AI that remembers: ChatGPT, Gemini and Claude compared

For a long time, every conversation with an AI started from scratch. You had to explain the tone, the project, and your own role all over again each time. That is over. In 2026, ChatGPT, Claude, and Gemini gained memory that remembers who you are and how you work beyond individual chats. That turns the tool into something closer to a colleague who already knows your preferences. It is useful, but it is worth looking closely at what is stored and who can access it. For European users in particular, one of the three has a clear limitation. More on that shortly.

What AI memory actually does

Memory means the assistant stores facts about you separately from the regular chat history and uses them in later conversations. Your name, your job, ongoing projects and your writing style. Tell it once that you want short answers without jargon, and the AI will follow that preference in the future without needing to be reminded. One thing is important to know and will matter later: memory and AI training are two different things, and each provider handles both differently.

The context window is separate from that. It is the working memory within a single chat, meaning everything that has been said so far in that one conversation and that the AI currently has in view. That window is limited. If a chat gets very long, the oldest parts drop out at the back, and the AI loses earlier details from that specific conversation. This is where memory comes in: anything that remains important over time is saved as a condensed fact in memory and survives the end of a chat. Two limits still remain. Memory does not store every word, only short notes about you. And something you mentioned only in passing in a long-closed chat is not automatically saved forever. How far memory technically reaches back within a chat and why it eventually hits limits will be covered in more detail in a later part of this series.

ChatGPT remembers a lot, users stay in control

ChatGPT works with two components. It remembers things you explicitly ask it to save, and it also learns from your previous chat history. A newer, self-updating system is being added as well, but it is being rolled out gradually and is not available everywhere yet. You can control all of this under Settings, Personalization, Memory. There, you can see the saved items, delete individual ones, or turn memory off entirely. For sensitive topics, there are temporary chats, which do not remember anything. The additional sources ChatGPT can learn from, such as files from your library or Gmail, are not enabled in the EU. Memory itself still works.

Claude shows its memory openly, now for free too

Claude takes the most transparent approach. The assistant keeps a summary of what it knows about you, and you can view and edit it directly. You can turn it off by pausing it or delete everything by resetting it. Since March 2, 2026, memory has been available to everyone on Claude, including the free plan. Anyone who wants to work unobserved can use an Incognito chat, which does not feed into memory or chat history. Unlike ChatGPT, this is the rule: if you delete a conversation, it also disappears from the memory synthesis. In return, the memory updates itself regularly.

Gemini remembers preferences, but deep Google access remains blocked in some regions

With Google, two functions need to be kept clearly separate. One is called Personal Intelligence, and it is the broadest access model of the three providers: with it, Gemini can actively look into your Google services, including Gmail, Docs, Drive, Photos, and Maps. Google is explicitly holding back that deep integration in the EU, including Germany, as well as in Switzerland and the UK. For expanded access to connected apps, Google's help pages also list Australia, Korea, and Nigeria as excluded. In practice, this means the request "Search my emails for next week's hotel booking" does not work in Germany for now.

Separate from that is the actual memory, meaning the ability to remember things you tell Gemini yourself, such as that you are vegetarian or prefer short answers. This memory feature does not carry that kind of country restriction in Google's help pages, and it is available with a German help page. The requirements are a personal Google account, a minimum age of 18, and activity turned on. Work or school accounts are excluded. Because Google rolls out features like this gradually, it is worth checking Gemini's settings before using it to see what is currently enabled for your account. For Germany, the bottom line is this: Gemini can remember your preferences. Convenient access to your own Google data, however, is not available.

What to watch out for

Memory is convenient, but it also remembers things that are better not stored permanently. Confidential client data, passwords, and health information do not belong in a saved profile. A common misconception: with ChatGPT, deleting a chat does not remove the facts remembered from it. Those have to be deleted separately in the memory settings. With Claude, it works the other way around: deleting a conversation also removes the remembered points derived from it. One point is often confused: whether the AI uses your inputs for training is controlled separately from memory.

Anyone who works with AI regularly can save real time with memory. They just need to know where it sits, how to clear it, and if there is no option at all, as with Gemini's deep Google access in the EU.