ChatGPT, Claude and Gemini privacy settings: a step-by-step guide

To protect your conversations with ChatGPT, Claude or Gemini, three moves cover most needs: opt out of model training, switch to a temporary or incognito chat for sensitive topics, and tidy up your history and memory. This tutorial walks you through each step, based on the site’s assistant profiles and the official help pages we consulted on October 22, 2026.

Before you start

What you need

• Your account, signed in on the website or in the assistant’s app.
• About five minutes per assistant.
• A little flexibility: providers rename their menus regularly. We quote the labels exactly as they appear in official sources.

Three controls worth knowing

1. Training. By default, on consumer plans, your chats may become training data, the text used to improve future models. There is usually a switch to opt out.
2. Ephemeral chats. Temporary Chat or incognito mode: the conversation stays out of your history and isn’t used for training.
3. History and memory. An assistant’s memory keeps details about you from one chat to the next. You can review it and clear it.

To understand what happens to your messages once they are sent, also read our guide on AI and privacy: what happens to your conversations.

Step 1: set up ChatGPT

The labels below come from the site’s ChatGPT profile, checked on October 22, 2026.

1. Open Settings, then Data controls.
2. Turn off “Improve the model for everyone”. New chats stay in your history but are no longer used to train OpenAI’s models.
3. For anything sensitive, start a Temporary Chat. It doesn’t appear in your history, doesn’t create memories and isn’t used for training. OpenAI says it may keep these chats for up to 30 days for safety reasons.
4. Review your memory. Under Settings, in the Personalization section, check what ChatGPT has saved about you and delete anything that shouldn’t be there. A Temporary Chat neither uses existing memories nor creates new ones.

What you should end up with: your everyday chats remain available, they no longer feed model training, and sensitive topics leave no trace in your history.

Worth knowing: according to the same profile, the Business plan doesn’t train on company data by default, and European data residency is limited to the Enterprise and Edu plans.

Step 2: set up Claude

For Claude, we rely on Anthropic’s privacy center, especially the page “Is my data used for model training?”, consulted on October 22, 2026.

1. Open your account settings and go to Privacy Settings.
2. Turn off “Model Improvement”. Anthropic says it will then not use your previous or new chats for future model training.
3. For anything sensitive, use incognito mode. According to Anthropic, incognito chats are never used to improve Claude, even if Model Improvement is turned on.
4. Think before hitting the thumbs buttons. When you send feedback, the related conversation is kept for up to 5 years, de-linked from your identity, and may be used to train models.
5. Delete what shouldn’t stay. According to the page “How long do you store my data?”, a deleted chat leaves your history immediately and Anthropic’s back-end systems within 30 days.

What you should end up with: your chats no longer go into the training pipeline. If you leave the setting on, Anthropic says de-identified data may be kept for up to 5 years. The site’s Claude profile also notes that Team and Enterprise plans don’t train on your data by default.

Step 3: set up Gemini

For Gemini, we follow the site’s Gemini profile and Google’s help page on Gemini Apps Activity, consulted on October 22, 2026.

1. Open Gemini Apps Activity.
2. Turn off “Keep Activity”. Your chats are then held for 72 hours, long enough to respond and keep the service safe, and are not used to improve the models unless you send feedback.
3. Or set auto-delete. If you’d rather keep your activity, Google offers auto-delete after 3 months, 18 months (the default) or 36 months, or no time limit.
4. Use Temporary Chats for sensitive topics: they don’t appear in your history and aren’t used for personalization or training.
5. Keep human review in mind. Google says a portion of chats is reviewed by trained people. Those reviewed chats are kept for up to three years, disconnected from your account, even if you delete your activity.

What you should end up with: Gemini works as usual, but your chats aren’t kept beyond 72 hours or used for training. Google advises not entering confidential information you wouldn’t want a reviewer to see.

The three assistants at a glance

	ChatGPT	Claude	Gemini
Training switch	“Improve the model for everyone”	“Model Improvement”	“Keep Activity”
Ephemeral chat	Temporary Chat	Incognito mode	Temporary Chat
Retention afterwards	Up to 30 days (temporary)	30 days after deletion	72 hours (activity off)
Watch out for	Saved memories	Feedback kept 5 years	Human review, 3 years

Sources: the site’s assistant profiles and Anthropic’s and Google’s help pages, consulted on October 22, 2026.

Check yourself You want an assistant to proofread a letter that mentions a health condition. Which setting should you use?

An ephemeral chat: Temporary Chat in ChatGPT or Gemini, incognito mode in Claude. The conversation stays out of your history and isn’t used for training. Still, remove your name, address and any ID or account numbers before pasting the text.

What if privacy is your top priority?

The table below lists the features of these three assistants, including memory, as published on the providers’ websites.

Assistant	Free	Paid from	Web search	Agent mode	Voice	Images
ChatGPT	Limited	$8/mo	Yes	Yes	Yes	Yes
Claude	Limited	$20/mo	Yes	Yes	—	—
Gemini	Limited	€4.99/mo	Yes	—	Yes	Yes

Some services make privacy their main selling point. According to its profile, Duck.ai, DuckDuckGo’s AI chat, works without an account, strips your IP address before sending requests to model providers and never uses chats for training. Lumo, from Proton, takes a similar approach: according to the site’s data, Proton says it never uses conversations to train models, encrypts your chat history so that it cannot read it itself, keeps no logs and runs its own servers in Europe. These services have fewer features than the big three, so weigh what you gain in privacy against what you lose in convenience. To compare several assistants on the same criteria, open our AI assistant comparison tool.

Good habits beyond the settings

• Never paste a password, a card security code or an ID number.
• Hide other people’s details. In the EU and the UK, the GDPR also protects the people you mention, and it gives you the right to access and erase your data with the provider.
• Sharing an account with your family? History and memory are visible to everyone. Our guide to kids and AI explains how to set boundaries.
• Does your assistant read web pages or your email? A booby-trapped text can try to hijack it: that’s prompt injection, explained simply.
• A message claiming to be an assistant’s support team asks for your login details? That’s a classic scam, covered in our guide to AI scams.