Gemini Shows What It Remembers About You: "Memory Summary" Lets You Edit and Delete Everything

Imagine this: for six months you discussed trip preparations with an assistant, then you changed your phone number, moved, and gave up running. Yet the assistant keeps "remembering" the old you. Worse, you don't know what exactly it remembers. Memory in chatbots has long been convenient, but it remains a black box. Google is taking a step toward opening that box: Gemini now has a "Memory summary."

What exactly has appeared

In the web version, the section is located under Settings → Personal data, or at gemini.google.com/personalization-settings. Inside is not a feed of scattered notes but a structured profile. It is divided into categories:

• The Essentials: a short summary of your background, location, and key connections;
• Career & Professional Projects: work and projects;
• Travel & Daily Gear: trips and everyday items;
• Media & Entertainment: what you watch and listen to;
• Fitness & Lifestyle: sports and lifestyle.

Each category contains specific items. There are two ways to manage them. The first is a text field at the bottom of the screen where you type a command in plain language. "Remember that I speak Japanese" adds a fact, and "Delete my old phone number from memory" removes an outdated one. The second is a pencil icon at the top right for larger edits. The three-dot menu contains Memory history, which shows when a specific piece of information was added or changed.

Why this matters more than it seems

At first glance this is just a settings page. But it addresses three familiar problems of any assistant with memory.

First, transparency. As long as users can't see what is known about them, they either distrust memory and turn it off, or forget about it and are surprised by strange answers. When the profile can be read, the question "where did it get that?" goes away.

Second, outdated data. Memory without an editing mechanism accumulates junk: an old number, a former place of residence, a project closed long ago. Being able to delete an entry with a single phrase lowers the cost of such an error.

Third, change history. Memory history turns memory from a set of facts into a log that shows what got into it and when. For a product that collects personal data, it is also an argument in the conversation about trust.

How the control is designed is also interesting. Instead of a separate interface for each operation, Google uses the same conversational channel as the assistant itself. Users don't need to hunt for the right button; it's enough to say what to forget.

Limitations and what to do about them

For now the feature is heavily restricted. It is available to Google AI subscribers in the US, and only if Keep Activity and Memory are turned on. It is not in the Android and iOS apps, though it is expected to work similarly there. The announcement doesn't make clear how exactly edits affect the model's answers or when the feature will expand beyond the US, so it's too early to draw conclusions about the depth of control.

For those building their own assistants or agents with long-term memory, the takeaway is practical. If you store data about a user, give them the ability to:

• see it in readable form, not as a raw log;
• group it by clear topics, not by the technical structure of the storage;
• edit and delete it with an ordinary phrase, not only through a form;
• trace when and what changed.

Product teams should take a closer look at the pattern itself: memory management is becoming part of basic UX rather than a hidden option in settings. And if you simply use Gemini and live in the US, open this section and delete what's outdated: an assistant that remembers you as you are today is more useful than one stuck in last year.