Quick answer: Free and Plus may train on what you type unless you turn it off. Pro does too — it's priced like a business tool but runs on the same consumer toggle. Team, Enterprise, Edu and the API don't train on your data by default; you don't have to do anything to get that protection. The toggle for the consumer tiers lives in Settings → Data Controls, though the exact label has moved before and will again.
This is the question that decides whether a business lets its team near ChatGPT at all, and most answers you'll find online are either years out of date or blur consumer and business tiers together as if they behave the same way. They don't. What follows is the current tier-by-tier picture, plus how to check it yourself in two minutes rather than trusting anyone's screenshot — because vendor settings drift and a guide that isn't dated is a guide you can't rely on.
Does ChatGPT train on your data, tier by tier?
Training depends entirely on which plan you're on, not on how much you're paying per seat.
| Tier | Trains on your data by default? | Can you turn it off? | Retention | DPA available? |
|---|---|---|---|---|
| Free | Yes | Yes, in Data Controls | Standard window, longer if flagged for abuse review | No |
| Plus | Yes | Yes, same toggle as Free | Same as Free | No |
| Pro | Yes | Yes, same consumer toggle | Same as Free/Plus | No |
| Team | No — excluded by default | Not applicable, already off | Admin-configurable, workspace-level controls | Yes |
| Enterprise | No — excluded by default | Not applicable, already off | Admin-configurable, longer options available | Yes |
| API | No — excluded by default | Not applicable, already off | Typically short, around 30 days, for abuse monitoring | Yes |
The pattern: three consumer tiers share one training toggle that defaults to on. Three business tiers exclude your data as standard, and back it with a data-processing agreement. The dividing line isn't price. It's contract type.
Does ChatGPT Pro train on your data?
Yes, by default, and this is the one nobody writes about clearly. Pro costs several times what Plus does and gets marketed around its stronger models and higher usage limits, which makes it easy to assume it also comes with business-grade data handling. It doesn't. Pro is still a consumer subscription. It sits under the same personal account settings as Free and Plus, including the same training toggle with the same default.
If you've put your best people on Pro because it's the "serious" tier, check the data controls before you assume anything about what happens to what they paste. The fix is identical to Free and Plus: switch off training in your account settings, or move the work onto Team or Enterprise, where exclusion is the default rather than something you have to remember.
Where's the toggle, and how do you check it yourself?
Settings → Data Controls → the option usually labelled "Improve the model for everyone." Switch it off and new conversations stop feeding training. It does not retroactively remove anything already sent.
Don't take that path as gospel, though — OpenAI has renamed and relocated this setting more than once, and every article that screenshots the exact click path is stale within a few months. Two minutes gets you a real answer instead of a guess:
- Open your account settings inside ChatGPT.
- Find the data or privacy controls section — the label has varied, but it's always grouped near account and privacy settings.
- Look for the model-improvement or training toggle and read its current state and wording. If the text describes what it does rather than matching an old label you remember, that's fine — read the description, not the name.
- If you manage a workspace, check the admin console too. Team and Enterprise settings live separately from personal account settings, and an admin default doesn't always reflect what an individual sees on a personal login within the same organisation.
Do this once now and put a reminder to check again in six months. Settings pages are not stable ground to build a policy on.
What "doesn't train" doesn't mean
Turning training off, or being on a tier where it's already off, does not mean your data disappears or stays only with you. It's still sent to and processed on OpenAI's servers. It's still retained for a period — 30 days by default on business tiers for abuse monitoring, even when nothing is used for training. And it's still discoverable in principle: subject to legal process, internal safety review, or a security incident, the same as data held by any third-party processor.
Training and privacy are two separate questions and it's worth keeping them apart. "Not used to train the model" answers one. It says nothing about who else can see it, how long it sits on a server you don't control, or what happens if that server is compromised. Anyone who treats "training is off" as "this is now safe" has stopped a step too early.
The temporary chat escape hatch
Temporary chats are the closest thing ChatGPT has to a private browsing window. They don't appear in your history, they aren't used for training, and they get cleared after the session. For a one-off question involving something sensitive, that's the fastest safe option — faster than digging through settings, and it doesn't touch your account-wide toggle.
It's not a substitute for a proper policy, though. Temporary chats solve a single conversation. They don't solve what your team pastes into ChatGPT every day, which is a settings-and-training-policy problem, not a mode problem.
What about Microsoft Copilot, Google Gemini and Anthropic Claude?
The same shape shows up everywhere, as of August 2026. Microsoft Copilot splits along the same consumer/business line: personal Copilot use falls under Microsoft's standard consumer terms, while Copilot for Microsoft 365 keeps your data inside your tenant's compliance boundary and excludes it from training. Google Gemini follows the identical pattern — the free consumer app may have conversations reviewed to improve the product, while Gemini for Workspace and Google Cloud exclude your data under enterprise terms. Anthropic Claude does the same again: the consumer app has its own data-control settings for training, and Claude for Team, Enterprise and the API don't train on your inputs by default, backed by a commercial agreement.
Every one of these vendors updates its policies on its own schedule, and a description that's accurate today can be wrong in six months. Treat every claim on this page, and every claim anywhere else, as something to verify in your own account before you rely on it — the tier and the contract matter far more than which logo is on the login screen.
FAQ
Does ChatGPT train on my data?
On Free, Plus and Pro it can, unless you turn training off in the data controls. Team, Enterprise, Edu and the API don't train on your data by default. Check your own account, because the setting and its default have both changed before and will again.
Does ChatGPT Pro train on my data?
Yes, unless you switch it off. Pro is priced like a business tool but sits under the same consumer data-controls toggle as Free and Plus. Paying more for Pro does not change the training default — you still have to turn it off yourself.
How do I turn off training in ChatGPT?
Go to Settings, then Data Controls, then switch off the option usually labelled Improve the model for everyone. It only applies going forward, not to conversations already sent. The exact wording and location have moved before, so check your own account rather than trusting a screenshot from an old article.
If ChatGPT doesn't train on my data, is it fully private?
No. "Doesn't train" only means your conversations aren't used to improve the model. Your data is still sent to and processed on OpenAI's servers, still retained for a period (30 days by default for abuse monitoring on business tiers), and still visible to OpenAI staff or systems under certain conditions. Training and privacy are two separate questions.
What is a temporary chat in ChatGPT and does it stop training?
A temporary chat doesn't appear in your history and isn't used to train the model, similar to a private browsing window. It's the fastest way to ask something sensitive without touching your settings, but it doesn't retroactively affect anything you've already sent in a normal chat.