When you type into an AI chatbot such as Gemini, ChatGPT or Claude, your message leaves your device, is processed on the company’s servers and is usually saved to your account, where it may be reviewed by people and, depending on your settings, used to improve future models. How long it is kept and who can see it depend on a handful of settings most people never open. With chatbots now connecting to email, photos and, since this month, bank and wallet data, those settings matter more than ever. This guide explains the basics in plain language and shows which switches to check first.
Where your words go when you press send
A chatbot is not a program running only on your phone or laptop. The large language model that writes the answer runs in the company’s data centers, so every message, photo or file you share is sent there. The service then does three separate things with it:
- Answers you. The model reads your message, plus recent parts of the conversation, and generates a reply. Our explainer on context windows covers how much of a chat the model can “see” at once.
- Stores it. Most services save your chats so you can return to them. That history sits with your account until you or an automatic timer deletes it.
- May reuse it. Companies can use saved conversations to check for abuse, fix problems and, in many cases, train or improve their models, either by default or only if you agree.
Keeping these three uses apart is the key to understanding any chatbot’s privacy settings. Turning off one, such as model training, does not necessarily switch off the others.
Can a human read my chats?
Sometimes, yes. Google’s Gemini privacy hub says a subset of chats is reviewed by trained human reviewers, including staff at service providers, who rate whether answers were accurate or harmful and suggest better ones. Google says chats are disconnected from your account before they reach those providers, and that reviewed chats are kept for up to three years, even if you later delete your activity. Its advice is direct: do not enter confidential information you would not want a reviewer to see.
Anthropic’s privacy center says conversations flagged for safety review may be analyzed to enforce its usage policy, regardless of your training setting. Most large chatbot providers describe some form of safety review, so the safest assumption for a beginner is that nothing you type is guaranteed to be seen only by a machine.
Is my data used to train the AI?
This is the setting people ask about most, and the answer differs by company and by plan.
- Gemini: a setting called Keep Activity controls whether your future chats are saved and can be used to improve Google’s AI. If it is off and you do not submit feedback, Google says it does not use your future chats to improve its models.
- Claude: for consumer plans (Free, Pro and Max), Anthropic uses your chats to improve its models if you allow it in your privacy settings, if a conversation is flagged for safety review, or if you join a testing program. Its Incognito chats are not used for training even when the setting is on.
- Business plans: company and team plans from the major providers generally come with stricter defaults, but read your employer’s agreement rather than assuming.
One detail catches many people out. Pressing thumbs up or thumbs down sends feedback, and Anthropic says it stores the whole related conversation for up to five years when you do, separated from your user ID. Google also treats feedback as data it can use. If a conversation was sensitive, skip the rating buttons.
Temporary and incognito chats
Most chatbots now offer a mode that does not keep a lasting history. Gemini’s temporary chats are kept with your account for 72 hours, Google says, so the service can reply using the last 24 hours of conversation and protect against abuse or system failures. They are not used to train Google’s AI models. Claude’s Incognito chats, as noted above, are excluded from model improvement.
These modes are useful for one-off sensitive questions, but they are not invisible. The company still processes the conversation, and safety systems still apply.
Connected apps: the new privacy question
Chatbots increasingly read data from other services. Gemini’s Personal Intelligence can connect Gmail, Photos, YouTube and Search, and on November 13 Google added Google Wallet, so Gemini can answer questions about your spending. Claude and ChatGPT have launched similar finance and app connections.
Three rules of thumb help here:
- Connect only what you need. Each connection gives the chatbot a wider view of your life. Turn on the finance link when you want a budget review, not by default.
- Check what the other service keeps. Google says that when Gemini works with services such as Wallet or YouTube, those services may save activity under their own policies, and deleting your Gemini activity does not delete it there.
- Prefer read-only access. Google says Gemini can only read Wallet data and cannot move or invest money. Know whether an assistant can act, such as making payments, before you connect it.
A five-minute privacy checkup
If you use a chatbot regularly, these steps take a few minutes:
- Open the chatbot’s privacy or data settings and find the training or “improve the model” switch. Decide whether you want it on.
- Find the history setting. Gemini lets you change its auto-delete period from the default of 18 months to 3 months, 36 months or indefinitely, and delete chats manually.
- Look for a memory page. Gemini’s Memory Summary and similar features in other chatbots show what the assistant remembers about you; delete anything you are not comfortable with.
- Review connected apps and switch off the ones you do not use.
- Use a temporary or incognito chat for anything health-related, financial or about other people.
If privacy is your top concern, a model running on your own computer keeps prompts on your machine. Our comparison of cloud APIs and local models explains the trade-offs in speed and quality.
What to watch next
- More chatbots adding bank, health and email connections, each with its own data rules.
- Regulators in the U.S. and Europe examining how long chatbot conversations may be retained.
- Whether providers make temporary chats or training opt-outs the default for sensitive topics.
Key takeaways
- Chatbot messages are processed and usually stored on the company’s servers, not just on your device.
- Answering, storing and training are separate uses with separate settings.
- Some chats are read by human reviewers; Google keeps reviewed Gemini chats for up to three years.
- Temporary or incognito chats limit history and training but are still processed by the company.
- Connected apps such as Google Wallet widen what a chatbot can see, so connect only what you need.
Frequently asked questions
Do AI chatbots save my conversations?
Usually, yes. Most chatbots store your chats with your account until you delete them or an auto-delete timer runs out; Gemini’s default is 18 months. Temporary or incognito modes keep them for a shorter time.
Can I stop a chatbot from training on my chats?
Often you can. Gemini stops using future chats to improve its models when Keep Activity is off and you do not send feedback, and Claude’s consumer plans use chats for training only if you allow it, apart from safety reviews.
Is it safe to share financial information with an AI chatbot?
It can be, if you use official, read-only connections and understand the provider’s retention rules, but avoid pasting account numbers or passwords into chats. Check totals the chatbot gives you against your bank’s own statements.




