AivexaNewsSearch
AI news for builders and product teamsChecked every hour

I checked Gemini’s new Memory summary to see what it knew about me – how to see and edit yours

Collected Nov 12, 2026

Google has added a Memory summary to Gemini that lets users inspect and change what the assistant has recorded about them over time. The capability is live across the Gemini website, the iOS and Android mobile apps, and the Windows and Mac desktop apps, so the same controls are reachable wherever the product runs.

The memory draws on two kinds of material: details picked up during past conversations and facts a user has stated directly to the assistant. Memory summary presents those items grouped by category. In one user's account the information was sorted under headings such as Essentials, Business & Career, Reading & Interests, and Culinary Preferences, with individual entries listed beneath each one. A separate Memory history view, reached from a three-dot menu at the top of the summary, shows recent changes to what Gemini remembers.

Reaching the feature takes a short path. From the left-hand pane of previous chats, choose the Settings icon at the bottom, select Personal Intelligence, then open Memory summary. If that option appears grayed out, the Memory feature itself is switched off; enabling it restores access, and it can be disabled again afterwards.

Editing works differently depending on the surface. On the website and the desktop apps, a pencil icon at the top of the summary opens the text for direct changes: details can be added, removed or rewritten, and the edits are saved with a Save action. On mobile there is no direct editing interface. Instead, users can simply tell Gemini what to change in the prompt, typing or speaking a request, and the assistant carries it out. The conversational route also works on desktop, so it serves as a fallback on every platform.

Saved edits sync across the website and the apps, so a correction made in one place carries over to the others. That synchronization is the detail worth noting for anyone maintaining an account across several devices: the memory is a single store, not a per-device setting.

The feature also offers a way to catch the assistant's mistakes. In one account, the summary claimed the user lived in Madison County even though the user lives in New York; rewriting that entry and saving fixed it everywhere. In another case, the summary recorded that the user ran a business called Batter Up selling collectible baseball cards. That was false, traceable to an earlier session in which the user had described such a store only to test Gemini's image generation by designing a logo for a baseball card seller. Telling Gemini the business did not exist and asking for the entry to be removed cleared it promptly.

This points to a broader quirk of assistant memory: entries are often inferred from context rather than explicitly stated, so passing hypotheticals, test prompts and role-play can harden into stored facts. A summary screen makes those inferences visible, which is useful precisely because they are hard to predict from the outside. It also means the list is worth a periodic skim rather than a one-time check.

Google does not have a strong reputation on privacy, and the same questions apply to Gemini. The memory is meant to personalize and steer responses, but stored personal details may also feed model training or be reviewed by people, so controlling what the assistant retains has a practical purpose beyond tidiness. Keeping memory on while pruning specific entries is one middle path between switching the feature off entirely and letting it accumulate unchecked. In practice, that is the trade-off this option is built around: personalization versus exposure, decided entry by entry.

For developers and product teams, the design is a useful reference point. Editing memory through a structured summary, and offering a natural-language route for platforms without a full editor, is a pattern that could carry over to other assistants with persistent user context. A likely consequence for product teams, and here I am inferring, is that users will treat stored memory as something they own and expect to audit; interfaces that expose it and keep changes consistent across devices will face fewer surprises than opaque ones. The flip side, also inference, is more support and design work around sync and conflict handling once users can rewrite memory from several places at once.

Why it matters: Anyone who uses Gemini for personal or work queries now has a direct way to find and correct the profile the assistant has built, without abandoning memory altogether. Teams building assistants with long-term memory should note that visible, editable storage is becoming a baseline expectation rather than a nice-to-have. Whether that increases trust in Gemini's memory or simply surfaces how much inference goes on behind it is an open question.

Read at ZDNet · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

A new option called Memory summary lets you see and control the things that Gemini AI knows about you.