Chatbot privacy settings: history, memory, training and ads

Which settings decide what happens to your chats?

With ChatGPT, Gemini and most other chatbots, four separate questions matter. Is your history saved? Does the chatbot remember details about you beyond a single conversation, its memory? May your chats be used for training future AI models? And does the service show you ads?

These settings are only partly linked. Turning off training in ChatGPT does not delete your history, and deleting a chat does not automatically delete a memory saved from it. In Gemini, one setting controls history and training together. What should never go into a chat at all is covered in Before you ask an AI: what to leave out of your text.

An everyday example

Fictional example: Clara lives in Brussels and uses ChatGPT for small tasks. Once she asked it to help her write to her son’s school about extra support for his dyslexia. Weeks later, while looking for birthday present ideas, ChatGPT suggests books ‘suitable for a child with dyslexia’. She had mentioned it only in that one conversation.

Clara opens her settings. The memory summary contains several details about her son. She deletes those entries and the old school chat, then turns off model training. Questions about her children now go into a temporary chat, without names. She keeps memory on, because she likes ChatGPT knowing she prefers short, plain answers.

How do you set up ChatGPT?

The labels come from OpenAI’s English help centre. Signed in, the training setting applies on all your devices; signed out, only in that browser.

New conversations are then no longer used for training but stay in your history. OpenAI names one exception: if you rate an answer with thumbs up or thumbs down, the entire conversation linked to that feedback may still be used for training.

Saved memories are stored separately from your chat history, so delete both the memory and the chat. After you turn off ‘Reference chat history’, information from past chats is scheduled for deletion within 30 days, OpenAI says.

Tip: For a one-off sensitive question, use a temporary chat. It does not appear in your history, does not create memories and is not used to improve models. Before starting, you can choose ‘Unpersonalized’. OpenAI may keep temporary chats for up to 30 days for safety purposes.

How do you set up Gemini?

In Gemini, almost everything depends on ‘Keep Activity’. While it is on, Google stores your chats and shared files and photos in your Gemini Apps Activity and uses them to improve its services, including AI training. It is on by default for users aged 18 and over.

Even with Keep Activity off, new chats are stored with your account for up to 72 hours so Gemini can respond; they do not appear in your activity. Google has a subset of chats read by human reviewers, some of them external service providers, after the link to your account has been removed. Reviewed chats are retained for up to three years, even if you later delete your activity.

Whether Gemini uses your past chats to personalise answers is set in the Personal Intelligence settings of the Gemini app. Google says this needs a personal account, an age of 18 or over and Keep Activity turned on. Gemini can also use information from connected Google apps such as Gmail or Google Photos; to check which services can access your accounts, see Connected apps: which services can access your account?.

Ads in ChatGPT: what is personalised and what can you change?

According to a Euronews report of 21 August 2026, ChatGPT ads were to be switched on in 31 European countries from 24 August; OpenAI itself describes a gradual rollout. Only the Free and Go plans are affected; Plus, Pro, Business, Enterprise and Edu stay ad-free. Accounts that OpenAI identifies as belonging to under-18s see no ads. An ad appears below a ChatGPT response.

OpenAI says personalised ads are not initially available in the European Economic Area (EEA) or Switzerland. An ad then relies only on the current chat and on basic context such as general location and language. OpenAI does not place ads near sensitive topics such as personal health, mental health or politics, and shows none in temporary chats. Advertisers, it says, receive no chats, memories, names or email addresses, only aggregated figures such as views and clicks.

• Ad settings: in Settings, the ad controls section offers ‘View History and Topics’ to see which ads you have been shown. ‘Delete ads data’ removes your ad data; this can take up to 30 days and does not affect your chats.
• A single ad: use the three-dot menu on the ad to choose ‘Hide ad’, ‘Report ad’ or ‘About this ad’.
• Ad-free on the Free plan: in the ad controls, select ‘Change plan to go ad-free’ and then ‘Reduce message limits’. You then see no ads, but you can send fewer messages and some tools, such as image generation and deep research, are not available.

What about Copilot, Claude or Le Chat?

In October 2025, the French data protection authority CNIL listed where major chatbots keep their training switch. English labels may differ:

• Microsoft Copilot: separate switches for model training on text and on voice, in the privacy section of the account settings.
• Claude by Anthropic: a switch to help improve Claude, also in the privacy section of the settings.
• Le Chat by Mistral: an option on whether your interactions may be used for training, in the preferences.
• Meta AI: no simple switch, but an objection through Meta’s privacy rights centre.

The CNIL stresses that the list does not assess whether these procedures comply with the GDPR. If you are still deciding which service suits you, see Choosing an AI chatbot: free, subscription or ads.

What are your rights, and who can help?

In the EU and the EEA, the General Data Protection Regulation (GDPR) applies. You can object at any time and free of charge to your data being used for direct marketing, and withdraw consent without detriment. Requests about your rights, such as access or erasure, should be answered within one month at the latest.

In the EEA and Switzerland, Gemini is provided by Google Ireland Limited. To object or ask for inaccurate personal data in answers to be corrected, select ‘Report legal issue’ in the More menu below a Gemini response, or use the request form in Google’s help centre.

The GDPR does not apply automatically in Switzerland. According to the FDPIC, people using a chatbot have a right to know whether the data they enter is used to improve self-learning programs. This guide explains settings and general rights; for a specific dispute with a provider, seek individual advice.