The last time you pasted something private into ChatGPT, a contract, a medical worry, a draft you would never want public, you were also, by default, handing it over as training material. As of June 2026, ChatGPT, Google Gemini, and the consumer version of Claude all train on your chats unless you actively turn it off, and paying for Plus or Pro does not change that. This is not a conspiracy. It is a single setting, sitting in a menu you have probably never opened. Every one of them has an off switch. Here is exactly what each assistant does with what you type, and the precise steps to opt out.
Quick takeaways, as of June 2026: ChatGPT, Gemini, and Claude all train on consumer chats by default, and a paid plan does not opt you out. Per Google's Gemini Apps documentation, a sample of chats goes to human reviewers, and Google states those reviewed copies can be kept up to three years and are not removed when you delete your activity. Per Anthropic's August 2025 consumer terms, Claude now trains by default unless you decline, with retention up to five years if you allow it. Each has a one-toggle opt-out, and enterprise and API tiers are excluded by default.
What training on your chats actually means
Start with the thing people get wrong. Training on your chats does not mean the model memorizes your message and recites it to a stranger. It means your conversations can be stored, sampled, sometimes read by a human reviewer, and used as raw material to improve future versions of the model. The risk is not that the AI quotes your tax return verbatim. The risk is that a copy of what you told it sits on a company server, inside a training pipeline you do not control, for months or years. Once you see it that way, the question stops being academic. It becomes: who has a copy, and for how long?
The default is on, and paying does not turn it off
Here is the part that surprises even careful users. Upgrading to a paid consumer plan does not opt you out of training. ChatGPT Plus and Pro, Claude Pro and Max, Gemini on a paid tier, and Perplexity Pro all train on your conversations by default, exactly like their free tiers. You are paying for capability, not privacy. The upgrade buys a smarter model, not a quieter one. The only consumer lever is the opt-out toggle, and you have to flip it yourself. The table below is the short version. Each assistant is covered in detail underneath.
| AI assistant | Trains on your chats by default? | How to opt out |
|---|---|---|
| ChatGPT (Free, Plus, Pro) | Yes, unless you turn it off | Settings, Data Controls, turn off Improve the model for everyone |
| Google Gemini | Yes, and a sample goes to human reviewers | Turn off Gemini Apps Activity |
| Claude (Free, Pro, Max) | Yes since 2025, unless you decline | Privacy settings, turn off the model training setting |
| Microsoft Copilot | Opt-out available | Account, Privacy, turn off Training on conversation activity |
| xAI Grok | Yes, including your public posts | X Privacy settings, Grok and third-party collaborators, uncheck |
| Perplexity | Yes when signed in, unless you turn it off | Settings, Account, turn off AI Data Retention |
Setting names and menu paths change often. The labels below are accurate as of June 2026, but confirm the live toggle in your own settings before you rely on it. If a path has moved, search the assistant's help center for the words model training or data controls.
ChatGPT: one toggle, plus a private mode
On a consumer ChatGPT account, your conversations help train OpenAI's models by default. To stop it, open Settings, go to Data Controls, and turn off Improve the model for everyone. That single switch keeps your future chats out of training. There is also a faster path for one-off sensitive conversations: Temporary Chat. Anything in a Temporary Chat is not used for training and is deleted within about thirty days. Deleting a normal conversation also removes it from your account, with OpenAI stating deleted chats are purged within roughly thirty days.
Gemini: the three-year reviewer copy
Google Gemini trains on your conversations by default, and a subset of chats is sent to human reviewers to improve Google's models. To opt out, turn off Gemini Apps Activity in your Google account. With activity on, Google auto-deletes after eighteen months by default, which you can shorten. With it off, your chats are kept only around seventy-two hours. Here is the detail worth knowing, because it is the strongest reason to think before you type. Per Google's Gemini Apps documentation, conversations selected for human review can be retained for up to three years, and Google states they are not removed when you delete your activity. Turning the setting off stops new sampling. It does not pull back the up-to-three-year copy a human reviewer already holds.
