How to use AI chatbots safely: the privacy settings that matter

The AI chatbot privacy settings that matter most decide three things: whether your conversations help train future AI models, how long your chats are stored, and what the assistant remembers about you. ChatGPT, Gemini, Microsoft Copilot and Claude all let you control these, but the defaults differ and the switches live in different menus. Here’s what to change, where to find it, and what to keep out of the chat box entirely.

Turn off AI training first

Most consumer chatbots can learn from what you type unless you say no. In ChatGPT and Claude, opting out doesn’t cost you your chat history. Gemini is the exception, as explained below. Menu names shift between app updates, so treat the paths here as a guide and look for words like “privacy,” “data” or “activity.”

ChatGPT

Open Settings, choose Data controls and switch off Improve the model for everyone. OpenAI says your new conversations then won’t be used to train its models, and they still appear in your history. When you’re signed in, the choice follows your account across devices.

Gemini

Google ties training to one setting, Keep Activity, which is on by default for users 18 and over. You’ll find it under Settings & help, then Activity. With it off, Google says future chats won’t be saved to your activity or used to train its AI models unless you send feedback. The trade-off is real: you lose saved chat history, Gemini’s memory of past chats and some connected apps. If you’d rather keep your history, leave it on and use temporary chats for anything sensitive.

Microsoft Copilot

Microsoft made an updated Copilot app available in August 2026. For personal Microsoft accounts on that version, Microsoft says prompts, responses and file contents aren’t used to train foundation models. On the older app, look in your profile’s privacy settings for two training switches, one for conversations and one for voice, and turn both off.

While you’re there, open Settings, then Personalization. The One shared experience switch lets your Copilot chats personalize Bing, Edge and MSN, and a separate ads switch controls whether ads in Copilot are tailored using your chat history and saved memories.

Claude

In Settings, open Privacy and switch off the model-improvement toggle. Anthropic says your chats then won’t be used in future training, though data already in a training run or a finished model stays there. If you leave training on, Anthropic may keep de-identified chats for up to five years.

The feedback catch

The thumbs-up and thumbs-down buttons are a gap in most opt-outs. OpenAI says rating a response can send the entire conversation for training even after you’ve opted out. Anthropic stores rated conversations for up to five years, and Google collects the last 24 hours of your chats as context when you send feedback with Keep Activity off. Skip the rating buttons on anything personal.

Use temporary chats for sensitive one-off questions

A temporary chat works like a private browser window. It stays out of your history and isn’t used for training, which makes it the right choice for a question about a health worry, a legal letter or anything else you don’t want sitting in your account. “Temporary” doesn’t mean “instantly erased,” though.

Assistant	Off-the-record option	Copy kept for	Good to know
ChatGPT	Temporary chat	Up to 30 days, for safety	Choose Unpersonalized before your first message to skip existing memories
Gemini	Temporary chat	72 hours	Personal accounts only; no Gems or personalized answers
Claude	Incognito chat (ghost icon)	30 days by default	Not available inside projects and can’t be saved later
Copilot	Temporary chat in Copilot Chat for work accounts	Your organization’s retention policy	Not listed in Microsoft’s personal-account privacy controls as of September 2026

Two details trip people up. ChatGPT’s temporary chat can be saved to your history later, at which point it follows your normal training and memory settings. And a Claude incognito chat can’t be reopened once closed, so copy anything you need first.

Check what the chatbot remembers about you

Memory lets an assistant carry details such as your job, family or preferences from one chat into the next. It’s useful, but it’s also where sensitive facts quietly pile up, and deleting a chat doesn’t always remove what was learned from it.

• ChatGPT: Go to Settings, Personalization, then Memory to review, correct or delete what’s saved, or turn off memory and chat-history references. OpenAI notes that deleting a chat doesn’t necessarily delete a saved memory created from it.
• Gemini: Memory of past chats has its own switch under Settings & help, then Personal Intelligence, and it only works with Keep Activity on. To remove something Gemini remembered, Google says to delete every chat that mentions it and disconnect any connected app, such as Gmail, that holds it.
• Copilot: Under Settings and Personalization, turn off Saved memories or select Manage to remove items. Microsoft notes that turning memory off doesn’t delete what’s already saved, so use Delete all memories for a clean slate.
• Claude: Settings, then Memory, lists what Claude has saved as topics you can edit or delete. Memory is on by default for Free, Pro and Max plans, and Anthropic says memories generated from a deleted conversation aren’t removed automatically. Pausing memory keeps what’s saved; resetting it erases everything permanently.

