AI assistant privacy: what ChatGPT, Gemini, Claude, Copilot, Perplexity and Alexa+ do with your data

Every AI assistant keeps something from your conversations. What matters is whether it trains on them, whether a person might read them and how long they’re kept. This guide answers those questions for ChatGPT, Gemini, Claude, Microsoft Copilot, Perplexity and Alexa+, using each company’s own help centers and privacy pages as of October 2026. For step-by-step deletion, see our separate guide to deleting AI chat history. Settings vary by country, plan and app version, so treat menu names as a guide.

The comparison at a glance

ChatGPT

Training. For personal accounts (Free, Go, Plus and Pro), OpenAI may use your chats to improve its models while Improve the model for everyone is on. You switch it off in Settings, Data controls. It applies to your whole account on every device. OpenAI says it doesn’t train on content from Business, Enterprise, Edu or Healthcare workspaces by default.

Human review. OpenAI says authorized staff can access content only for set reasons: abuse or security investigations, support, legal matters and, unless you’ve opted out, model improvement.

Retention. Deleted chats are scheduled for permanent deletion within 30 days, with exceptions for de-identified data and security or legal holds. In Enterprise and Edu workspaces, admins set how long content is kept.

Temporary Chat. It stays out of your history, doesn’t create memories and isn’t used for training. OpenAI says temporary chats are deleted from its systems after 30 days and may be reviewed only to monitor for abuse.

Voice. OpenAI’s Voice FAQ says audio and video clips from voice chats are stored with the chat transcript and removed within 30 days of deleting that chat. Clips aren’t used for training unless you choose to share them, but if Improve the model for everyone is on, transcripts of your voice chats can still be used. Our ChatGPT Voice guide covers the modes.

Memory and ads. Memory has its own settings, explained in ChatGPT memory explained. Free and Go users may also see ads; ChatGPT ads explained covers what data they use and the Ad Controls menu.

Privacy Center. OpenAI is rolling out a Privacy Center that explains each option and links to the matching setting. On the web it’s under your account menu, Help, Privacy center.

Gemini

Training and Keep Activity. Google ties almost everything to one setting, Keep Activity. Its Gemini Privacy Hub says that when it’s on, your activity is used “to provide, develop, and improve its services (including training generative AI models).” Turn it off and future chats aren’t used for training unless you send feedback, though Google keeps them for up to 72 hours to respond.

Human review. Google says a subset of chats is read by human reviewers after being disconnected from your account, and reviewed chats are kept for up to three years, even if you delete your activity. Its advice is plain: “Please don’t enter confidential information that you wouldn’t want a reviewer to see.”

Retention. Gemini activity auto-deletes after 18 months by default. You can change that to 3 or 36 months, or keep it indefinitely.

Temporary chat. Temporary chats aren’t used to train Google’s models or to personalize your experience, and are kept for 72 hours.

Voice and Gemini Live. Google says your audio, and Gemini Live video and screen shares, aren’t used to improve its services by default. That’s a separate opt-in setting.

Memory and ads. With memory on, Google says Gemini “may use sensitive info you provided in your chats” to personalize responses; check Settings & help, Personal context. Google also says: “Your Gemini Apps chats are not being used to show you ads.”

Work and school accounts. Google’s Workspace help center says Gemini app chats from qualifying Workspace editions aren’t used to train models and aren’t reviewed by humans, and admins control retention. To switch Gemini off on your phone, see how to turn off Gemini.

Claude

Training. Since Anthropic’s 2025 terms update, Claude Free, Pro and Max users choose whether their chats and coding sessions help train Claude, and Anthropic’s pricing page lists model training as an opt-out on those plans and on Ultra. The toggle is Help Improve our AI models under Settings, Privacy. Team, Enterprise, education, government and API use aren’t trained on by default.

Retention. This is where the setting matters most. With training on, Anthropic keeps de-identified data from new or resumed chats for up to five years in its training pipelines. With it off, a deleted chat is removed from back-end storage within 30 days. Either way, chats flagged by Anthropic’s safety systems can be kept for up to two years, and anything you send with the thumbs up or down buttons is kept for five.

Incognito chats. Click the ghost icon in a new chat outside a project. Incognito chats aren’t saved to your history, don’t use or add to memory, and aren’t used to improve Claude even if training is on. Anthropic keeps them for 30 days by default. On Team and Enterprise plans, owners can still get them through a data export.

Memory. Memory is on by default on Free, Pro and Max, but Anthropic says Claude doesn’t store sensitive topics such as health, religion or politics unless you turn on Include sensitive topics in memory. On Team and Enterprise, memory is off for each member until they turn it on. Our beginner’s guide to Claude explains how memory works in projects.

Enterprise retention. Enterprise owners can set a custom retention period of at least 30 days; otherwise Anthropic says data is kept indefinitely.

Microsoft Copilot

Microsoft released an updated Copilot app on August 18, 2026, with its own privacy pages. These rules cover that version.

Training. Microsoft’s activity history page says: “Prompts, responses, and your file contents when using the Microsoft Copilot app aren’t used to train foundation models.” Optional feedback may be used to improve Copilot, but not to train those models. The older app had separate opt-out toggles for text and voice training; Microsoft’s pages for the updated app don’t describe them.

Memory. Go to Settings, Personalization and use the Saved memories toggle. Manage lets you delete individual memories or all of them. Microsoft notes that turning memory off doesn’t delete what’s already saved.

