for 18 months and trains its AI on them. But it says it isn’t using your conversations to target ads … at least so far.
Gemini is also now integrated directly into Gmail and Docs, which raises different kinds of concerns.
What you should do now:
- Make past chats delete automatically. This is something Gemini does better than the rest. Tap on Settings & help → Activity. The best setting is to toggle “Keep activity” to off, which stops Google from keeping a file on you, and also stops it from using your conversations to train its AI. If you’d like to keep some past chats for reference, set them auto-delete after a fixed interval, such as 3 months. But just know: There is no way to stop Google from training its AI on your text if you keep it in your history.
- Turn off its memory. Go Settings & help → Personal context and toggle off “Your past chats with Gemini.” A better way to make Gemini more personalized is to go to Settings & help → Instructions for Gemini and tell the bot how you want it to tailor its answers.
- Decide whether to turn off Gemini built into Gmail and Docs. Google says it does not use the contents of your emails or documents to train Gemini. And when you interact with Gemini in those applications, Google doesn’t keep data about it, so there’s no setting to adjust. But if you don’t like the idea of Gemini interacting with your personal data, you can deactivate it. Go to Gmail → Settings → General and scroll down to Google Workspace smart features, tap Manage Workspace smart feature settings and switch it off. This covers Gmail, Docs, Calendar and Drive.
Meta AI
Meta’s homegrown chatbot is hard to avoid: it exists both as its own app and website and is also built into Instagram, Facebook and WhatsApp. And Meta gives you few ways to limit what data it keeps, or what the tech giant does with your data. For one: There is no setting to tell it not to use the contents of your chats to target ads.
What you should do now:
- Delete past chats. In the Meta.ai website or app, go to Settings → Data & privacy → Manage your information → Delete all chats and media. Meta offers no way to automatically delete chats in a regular fashion.
- Delete your public posts. Some people have inadvertently been sharing their Meta AI chats and AI-generated pictures and video with the public. To make sure you haven’t done that, go to Settings → Data & privacy → Manage your information → Remove all public posts.
- Review and delete your memory. You can’t stop Meta AI from building memories of you. But you can delete individual memories. Go to Settings → Memory and select the memories you want it to delete.
Microsoft
This gets a little confusing: There are different Copilots for personal use versus company employees. And they offer different privacy options.
Also, it’s worth noting that the personal version of Copilot targets ads based on the contents of your chat — unless you turn it off.
What you should do now:
For personal users, including those with both free copilot accounts and paid Microsoft 365 subscription:
- Delete your chat history. Go to your Microsoft account → Privacy → Copilot → Copilot apps → Delete all activity history. There is no way to auto-delete after a certain period of time. While you’re at it, also delete your use of Copilot in other Microsoft apps in that same menu.
- Tell Microsoft to stop targeting ads based on your chats. Go to your Microsoft account → Privacy → Personalized ads & offers and switch it to off.
- Turn off personalization and memory. Go to the Copilot website, then tap your name in the bottom left corner, then in the menu that pops up tap on your name again. Then go to Privacy, and scroll down to Personalization and memory and switch it off. While you’re there, also tap Delete memory to clear what’s already stored.
- Turn off model training on text and voice. Go to the Copilot website, then tap your name in the bottom left corner, then in the menu that pops up, tap on your name again. Then go to Privacy and switch model training to off.
For corporate users, there are higher privacy standards by default about training on your data and using it for ads. But you shouldn’t type anything into an enterprise Copilot you wouldn’t want your boss to see — all the data belongs to your company.
Here’s what you should adjust:
- Delete your chat history. Go to your Microsoft account → Settings & Privacy → Copilot activity history and tap Delete history. That removes it from your view, but your IT administrator might be able to still access a copy.
- Turn off personalization. Go to the M365 Copilot web interface and look for the three dots in the upper right corner and tap Settings → Personalization. Switch off Saved memories and chat history. A better way to make Copilot more personalized is to go on the same menu to Custom instructions, and tell the bot how you want it to answer you.
Heather Kelly contributed reporting.
The post Your chatbot keeps a file on you. Here’s how to delete it. appeared first on Washington Post.