A quick check works in most assistants: ask, “What do you remember about me?” If the answer includes something you’d rather it didn’t know, delete it at the source.

Know how long chats are kept, and delete what you don’t need

Every assistant stores your regular chats, but for different lengths of time.

Assistant	Regular chats	After you delete
ChatGPT	Stay in your account until you delete them; archiving only hides them	Scheduled for permanent deletion within 30 days, unless security or legal obligations require longer
Gemini	Auto-deleted after 18 months by default; you can pick 3 or 36 months, or never	Removed from your account, but chats already seen by human reviewers are kept for up to three years
Copilot	Kept in your chat history so you can revisit them	Delete single chats from the chat list, or everything through Microsoft’s privacy dashboard
Claude	Stay in your chat list until you delete them	Removed from history immediately and from back-end systems within 30 days

A few habits keep your history lean:

• Set a short auto-delete in Gemini. Three months is enough for many people. The option sits on the same Activity page as Keep Activity.
• Bulk-delete when you’re done with old chats. ChatGPT has Delete all chats under Data controls, and the ChatGPT deletion guide explains what’s removed. Gemini’s activity page and Microsoft’s privacy dashboard offer similar “delete all” options.
• Clean up shared links. Anyone with a shared chat link can forward it. Deleting a shared link in ChatGPT doesn’t delete the chat itself, and Google says deleting a shared Gemini chat doesn’t remove its public link.

What never to type into an AI chatbot

Settings reduce risk, but the safest data is data you never share. Google’s Gemini privacy notice puts it plainly: “Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services.” That advice applies to every assistant.

• Passwords, PINs and one-time codes. No chatbot task needs them. Keep them in a password manager instead.
• Full ID and account numbers, such as Social Security, passport, bank and card numbers. Replace them with “XXXX” before you paste a document.
• Medical, legal or financial records with your name on them. Remove identifying details or describe the situation in general terms.
• Other people’s private information, including screenshots of group chats or a friend’s diagnosis.
• Confidential work files in a personal account, which your employer’s rules may forbid.

Take extra care with connected apps and AI agents that act on your behalf. Google warns that Gemini can make mistakes such as sharing your data with third parties unexpectedly, and advises supervising its tasks closely. Finally, protect the account itself: your chat history is only as safe as your login, so turn on passkeys or two-factor authentication.

Work and school accounts follow different rules

Business versions of these assistants come with stronger data promises, but your employer, not you, holds the controls.

• ChatGPT Business, Enterprise and Edu: OpenAI doesn’t train on this content by default, and workspace owners set retention and memory rules.
• Gemini with a Google Workspace account: Google’s Workspace privacy hub says your content isn’t human-reviewed or used to train models outside your organization without permission. Admins decide how long chats are kept.
• Microsoft Copilot Chat with a work account: Prompts and responses aren’t used to train foundation models, but Microsoft says they’re logged, and your IT admin can view them with audit tools.
• Claude Team and Enterprise: Anthropic doesn’t train on this data by default, memory stays off unless an owner enables it, and incognito chats still appear in organization data exports.

The rule of thumb: assume your employer can see anything you type into a work account, and keep personal matters out of it. The reverse matters too. Pasting company files into a personal account to get around workplace rules strips away those business protections.

The bottom line

A few minutes in settings closes the biggest gaps: turn off training, review what the assistant remembers, set an auto-delete period where one exists, and use temporary chats for anything you wouldn’t want stored. Then keep passwords, ID numbers and other people’s secrets out of the chat box entirely. Menus move often, so if a label here doesn’t match your app, search its settings for “privacy,” “data” or “memory.”

Ready to get more out of these tools? Read our guide on how to write better AI prompts, browse our AI coverage and cybersecurity guides, or find more plain-English explainers on RedAndWhiteMagz.com.