Ads. The same Personalization menu has Allow ads personalization. Microsoft says Copilot doesn’t show personalized ads to users under 18, whatever the settings.

What’s not documented. The updated app’s privacy pages don’t state a retention period or describe human review, and they don’t mention a temporary chat mode. Microsoft’s FAQ for the older app said conversations were kept for 18 months and some were reviewed by people, with no opt-out.

Work accounts. When you sign in with a work or school account, Microsoft applies enterprise data protection automatically, and prompts and responses aren’t used to train foundation models. Your admin can review logs. For turning Copilot off in Windows and Office, see how to turn off Copilot.

Perplexity

Training. Perplexity’s setting is called AI Data Retention, and its help center says it’s on by default for Free, Pro and Max. It lets Perplexity use your searches to improve its models. Turn it off under Settings, Preferences. Perplexity says opting out only affects data collected afterwards, and data already collected can’t be removed from training.

Memory. Perplexity says it may also use memory content to improve its models unless AI Data Retention is off. You can switch search history and saved memories off separately and delete them. It may keep a log of cleared memories for up to 30 days. Our Perplexity tutorial walks through the Memory screen.

Incognito. Perplexity says incognito searches aren’t stored or used for memory. Its help center says incognito threads expire after 24 hours, and threads you start while logged out expire after 14 days.

Enterprise. Perplexity says Enterprise data is never used to train its models, and admins can force incognito mode for everyone and set custom retention.

Ads. Perplexity’s privacy notice says you can opt out of targeted advertising through its cookie settings.

Alexa+

What gets recorded. Amazon says an Echo only streams audio to its cloud after it hears the wake word, or when you press the action button, and the light ring turns blue while it’s sending. Amazon says joining Alexa+ switches on Follow Up Mode, which lets you make a follow-up request without repeating the wake word. Every Echo also has a microphone-off button.

Retention. Amazon’s help page says recordings and transcripts are saved by default. In the Alexa app, go to More, Alexa Privacy (or amazon.com/alexaprivacysettings) to delete them automatically after 3 or 18 months, or choose not to save recordings at all. With no recordings saved, Amazon still keeps text transcripts and typed requests for 30 days after your last interaction in a chat, and your voice ID is deleted.

Training and human review. Amazon says it uses Alexa interactions to train its machine learning models, and that humans review “an extremely small sample” of interactions. You can manage whether your voice recordings are used to improve its services in the same Alexa Privacy menu.

What deletion doesn’t cover. Deleting recordings doesn’t remove attachments, Remember This notes, records of actions Alexa took or transcripts tied to feedback. Amazon notes that Remember This notes can be accessed by anyone using your devices, and that Alexa+ may share relevant details from your interactions with skill developers.

Ads. Some Alexa content, such as music or news on an Echo Show, can carry interest-based ads, which you can turn off. See how to use Alexa+ and the Alexa+ profile for setup.

How to choose the right settings

We compared the six on the training default, human review, retention, private modes and business plans. A few patterns stand out.

• Paying doesn’t buy privacy on personal plans. ChatGPT Plus, Claude Pro and Perplexity Pro follow the same training settings as the free tiers. Only business, enterprise and education plans change the default.
• Human review is the biggest difference. Gemini is the most explicit that people read a sample of chats and keep them for years, which makes Keep Activity the most important setting in this guide.
• Turning off training isn’t retroactive. Anthropic, Perplexity and Amazon all say data already used for training can’t be pulled back out.
• Private modes aren’t instant deletion. They range from 24 hours (Perplexity) to 30 days (ChatGPT and Claude).

The 10-minute privacy checklist

1. ChatGPT: turn off Improve the model for everyone in Settings, Data controls, and check that voice audio sharing is off.
2. Gemini: decide on Keep Activity. If you keep it on, shorten auto-delete to 3 months.
3. Claude: check Help Improve our AI models in Settings, Privacy, and leave sensitive topics out of memory.
4. Copilot: update the app, then review Saved memories and Allow ads personalization in Settings, Personalization.
5. Perplexity: turn off AI Data Retention and review saved memories.
6. Alexa+: in Alexa Privacy, set recordings to 3 months or don’t save, and review Remember This notes.
7. Read your memories everywhere and delete anything out of date or too personal.
8. Disconnect apps you no longer use, such as email or calendar connectors.
9. Find each private mode so it’s one tap away.
10. Clear old history with our deletion guide.

What not to share with any AI assistant

Even with training off, chats are stored for a while and may be checked for abuse. Keep these out:

• Passwords, PINs, recovery codes and one-time codes.
• Government ID, passport, Social Security and bank or card numbers.
• Full medical records or test results tied to your name. Ask the general question instead.
• Confidential work material on a personal account. Use your employer’s approved business plan.
• Anything you’d regret a reviewer reading, to borrow Google’s own advice.

Verdict

If you want the fewest settings to manage, Claude asks you up front, offers incognito chats on every plan and keeps sensitive topics out of memory by default. The updated Copilot app says it doesn’t train its foundation models on your chats at all, though Microsoft documents less about retention. ChatGPT and Perplexity are fine once you flip one switch. Gemini needs the most thought because of human review, and Alexa+ needs a visit to Alexa Privacy to stop recordings piling up.

Where to go next

Start with how to delete your AI chat history to clear what’s already stored, then read ChatGPT memory explained and ChatGPT ads explained if ChatGPT is your main assistant. For a wider comparison, see ChatGPT vs Claude vs Gemini.