Gemini 3, Google's latest AI model, offers a unique approach to memory compared to ChatGPT and Claude. While ChatGPT focuses on continuous personalization, Gemini opts for a more cautious method, only activating personalization when explicitly prompted by the user. Google’s memory system serves over 650 million monthly users, and its design reflects a careful balance between personalization and user control. The memory architecture in Gemini consists of a conversation summary called user_context, which includes demographic information, interests, relationships, and dated events organized in a structured format. This allows for efficient updates, as different sections can be refreshed independently based on user interactions. Notably, each memory statement is linked to a specific source interaction, providing a timestamp and rationale that enhance transparency. This feature helps avoid confusion about outdated information, allowing the model to treat statements as historical rather than permanent. Gemini's approach to memory also emphasizes user control. Unlike ChatGPT, where personalization is constant unless disabled, Gemini requires users to explicitly request it through specific phrases. This level of restraint means that the AI won’t incorporate past interactions unless users invite it to do so, potentially reducing the instances of irrelevant or confusing responses. Overall, Gemini's design prioritizes user trust and clarity, aiming to deliver a more tailored experience while still maintaining strict boundaries around when and how memory is utilized.