ChatGPT’s data retention and model training
Some weeks ago now, Sam Altman, CEO of OpenAI, the maker of ChatGPT in all its variations, told a podcast host that he wished people would stop using ChatGPT as a therapist because the conversations weren’t actually as private as people thought they were.
And then ChatGPT 5 was released.
Guess which one of those two things made headlines longer?
Here’s what I learned about data privacy on ChatGPT: that “data is deleted” clause for free accounts actually doesn’t apply any more because of a US federal ruling in the New York Times vs OpenAI case that called for keeping the data around. That happened in May 2025.
The only accounts where data privacy actually still holds? ChatGPT Teams, Education and Enterprise.
Any other type of account, your data has not been deleted since May 2025, and won’t be unless that ruling changes.
But that’s just part of a puzzle about data privacy on GenAI. The other part has to do with whether or not your data is being used to train ChatGPT.
If you’re on a ChatGPT Team, Education, or Enterprise account, you’re defaulted into your data not being used to train the model.
Otherwise? You need to opt out.
What does this mean for small businesses?
It means that:
Data you enter into ChatGPT is being held indefinitely.
Unless you specifically opted out of it (Settings > Data Controls > Improve the Model for Everyone = Off), your data is being used to train the model.
In short, just like your old Facebook posts that you wish you could erase, anything you put into ChatGPT stays there. So be mindful what you’re entering. If it’s proprietary to your business, leave it for something else.
Other data retention and training info
A quick look at other GenAI tools’ data retention and model training policies:
Tool | Data retention and privacy | Model training |
|---|---|---|
Perplexity | Files are deleted after 7 days; conversations are not. They may share your data with advertisers and business partners. Not subject to the same federal ruling as OpenAI. | Need to opt out for your data being used to train Perplexity; data is not used to train any external models |
Claude | Default storage is 2 years. You control when conversations are deleted. Once you delete a conversation, it is then deleted from storage in 30 days. Data is not shared with any third parties. Some enterprise accounts have “zero data retention” agreements. | Need to opt-in to have your data used to train their model. Commercial users have explicit ownership of outputs from the AI. |
Gemini | Free/Pro stores Gemini Apps activity for up to 18 months by default, but users can change that to 3-36 months. Anonymized versions of your conversations are reviewed by humans to help with quality and to improve products. Apps chats are not being used to show you ads. | Conversations are used for model training, but the training is done via anonymizing the conversations and then having humans review them, not automatically uploading them. Workspace and Enterprise accounts default to just not having conversations used to train models. |
Manus.im | No clear data retention information. Your data may be shared with advertisers, though. | Data is not used to train their or any third-party models, whether LLM or machine learning. |
One customer called it, “murky waters.”
Market research doesn’t have to be confusing. If you’ve been stuck on a project and need some direction, or have a project you need to kick off but don’t know where to start, join my market research training and coaching pilot program!
Limited spots are open through September for organizations wanting 1:1 training and coaching through their own market research project. For $3600, you’ll get 12 hours of expert guidance and access to world-class Market Research 101 training from Research Rockstar! To qualify, you just need a market research project to run. Email [email protected] to learn more and schedule a free introductory session.
Mental Health Moment
Have you ever wondered why there are songs about life being a “road” or a “highway” when it seems like it’s more like one of those video games where you start in the middle of nowhere, with absolutely no directions on what to do in the game, what the objective of the game is, how to move or do anything, and you have to figure everything out?
Well, it turns out that, just like with those games, there is a cheat code. It’s just a tough one for most of us: asking for help.
In Western society, independence is lauded and praised. Asking for help isn’t. And honestly, that’s kind of backwards, because independently trying to make your way through things just makes it all harder. Getting help makes things easier. Easier to bear because someone else is sharing the pain, even if it’s just by listening. Easier to handle because someone else is helping you figure it out, even if they’re just looking things up like you are. Easier to understand because someone else has been through it and can help you navigate it.
Lauren Howard is one of my favorite people to follow on LinkedIn, because she constantly reminds people that isolation is a lie. You’re not the only one feeling the way you do. and You don’t have to do it all alone.
Confession: I am awful at asking for help. I was raised to be “fiercely independent.” And while there are times I’ve been proud of figuring something out on my own, I inevitably learn later that I could have done it 20x easier and faster by asking someone else for help. You’d think after a few times, I’d figure out it was easier to look for help before trying to figure something out on my own, but, no, my default setting seems to be a toddler’s stubbornness level of “I do it myself!”
So let’s change the script a bit and remember isolation is a lie, independence is better with friends, and ask for help through life’s twists and turns, ups, downs, and sideways.
And always remember to breathe.