How to stop AI tools from training on your data
- You can opt out of training in most AI tools' privacy settings, but the only guaranteed way is to use AI that never uploads your data at all — on-device AI.
- In ChatGPT, Claude, and Gemini you can usually turn off the "improve the model for everyone" toggle and use temporary/incognito chats. Check each tool's current setting — they move and reset.
- Business and enterprise tiers often exclude training by contract; consumer free tiers usually default it on.
- A toggle is a request to a company. The only thing a company can't train on is data that never reaches it.
You can opt out of training in most AI tools' privacy settings, but the only guaranteed way to keep your data out of a model is to use AI that never uploads it — on-device AI. Here's the honest version of both: the settings to flip today, why they aren't a guarantee, and the one approach that is. The toggles work and you should use them. Just know what they are and aren't.
The fast version: turn off training in the settings
Every major consumer AI tool has a setting that controls whether your chats are used to improve its models, and on the free tiers it usually defaults to on. The opt-out is real and it's worth doing. The wording moves around, so don't trust a screenshot from a blog — open the tool and look. The general steps:
- Open the tool's settings, then find the section called privacy, data controls, or something similar.
- Find the toggle that says it uses your conversations to "improve the model for everyone" (or "train our models," or similar) and turn it off.
- Use a temporary or incognito chat when you want a single conversation kept out of training entirely — most tools now offer this.
- Check business and enterprise terms. Paid team, business, and enterprise tiers frequently exclude training from your data by default — if you're on one, you may already be covered, but read the data-use terms to confirm.
- Re-check the setting later. These options get renamed, moved, and occasionally reset on a new sign-in or a new feature rollout. Treat it as a recurring chore, not a one-time fix.
Two honest caveats. First, opting out of training is not the same as opting out of storage — your conversations may still be saved and reviewed for abuse or safety even when they're excluded from training. Second, "off" depends on the company keeping its word, its settings stable, and its terms unchanged. Useful, not airtight.
Where the opt-out usually lives (general, honest)
This is a map, not a manual. Exact menu names change often, so use this to know roughly where to look — then confirm the current wording inside the tool itself.
| Tool | Roughly where the training opt-out lives |
|---|---|
| ChatGPT | Settings → Data controls (a toggle for improving the model); temporary chats for one-off privacy |
| Claude | Settings → Privacy / data-use controls; check your plan's data terms |
| Gemini | Settings → your activity / app-activity controls in your Google account |
| Most others | Settings → Privacy or Data; look for a "use my data to improve the model" toggle |
If you can't find a toggle, that itself is the answer: a tool with no opt-out is using your data and you can't stop it without leaving. Which is the cleaner question to ask anyway — not "how do I ask them not to," but "why is my data there at all?"
The only guaranteed way: keep the data on your machine
Every opt-out above shares one weakness: your prompts still leave your computer. They travel to a server you don't control, governed by a policy that can change, enforced by a company you have to trust. Turning off training reduces what they do with the data. It doesn't change the fact that they have it.
On-device AI removes the weakness instead of managing it. The model file sits on your disk and the math runs on your own chip, so the conversation never reaches a server. There is nothing to train on because nothing was uploaded. There is no opt-out to maintain because there was never an opt-in. Pull the Wi-Fi and it still answers — proof the data isn't going anywhere. Files, as we like to say, don't have terms of service.
That's the structural point: an opt-out is a promise; on-device is a fact you can verify by switching off your network. It's also why a developer who can't paste proprietary code into a cloud tool can paste it into a local one without a second thought — there's no transmission to leak.
So which should you do?
Both, in order. Flip the training toggles in the tools you already use today — five minutes, real benefit. Then, for anything you'd actually mind a company keeping (client work, health notes, code, anything you'd hesitate to email a stranger), move it to AI that runs on your own machine, where the opt-out question never comes up. The first is good hygiene. The second is the only version that doesn't depend on anyone's promise.
Frequently asked questions
Does ChatGPT train on my conversations?
On consumer tiers it can, unless you turn training off in the privacy settings. Most major AI tools default to using your chats to improve their models and offer an opt-out toggle; business and enterprise tiers usually exclude training contractually. Settings vary by tool and change over time, so check the current option in each tool. The only way to be certain your conversations are never trained on is to use AI that runs on your own device and uploads nothing.
How do I opt out of AI training?
Open the tool's settings, find the privacy or data-controls section, and turn off the toggle that lets it use your chats to improve the model. In some tools, using a temporary or incognito chat keeps that conversation out of training. Then verify the setting later, because these options move and reset. Business or enterprise plans often exclude training by default. None of these guarantee zero upload; only on-device AI does.
Is there AI that never uses my data?
Yes — AI that runs entirely on your own machine. If the model and inference live on your device and nothing is sent to a server, there is no conversation to train on and no opt-out to manage. Outlier is a Mac-native app whose free Nano and Lite tiers run locally and work with Wi-Fi off, so your prompts never leave the device. Opt-out toggles are a request to a company; on-device is a guarantee by physics.
Is opting out of training the same as deleting my data?
No. Opting out of training stops your chats from being used to improve future models, but it doesn't necessarily delete them — conversations may still be stored and reviewed for safety or abuse. Deletion is a separate control, and where your conversations go in the meantime is worth understanding before you rely on any single toggle.
Try Outlier free
Free Nano + Lite — local, private, no account. Pro $20/mo or $149/yr adds everything (all 7 model tiers incl. Plus 397B). Lifetime Pro from $99 (Founding 200, first 200 seats) or $200 (Founders 500). Apple Silicon only.
Download for Mac