Copilot Memory Not Working? Settings and Recall Checks
Start by identifying which Copilot you use, then check its memory controls and the saved fact. A small recall test is more useful than repeatedly asking the assistant to remember everything.
Updated 5 min read
If Copilot memory is not working, first identify whether you mean Microsoft Copilot, Microsoft 365 Copilot or GitHub Copilot. Then check that product’s controls, confirm the needed detail was saved and test it in a new conversation on the same account. fenbs can keep explicit project context and approved rules available separately when work should not depend on personal recall.
What should you check first?
- Write down the app or website, account and sign-in type. A personal Microsoft account, a work account and a GitHub account lead to different controls.
- Find the product’s memory or personalization settings. Use the documentation linked below for that product instead of following a screenshot from a different Copilot.
- Inspect the saved detail where a memory list is available. Check the wording: mentioning a fact in an old conversation does not establish that it was saved for later.
- Try one harmless, specific recall question in a fresh conversation. Keep the same account and avoid adding the answer to your test question.
- Compare the result with the saved record. If it fails, note the missing setting or incorrect detail before trying more changes.
Do not start by deleting all your conversations. That removes useful evidence and may not address the control responsible for the behavior. You want to isolate the failure: wrong account, unavailable feature, missing saved detail, conflicting instruction or an answer that failed to use available context.
Which Copilot product are you troubleshooting?
| Product | Start with | Separate issue to avoid confusing with memory |
|---|---|---|
| Microsoft Copilot for individuals | Your account’s personalization and saved-memory controls. | Whether an earlier conversation is still in history. |
| Microsoft 365 Copilot | Copilot Chat settings and Personalization for the work or school account. | Access to the work content needed for the answer. |
| GitHub Copilot | Personal Copilot settings, supported feature and repository memory. | Whether repository instruction files were supplied. |
Memory is not a general switch that makes every document, chat and app visible. A request to recall your preferred writing style differs from a request to retrieve a spreadsheet or follow a codebase rule. Name the missing information before deciding which setting could help.
If the missing information is a repository instruction, use Copilot instructions not working. If the coding session itself fails to run, use GitHub Copilot agent mode troubleshooting. Those checks address different failures.
Was the Microsoft Copilot detail actually saved?
For the personal product, Microsoft’s privacy controls guide (opens in a new tab) describes Settings, Personalization and the Saved memories controls. Review the saved items rather than assuming every detail in a past conversation became memory. Follow the guide that matches the controls visible in your account.
For Microsoft 365 Copilot, the memory management guide (opens in a new tab) describes Chat settings, Personalization and Manage saved memories. It also describes a memory-updated notice when a detail is saved. Check the list for the exact preference you expected it to retain.
Use a low-risk example: “Remember that I prefer a short summary before detailed notes.” Then inspect the relevant controls and try a new ordinary conversation. A long project plan is a poor first test because you cannot easily tell which parts were retained or why the answer omitted one.
Microsoft also distinguishes saved memories, chat-history personalization, work insights and custom instructions in its personalization explanation (opens in a new tab). Temporary chats do not use or update personalized information. Check the conversation type before treating absent recall as a broken setting.
Does the GitHub Copilot session support the memory you need?
GitHub’s memory overview (opens in a new tab) separates repository facts from personal coding preferences and lists the supported features. Check the exact surface rather than assuming memory applies identically to every editor suggestion, review or coding session.
Open personal Copilot settings to review the memory setting and your preferences. Repository owners can inspect repository facts under the repository’s Copilot Memory settings. GitHub’s management instructions (opens in a new tab) explain those routes and how managed access can affect availability.
Test against a fact the current code supports, such as a package’s test command. Ask the assistant to identify its evidence. A remembered claim that conflicts with the repository should be corrected before you let it drive changes. The goal is accurate work, rather than proving that an old statement survived.
Instruction loading has its own rules across coding apps. The CLAUDE.md troubleshooting guide is useful when the missing guidance belongs to that file, but changing it will not repair a Microsoft Copilot personalization setting.
What if the setting is on but the answer is still wrong?
Reduce the test to one detail. Confirm that it is current, relevant and not contradicted by another instruction. Start a fresh conversation and ask the same small question again. Record whether the saved item is absent, wrong, or present but not reflected in the answer; those findings lead to different corrections.
For example, a saved preference for short answers does not answer which proposal a client approved. Give the assistant the approved decision and its scope. Ask it to separate established facts from questions still waiting on someone, then check that summary before requesting the next piece of work.
If a repeatable failure remains, use the product’s support route with the app, account type, visible setting and a sanitized example. Leave out private conversations and customer information. A reproducible small test gives support more to investigate than “it forgot my project.”
For ongoing work, fenbs project context gives you a deliberate record to inspect and correct when a useful decision gets lost among preferences.
Keep the approved work context somewhere you can check
Save the project goal, current tasks and decisions in fenbs instead of asking personal memory to hold the whole job. Supported connected assistants can read Rules and Where we left off before working. You can also copy a reviewed summary manually into an app that is not connected.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account and add the project whose context keeps going missing.
- Record the decisions a person approved in Decisions and rules. Separate ongoing rules from choices for a single task, and leave unanswered questions open.
- Add tasks and save Where we left off with completed work, checks and the next step. Review that record before reusing it.
- Use a supported connection from Integrations, or manually copy the relevant summary into Microsoft Copilot. After working, update the project record with what actually changed.
fenbs does not repair Copilot settings or import Copilot memories, and it does not connect to Microsoft Copilot or Microsoft 365 Copilot. It is not the right fit if you only need the assistant to remember a personal preference. It fits project work that needs an explicit shared record.
For another app’s recall controls, use ChatGPT memory troubleshooting. Explore the AI memory hub when you want to distinguish personal recall from the work you deliberately maintain.