AI · Privacy · Data protection · Cybersecurity
The real price of "free" AI
Chatting with a consumer AI assistant can mean your conversations help train its models, whether you pay or not. Here is what ChatGPT, Claude and Gemini actually do with your chats, the fine print most people miss, and five steps to take back control.
There is a saying in tech: if a product is free, you are the product. With AI chatbots, the saying is close to literal. The questions you ask, the drafts you paste in and the problems you talk through can become raw material for the next version of the model.
A popular claim goes one step further: using a free AI model means you have agreed to your data being used for training. That claim is often true, but it is incomplete in ways that matter. The real dividing line is not free versus paid. It is personal versus business accounts, and the default setting versus the choice you make.
Consent by default
Most people never read privacy terms. They accept a service and inherit its settings, usually with one tap on a large Accept button. That makes the default setting the real policy for most users.
On ChatGPT and Gemini, using your conversations to improve the models is switched on until you switch it off. This is called an opt-out model: the company does not ask you to say yes, it assumes yes and lets you say no, in a settings menu few people open.
Claude works differently. Since Anthropic updated its consumer terms in 2025, it asks each user to choose whether their chats may be used for training. Existing users had to decide by 8 October 2025. The choice is yours, but it is still one tap on a pop-up, often made in a hurry.
What the big three actually do
| Service | Used to train models? | How to switch it off | Worth knowing |
|---|---|---|---|
| ChatGPT (OpenAI) | Yes by default on personal Free, Plus and Pro accounts, and when you use it without logging in | Settings > Data Controls > “Improve the model for everyone”. Logged out: the ? icon > Settings | Temporary Chats are not used for training, but may be kept for up to 30 days for safety |
| Claude (Anthropic) | Only if you allow it, on Free, Pro and Max, including Claude Code on those accounts | Settings > Privacy > “Help improve Claude” | If you allow it, chats are kept for up to 5 years; if not, about 30 days |
| Gemini (Google) | Yes while Keep Activity is on, which is the default for personal accounts of users 18 and over | Turn off Keep Activity (formerly Gemini Apps Activity) | Human reviewers, including service providers, read some chats |
Business and developer accounts are different. OpenAI does not train on data from ChatGPT Business, Enterprise and Edu, or from its API, by default. Anthropic’s update does not apply to Claude for Work, Claude for Government, Claude for Education or its API, which fall under commercial terms that do not train on customer content.
The fine print most people miss
Switching training off helps, but it is not a clean slate.
- It only works going forward. Changing the setting stops your future conversations being used. It does not remove what a model has already learned from your earlier chats.
- Deleting is not always deleting. On Gemini, chats picked for human review are disconnected from your account and kept for up to three years, even after you delete your activity. With Keep Activity off, Google still keeps chats for 72 hours.
- Feedback can be an exception. On Gemini, chats you send feedback on can be used even with Keep Activity off. Anthropic keeps feedback you submit about Claude’s answers for five years if you have allowed training.
- Old chats can come back into scope. On Claude, old chats with no new activity are not used. Continue one, and it falls under your current setting.
- Temporary does not mean instantly erased. Temporary and incognito modes are excluded from training, but ChatGPT may keep them for up to 30 days and Gemini for 72 hours.
Myth versus fact
| Common belief | What is actually true |
|---|---|
| Only free users have their data used | Paid personal plans, such as ChatGPT Plus and Pro, follow the same rules as the free plan |
| Paying guarantees privacy | Protection by default comes with business, enterprise and API plans, not with a higher personal tier |
| You have no choice | ChatGPT, Claude and Gemini all let you switch training off |
| Switching off erases your history from the model | It only stops future use; past training cannot be undone |
| Deleting a chat removes every copy | Reviewed, flagged or feedback chats can be kept separately |
How to protect yourself
- Check your settings today. Find the training setting in every AI app you use and decide deliberately, rather than living with the default.
- Use temporary chats for sensitive questions. ChatGPT’s Temporary Chat and Gemini’s temporary chats are not used to train models. Remember they are still kept for a short time.
- Keep secrets out. Treat any chatbot like a public space: no passwords, ID numbers, medical records or confidential work documents.
- Use business plans for work. If your organisation handles client or company data, use a business, enterprise or API plan whose terms exclude training.
- Recheck after policy updates. Terms change. When an app shows an “updated terms” pop-up, read the settings before tapping Accept.
The bottom line
Using a personal AI chatbot often means agreeing, by default or with a quick tap, that your conversations can help train its models. That is true whether you pay or not. The consent is real in a legal sense, but it is rarely an informed choice.
The good news is that the choice still exists. A minute in your settings, and some care about what you type, puts you back in control of your own words.
Sources
- Data Controls FAQ, OpenAI
- Temporary Chat FAQ, OpenAI
- Enterprise privacy, OpenAI
- Updates to our consumer terms, Anthropic
- Gemini Apps Privacy Hub, Google
Policies and settings were checked against these pages on 10 October 2026. AI companies change their terms often, so check the current settings in each app.
