What AI Tools Actually Do With Your Data
Every AI tool asks you to paste something in. We look at who trains on your text, who does not, what the settings actually change, and what you should never paste at all.
Eddie Ochieng
July 16, 2026
Every AI tool works the same way from your side. You paste something in, and something useful comes back. The question almost nobody asks in that moment is where the thing you pasted just went, who can see it, and whether it will end up shaping the model that answers somebody else tomorrow.
The honest answer is that it depends enormously on which tool you used and which tier you are on, and the difference is much bigger than most people realise.
How we compared
This piece is drawn from the published privacy policies, terms, and data controls of each vendor, and from their own documentation on training. Policies change, sometimes quietly, so treat the specifics as a snapshot and the principles as durable. Where a vendor is ambiguous, we say that rather than guess.
The rule that explains almost everything
There is one distinction that predicts most of the answer. Consumer tiers tend to train on your conversations by default. Business, enterprise and API tiers tend not to, because companies would not buy them otherwise.
That is the whole game, most of the time. If you are using a free consumer chatbot, assume your text may be used to improve the model unless you have turned that off. If your employer pays for a business tier, your text is usually walled off contractually. The gap between those two situations is enormous, and it is invisible from the chat box, which looks identical either way.
What the settings actually change
Most major assistants now offer a control that stops your conversations being used for training. It is worth finding and turning on, and it is worth understanding what it does not do.
Turning off training does not mean your conversation vanishes. Providers still typically retain conversations for a period, for abuse monitoring and legal reasons, and staff can in defined circumstances review them. So the setting changes whether your words shape the next model. It does not make the exchange private in the way a locked notebook is private.
Delete is not a magic word
Deleting a chat is not the same as deleting the data. Retention windows, backups, and abuse review systems all mean that hitting delete removes it from your view rather than from existence. Assume anything you paste has a life beyond your screen.
What you should simply never paste
You do not need to become paranoid to be sensible. There is a short list where the calculation is not close, because the downside is severe and the convenience gained is small.
Do not paste customer data, medical records, or anything covered by a confidentiality agreement. Do not paste credentials, API keys, or passwords, ever, into any chatbot. Do not paste proprietary source code you are not permitted to share, which is a real and common problem in engineering teams. And think hard before pasting personal information about other people, because they did not consent to be part of your prompt.
Where the tools differ
The consumer assistants from OpenAI, Google and Microsoft all offer training controls, and all default consumer tiers toward using your data unless you say otherwise. Anthropic has taken a comparatively conservative default position with Claude on consumer training, though the safest habit is still to check the current setting rather than trust a summary written months ago.
The pattern that matters is not which vendor is virtuous. It is that defaults favour the vendor, controls exist but are not loud, and the tier you are on changes the answer more than the logo does.
| Tool | Price | Best for |
|---|---|---|
| ChatGPT | Free tier, Plus about $20/mo | Check the training control in data settings |
| Gemini | Free tier, paid about $20/mo | Review activity settings in your Google account |
| Claude | Free tier, Pro about $20/mo | Conservative consumer defaults, still worth checking |
| Microsoft Copilot | Free tier, paid tiers via Microsoft 365 | Work accounts differ sharply from personal ones |
FAQ
Do AI companies train on my conversations?+
On consumer tiers, often yes by default. On business, enterprise and API tiers, usually not, because that is what companies pay for. Check the setting in the tool you actually use, since defaults differ and change.
Does turning off training make my chats private?+
No. It stops your text being used to improve the model. Providers still typically retain conversations for a period and may review them for abuse. Private in the sense of unseen is not what any of these settings promise.
Is it safe to paste work documents into ChatGPT?+
Only if your employer has a business agreement covering it, or the document is not confidential. Pasting customer data or proprietary material into a personal account is exactly the mistake that gets people disciplined.
What about code?+
Never paste credentials, API keys, or code you are not permitted to share. For everyday code that is not sensitive, it is usually fine, and a paid business tier is the right answer for teams who need certainty.
For assistants generally, see our comparison of ChatGPT vs Gemini vs Claude. For pasting code specifically, see how to use ChatGPT for programming help.