Claude: the assistant that quietly changed sides
Claude is the surprise on this list. For most of its life, Anthropic did not train on consumer conversations by default, which earned Claude a reputation as the privacy-conscious choice. Per Anthropic's own announcement, that changed on August 28, 2025. Under the updated consumer terms, Anthropic now uses your chats to train its models unless you decline, and existing users had to make a choice by October 8, 2025. The retention difference is large and worth stating plainly. If you allow training, Anthropic can keep your data for up to five years. If you decline, it reverts to the previous thirty-day window. New users pick at signup, and you can change the setting later in your privacy controls. The reputation has not fully caught up with the policy, so a lot of people who chose Claude for privacy are now training it without realizing the default flipped.
Grok, Copilot, and Perplexity, in brief
- xAI Grok trains on your interactions and, through X, on your public posts and profile by default. On X, go to Privacy and safety, open Grok and third-party collaborators, and uncheck the data-sharing option. Note that, per X's terms of service, the content license you grant X is separate from this toggle, so unchecking it does not revoke that license.
- Microsoft Copilot offers a training opt-out. In the app or on the web, open your Account, then Privacy, and turn off Training on conversation activity. Microsoft states it does not train on signed-out sessions, and consumer data in the EEA is excluded.
- Perplexity trains on conversations when you are signed in, unless you turn off the AI Data Retention toggle under Settings, Account. Its enterprise tier and Sonar API are never used for training.
Enterprise and API are a different deal
If you use AI through a business, the math flips. OpenAI does not train on Enterprise, Business, Education, or API traffic by default, and has not trained on API data since early 2023 unless a customer opts in. Anthropic excludes its commercial and API products from consumer training. Perplexity never trains on enterprise content. The pattern is consistent across the industry: the paid business tier you reach through an API or an admin console is treated as private by default, while the consumer app you use at home is treated as training fuel until you say otherwise.
Opting out is the floor, not the ceiling
Turn off training on all six and you have done the right thing, but be honest about what you have achieved. You have stopped your words from feeding the next model. You have not made your data yours. The transcripts still live on each company's servers, under each company's retention clock, readable under the right legal request, and tied to whichever assistant you happened to use that month. Opting out controls one use of your data. It does not give you ownership of it.
This is the gap a memory you own is meant to close, and it is the reason MemX exists. Instead of scattering your documents, decisions, and notes across six assistants that each train on them by default, you keep them in one place that belongs to you, and let whichever model is best this month read from it on demand. The model becomes the replaceable part. The memory stays yours, opted out of everyone's training pipeline by design, because it never lived inside their product in the first place. You should still flip every toggle in this post. Just do not mistake an opt-out for control.
Do the five-minute version today: turn off Improve the model for everyone in ChatGPT, turn off Gemini Apps Activity, and check your model training setting in Claude. Then ask the bigger question of where your knowledge should actually live. Start with a memory you own at memx.app.
Key takeaway: the major AI assistants train on your conversations by default, and paying does not stop it. Each has a one-toggle opt-out, but opting out only stops the training. Your data still sits on their servers. Owning your memory is the only real fix.
01Does ChatGPT train on my conversations?
Yes, by default on consumer accounts. To stop it, open Settings, go to Data Controls, and turn off Improve the model for everyone. For one-off sensitive chats, use Temporary Chat, which is not used for training and is deleted within about thirty days. This is accurate as of June 2026, per OpenAI's data controls documentation.
02Does paying for ChatGPT Plus stop AI training?
No. ChatGPT Plus and Pro train on your conversations by default, just like the free tier. Paying upgrades capability, not privacy. The only way to stop training on a consumer plan is to turn off the data-sharing toggle in Settings, Data Controls yourself.
03Does Claude train on my chats?
Yes, since a policy change Anthropic announced on August 28, 2025. Per those updated consumer terms, Anthropic now uses consumer chats to train its models unless you decline, with retention up to five years if you allow it and thirty days if you decline. You can change this in your Claude privacy settings. This is accurate as of June 2026.
04How do I stop AI from training on my data?
Open each assistant's settings and turn off the training toggle: ChatGPT (Data Controls, Improve the model for everyone), Gemini (turn off Gemini Apps Activity), Claude (model training setting), Copilot (Privacy, Training on conversation activity), Grok (Grok and third-party collaborators on X), and Perplexity (AI Data Retention).
05Is my data deleted when I delete an AI chat?
Usually within a retention window, but not always completely. ChatGPT purges deleted chats within about thirty days. Gemini is the exception to watch: conversations already sampled for human review are kept up to three years and are not removed when you delete your activity. Always check each service's stated retention.
