On July 31, 2025, "private" ChatGPT conversations went very much public on Google search results pages (SERPs).
Any chat that had been previously shared between two people via a link—similar to how you can share a Google Doc link with someone—wound up public, on the internet, and indexed in Google search. In a matter of about five minutes, I came across:
- Some pitch deck outlines
- Someone's movie script idea
- Personality analysis, including full names of users
- What someone was dreaming about last night
- Content about mood and emotions
- An analysis of some strangely old financial reports (if they were running a test using outdated information, well done, internet stranger!)
OpenAI and Google have since removed these chats from SERPs and if OpenAI did the right thing, they've permanently blocked these shared chats from appearing in search indexing. (The company has also removed the specific link sharing option that made the chat publicly searchable — while there was a very small warning this could happen when sharing chats via link, I expect most people weren't aware.)
But that doesn't mean your chats are private. Almost NONE of your large language model (LLM) use is truly private … even when you have "improve the model for everyone" turned off or an AI tool says "we don't use the contents of this chat to train our model."
There are two things to know about your chats with an AI:
- Model training: If your sensitive data is used to train or "improve" the functionality of an artificial intelligence model, it means that the computer algorithm that powers the AI is processing your chat and/or personal data. You could also discover that an AI tool like ChatGPT uses elements of your work in future outputs. With your content in the generative AI model’s training data, the AI could take on bits of your tone, style, and even the words that you've written — one of the reasons why creative professionals are very concerned about AI and copyright.
- Data privacy: Your Meta / Claude / ChatGPT chats may not be encrypted. Encryption scrambles the contents of a file in a way that is unreadable unless someone has the 'key' to unlock it. (A lot of this happens between approved devices — iMessages are encrypted between iPhones, for example, and you don't even realize it's happening.) ChatGPT conversations live on the company’s servers, and you have to take their word for it that customers on certain pricing tiers get encryption. And even if they do encrypt some information, that all goes out the window when they, you know, accidentally push your private chats out to Google search results for ✨everyone✨ to read! Why wait for a hacker to enact a data breach when it can just get shoved out there by the service provider you pay, eh?
How to tweak your ChatGPT settings to make them a little safer
If you want to be just a little safer
- Open up your account settings
- Click on "data controls"
- Select "improve the model for everyone" and turn it off
- Click "manage" next to "shared links" and delete any shared chats
- Select the "security" section of your settings
- Turn on multi-factor authentication, which makes it harder (though not impossible) for someone else to sign into your account without you knowing
- If you haven’t made an account yet, you can also consider using a fake phone number / name / email address during setup (though, honestly, you can still be identified through your IP address … but that’s another conversation)
If you want to balance convenience and safety
- Do everything in the previous section
- Open up your account settings
- Click on "personalization" and do the following to reduce what ChatGPT stores about you:
- Toggle "reference saved memories" off
- Toggle "reference chat history" off
- Click on "manage memories" and clear your memories
If you want to prioritize safety over convenience
- Do everything in the previous two sections
- Open your account settings
- Click on "connected apps" and disconnect from every option, if connected. This reduces the amount of extra information ChatGPT might have about you and your work or personal life.
- Pay for a ChatGPT account. While it's not foolproof, generally speaking, paying for software makes the company a little less likely to use your data for other monetization purposes. No corporation is giving you a free service of the goodness of their hearts.
If you say "screw it, I want to be as safe as possible"
...Stop using ChatGPT and other generative AI tools. Or at the very least, stop using it for anything related to your personal life and business intellectual property. In some ways, these chats are more like a Reddit thread than a private Whatsapp message.
I'm sorry. I know that sounds like a flippant answer, but it's not. Neither the free nor paid versions of ChatGPT are truly private. It's up to everyone to create their own threat model , but if you really want to be sure that your data won't end up in the wrong hands, the best way is to stop chatting with the AI.