Everything you type into an AI chatbot gets read, stored, and in most cases used to train the next version of the tool. ChatGPT, Gemini, Microsoft Copilot, and DeepSeek are genuinely useful on a Mac. They draft your emails, write content, and sort your grocery list to a budget. They also keep a record of what you tell them, and that record can end up in places you'd never agree to if anyone asked.
So the real question is how much each one collects, and where that data goes once it leaves your Mac.
What each chatbot actually keeps
When you talk to a chatbot, your words aren't thrown away after the answer comes back. The tool reads your input to write a reply, and that input can include personal details, sensitive notes, or work you'd rather not hand to a vendor. What happens next depends on the platform.
- ChatGPT: OpenAI logs your prompts, device details such as Mac hardware identifiers, location data, and how you use the tool. That data can be shared with vendors and service providers to improve the service.
- Microsoft Copilot: Built into Microsoft 365 on Mac, Copilot collects much of what OpenAI does, plus your browsing history and what you do inside Mac apps. That can be shared with third parties and used for personalized ads or AI training.
- Google Gemini: Gemini records your conversations to improve Google products and its machine learning models. Human reviewers may read chats. Google can keep that data for up to three years even after you delete your activity. It says the data won't be used for targeted ads, but policies change.
- DeepSeek: This one collects the most. Prompts, chat history, location, device info, even your typing patterns, all used for AI training, product tweaks, and targeted ads. And it's stored on servers in the People's Republic of China.
The common thread is training. Almost everything you type feeds the model and shapes future answers, often without a clear moment where you agreed to it.
Where this bites you
Three things tend to go wrong.
- Privacy. Sensitive details you paste into a chatbot can reach developers or third parties, which raises the odds of a breach or misuse. Copilot has been criticized for exposing confidential data because the permissions it gets inside Mac apps are so broad.
- Security. A chatbot wired into your apps is another way in for an attacker. Researchers have shown Copilot can be manipulated into spear-phishing and data theft.
- Compliance. If a tool handles data in a way that breaks rules like GDPR, that's your legal problem, not the vendor's. Some organizations have already blocked ChatGPT on company devices over where and how it stores data.
How to use these tools without giving away the store
- Don't paste anything you'd regret. Keep confidential and personally identifiable information out of a chatbot unless you trust exactly how it's handled.
- Read the privacy settings, not just the marketing. Each platform handles your data differently. ChatGPT, for one, lets you opt out of data retention and sharing.
- Use the controls you already have. On a Mac or in a Microsoft 365 environment, tools like Microsoft Purview let you set rules around AI use and cut your exposure.
- Check back on the policies. The way these tools handle data shifts, so a setting you trusted last year may not hold today.
These tools save real time, and you don't have to give them up to stay safe. Treat the chat box like a shared room: useful, but not the place for anything you wouldn't say out loud. Watch what you put in, know where it goes, and you get the productivity without handing over more than you meant to.