Google AI Gemini now automatically remembers past chatbot conversations

Google is updating its Gemini AI model with an automatically activated memory. While users previously had to signal to the chatbot what it should remember, this is no longer necessary. Gemini now remembers past conversations automatically. However, this function can also be switched off. Google is also introducing temporary chats that are neither remembered nor used for AI training. This should make it possible to clarify “private questions” virtually in secret.

Google's Gemini chatbot was given a memory back in November last year, but only for paying AI customers. What the AI chatbot is supposed to remember, however, had to be explicitly specified until now, for example with phrases such as “Don't forget” or “I can only” or “Always mention.” However, the function can also be switched off, and reminders can be deleted individually by the user. In addition, this function was initially limited to “Gemini Advanced,” so a “Google One AI Premium subscription” was required.

AI chatbot with automatic reminders

In the future, the aforementioned phrases for details that the AI chatbot is supposed to remember will no longer be necessary, as Gemini will activate its memory. This can be useful in many cases, for example, when it comes to recommendations that suit the user. This is because the AI can then give tips based on the user's preferences that emerge from past conversations. Google cites book recommendations as an example. If book topics have been discussed in previous conversations, the chatbot should be able to refer to suitable books.

Gemini's memory is activated automatically but can be deactivated in the Gemini app settings if you do not want to disclose your data. However, when introducing the conversation histories last year, Google already stated that the “stored information will never be shared or used to train the model.” This is intended to guarantee the user's privacy.

Google will be rolling out the reminder function in various countries over the next few weeks, with the personalized conversations initially being available in the Gemini 2.5 Pro AI model. The expansion to other countries and to the Gemini 2.5 Flash AI model is promised for later.

Temporary chats and new privacy settings

A new Gemini function is the temporary chat. These conversations are not listed in the conversation histories and are only saved for 72 hours so that they can be revisited at a later date. The AI is also not supposed to remember the topics discussed in temporary chats, so that “private topics” can be discussed there that users want to keep secret but for which they still need help. Google explicitly promises that these conversations will not be used to train the AI models, but that they will also not contribute to the personalization of AI usage.

In addition, the “Gemini Apps Activit” in the Google settings will be renamed “Keep Activity.” If this is activated, Google will use parts of files or images shared with Gemini to “improve Google services for everyone” from September 2. This should amount to the use of this data for AI training. If the “Gemini Apps Activit” has already been switched off manually, this should also apply to the “Keep Activity.”

(fds)