Gemini Saved Info and Memory: What Carries Between Chats

Gemini saved info refers to information or instructions used to personalize later responses. Check Instructions for Gemini, Memory, and Connected Apps separately: a response preference, an earlier conversation, and information from another app are different sources of context.

Personalization can save repetition, but it does not tell a teammate what was approved or where a project stopped. fenbs keeps those selected work rules and progress outside personal Gemini memory. Use Gemini’s controls for how it responds to you and a project record for work another session must understand.

Where should you start with saved information?

1. Check the Google Account you are using, then open Gemini’s settings and look for the personalization area. Labels can include Personal Intelligence or Personal context.
2. Find Instructions for Gemini and review the instructions already present before adding another one.
3. Check Memory separately if you want personalization based on earlier conversations.
4. Review Connected Apps when you expect Gemini to use information from another Google service.
5. Test one small preference in a fresh ordinary chat, then check whether the response reflects it.

Google’s personalization overview↗ (opens in a new tab) distinguishes past-chat memory from instructions and says availability varies. These personal features require a personal Google Account rather than a work, school, or supervised account. Use the current feature pages to check eligibility for your situation.

Write down which control you changed. “Memory on” and “added a writing instruction” describe different tests. Changing several controls at once makes it difficult to tell why a later answer improved or why information appeared that you did not expect.

How do instructions differ from memory and connected data?

The table separates the questions to ask; it does not promise that every account has the same controls. Google’s memory guide↗ (opens in a new tab) explains that past-chat personalization requires Keep Activity and is unavailable in certain experiences, including Gems and Live chats.

A fresh chat can therefore behave differently from another type of Gemini experience. Test the place where you actually plan to work. For broader differences between remembered facts and an ongoing work record, see what agent memory means.

The distinction matters when a sentence sounds like an instruction but contains a project decision. “Give concise answers” is a response preference. “Use the client-approved accessibility checklist for the museum website” needs the checklist, its approval, and the project scope. A short personal preference cannot carry all of that reliably.

What makes a useful instruction for Gemini?

Use short instructions you can check in an answer. Start with one behavior, such as “Lead with a two-sentence summary, then give the supporting detail.” Avoid stacking broad requests to be concise, exhaustive, casual, formal, creative, and exact in the same entry.

Google’s instructions and remembered information guide↗ (opens in a new tab) shows how to add, edit, delete, and switch instructions on or off. It also distinguishes adding instructions from asking Gemini to remember specific information, whose eligibility and rollout can differ.

• Prefer observable behavior: “Use American spelling” is easier to check than “write better.”
• Give each instruction a clear purpose. Keep personal formatting preferences apart from changing project facts.
• Remove an obsolete instruction rather than adding a competing correction beneath it.
• Leave approvals and unfinished work in a record that can name the project and the person responsible for the choice.

For example, a student might prefer definitions before examples across ordinary questions. A design studio’s approved delivery requirements apply to one job and need a different home. Decisions and rules in fenbs preserve that scope and show rules first when a connected assistant reads the project context.

How do you review or remove information responsibly?

Begin with the source of the information. Is the detail written in an instruction, present in an earlier chat, or available through a connected service? Review that source rather than relying on a broad request to stop mentioning the topic.

For connected information, Google’s Connected Apps guidance↗ (opens in a new tab) explains why disconnecting an app and deleting chats are separate actions. A detail can remain available through one source after you remove it from another. Review the vendor’s current removal instructions before assuming a change is complete.

Keep work you are authorized to retain before removing the only copy of it. That may mean saving an approved paragraph in its document or recording a decision with its reason. Do not copy a whole personal conversation into a team record when only one project fact is needed.

After editing a preference, test a new request that would make the preference relevant. Inspect the actual behavior instead of asking only whether Gemini remembers you. A confident statement about memory is less useful than an answer that follows the instruction you intended.

What should stay outside personal Gemini memory?

Keep changing project status, approved decisions, and work that another person must continue in an explicit project record. Personal memory may help an assistant tailor an answer, but the record should make it possible for a person to read what was chosen and correct it.

Say you are preparing a community workshop. Save the approved audience, the agreed handout format, and the fact that the room setup still needs checking. Mark a suggested activity as a suggestion until someone chooses it. These details explain work, rather than your general preferences.

Other Google workspaces have their own purposes; see the guides to Gemini Notebook and Gemini Gems for those separate workflows. The AI memory hub covers the wider question of continuity between sessions.

How to do this with fenbs

1. Create a free account and add one project whose rules you currently repeat in chats.
2. Record the decisions people made, including their scope and reason. Mark continuing instructions as rules in Decisions and rules.
3. Add the remaining tasks and write Where we left off. State what is complete, what was checked, and what needs a person’s answer.
4. Use Integrations to connect a supported assistant and ask it to read the project context. The Gemini guide is for Gemini CLI; it is not a Gemini web-app memory connection.
5. At the end of the session, save current progress so the next supported assistant receives the updated starting point.

fenbs is not the right fit for remembering your favorite meals or changing how the Gemini app formats answers. Its purpose is to keep the work. Start with the one project rule or unfinished task you would otherwise have to explain again.