Gemini’s Memory Summary Lets Users View and Edit What It Knows
Gemini’s new Memory summary shows what the chatbot remembers, lets users correct false or sensitive details and keeps those changes in sync across web and app.

Why it matters
- Gemini now offers Memory summary on its website and in its mobile and desktop apps.
- Users can add, remove or revise stored details and check Memory history for recent updates.
- The website and desktop apps support direct pencil-icon editing; mobile apps rely on typed or spoken requests.
- ZDNET’s tester saw entries under four named categories: Essentials, Business & Career, Reading & Interests, and Culinary Preferences.
- Corrections to a mistaken Madison County location synced across the Gemini website and apps in ZDNET’s test.
By Sophie Lindqvist
Gemini now lets users view and edit the personal details it remembers through a new “Memory summary” option available on the Gemini website and in its mobile and desktop apps.
ZDNET reports that the summary can contain tidbits Gemini gathered from past conversations or facts users told it directly. The stated purpose is to tailor and personalize responses, but the source also warns that personal details may end up training the AI or receive human review.
That makes direct inspection and removal more than a convenience for people deciding how much information they want a chatbot to retain. In a hands-on assessment, ZDNET’s tester wrote: “I give the company credit for offering this easy way to control the AI’s memories.”
The tester found the feature useful for recalling details that guide responses, while preferring to keep Gemini and Google from knowing more than necessary. Users can keep personalization and remove what they do not want retained.
What does Gemini’s Memory summary show?
Users reach the feature through Gemini’s settings on the web or in its mobile or desktop apps. They open the left pane containing previous chats, select the Settings icon at the bottom, and then choose Personal Intelligence followed by Memory summary.
If that option is grayed out, the underlying Memory feature is off. Users can switch it back on and disable it again later if they wish.
In ZDNET’s test, the summary arranged details into sections including Essentials, Business & Career, Reading & Interests, and Culinary Preferences. Users can open each category and inspect the individual items inside it. These examples show the scope of the profile: Gemini may hold not only practical facts, but also career information, interests and food preferences.
The summary acts as a readable record of what Gemini believes it knows about a user. Its usefulness depends on both coverage and accuracy. ZDNET’s testing uncovered errors in both categories, including a wrong location and an unsupported business affiliation.
How do you edit what Gemini remembers?
On the Gemini website or in a desktop app, users click the pencil icon at the top of Memory summary. The interface lets them directly edit the text by adding, removing or revising specific details, then save the changes. In ZDNET’s test, a saved correction synced to both the Gemini website and its apps.
Users can also add or remove details by asking Gemini directly, rather than changing the stored text. This method works on mobile and desktop, and users can type or speak the request at the prompt. That matters because the mobile apps do not offer direct pencil-icon editing, leaving conversational instructions as the editing route there.
ZDNET also describes a separate history control for checking whether a correction stuck. Selecting the three-dot icon at the top of the summary opens Memory history, which shows the latest changes and updates. The record gives users a place to verify new or revised information rather than trusting the request alone.
Two ways to change the summary, direct editing and conversation, address different needs. The first is precise for desktop users; the second gives mobile users a way to make the same correction without navigating a text editor.
Can you correct a false memory?
Yes in practice, according to ZDNET’s testing. The tester’s summary said they lived in Madison County, while the tester actually lives in New York. They edited the text, saved it, and found the correction had synced across the Gemini website and apps.
The second example tested Gemini’s handling of a false business relationship. The summary claimed the tester ran a business called Batter Up that sold collectible baseball cards. In reality, the tester had mentioned such a store only while testing Gemini’s image-creation skills, specifically to design a logo for a baseball-card seller.
The tester told Gemini they did not run a business called Batter Up and asked it to remove that item from memory. ZDNET’s tester wrote: “Poof, that detail quickly disappeared.” This case shows that natural-language editing can remove a conclusion Gemini drew from an earlier conversation, even when the user never stated that conclusion directly.
Why does this matter for privacy?
ZDNET’s framing is blunt: Google is not known for privacy-friendly behavior, and Gemini’s memory carries that concern. The source nevertheless credits the company for making stored details visible and editable. Memory can make responses more relevant through personalization, but users should still be able to decide which details remain attached to their account.
The ability to remove a wrong or unwanted detail matters even when users keep memory enabled. ZDNET’s tester used that approach: preserve personalization for useful details, but edit or delete entries that were inaccurate, overly personal or simply undesirable. The feature does not require users to choose between a blank profile and an unchanged profile.
There are limits to what the source demonstrates. The test focuses on the viewing and editing workflow described here, along with the Memory history control. It does not measure how personal details may affect Gemini’s training or human review outside the summary.
The test also does not show how quickly deleted details leave Google’s broader systems or whether every downstream use disappears at once. The source provides no technical account of that process. The demonstrated control is the visible profile, not Google’s unseen data systems.
ZDNET found that users can correct the profile Gemini presents and inspect recent updates through Memory history. It did not test the company’s retention systems, training-data removal or access by human reviewers.
What should users check first?
Start with Memory summary and read the entries rather than assuming Gemini’s recollection is exact. ZDNET’s examples show two simple failure modes: an incorrect place of residence and a business relationship inferred from a design test. Both appeared as confident summary entries.
Then use Memory history to check the latest changes after an edit or request. On desktop, direct text editing offers the clearest control; on mobile, a typed or spoken instruction provides the equivalent route. Users can also turn the underlying Memory feature back off if they no longer want it active.
The practical result is a more manageable form of personalization. Users can keep useful memory while pruning false, sensitive or unwanted details, and ZDNET’s test suggests those edits can travel across Gemini’s website and apps. The next measure of the feature is whether accuracy and synchronization remain reliable after repeated corrections.
Original: support.google.com
More from Sophie Lindqvist
Show full bio
Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.
555 articles
Related articles
- Google Says Gemini Doesn't Train on Your Emails — With Exceptions
- Google rolls out Gemini Personal Intelligence to Android Auto
- Apple's 27.2 updates bring Siri conversation recall to iPhone today
- OpenAI's 'Dreaming' memory system goes mainstream in ChatGPT
- Georgia Tech tool recovers deleted local AI chats from phone memory