Table of Contents
Copilot Memory and Copilot Communication Memory or Live Memory
On August 6, I reported about Copilot Memory, a way for users to personalize how Microsoft 365 Copilot responds to prompts. The article mentions message center notification MC1127234 (1 August 2025). In the text I noted some conflicts between the reported dates in the Technical community post about Copilot Memory and MC1127234.
As it turns out, some confusion existed about what Microsoft described in the message center notification because it doesn’t refer to the Copilot Memory feature. Instead, MC1127234 discusses Copilot “communication memory,” or information derived from a user’s personal communication used by Microsoft 365 Copilot to deliver better results to user prompts.
On August 29, 2025, Microsoft reissued MC1127234 (Microsoft 365 roadmap item 499153) to clarify what they meant, saying that the restatement was due to customer feedback, and because the change requires new processing of user data.
Learning to be More Context-Aware from User Communications
The updated text describes an “extended capability” for how Microsoft 365 Copilot gleans information about what’s important to a user from their chats, emails, and meetings. In other words, Copilot learns about what people do from the information contained in the Microsoft Graph and uses that information (referred to as “memories”) to deliver more personalized and “context-aware” responses to user prompts. “Live memories” (the same as communication memory) for individual users are formed by using AI to analyze and summarize the information held in items accessible to individual user, like transcripts from meetings they attend.
Easy as it is to become confused with so many references to different types of memory, the serious point is that communication memory is “a unified view across communication channels.” In other words, where Copilot integration in Microsoft 365 apps used to focus exclusively on information from a specific app, now Copilot draws in information from multiple sources to create a more rounded picture. For instance, the Facilitator agent in Teams operates on messages posted to a chat. With communication memory, the agent can refer to other sources available through the Graph (like previous emails between the chat participants) to deliver better advice.
People-Related Questions First
Microsoft says that Copilot communication memories will initially be used to improve the quality of Copilot responses to people-related questions. I guess this means that if you include someone’s name in a Copilot prompt, it will automatically query the Graph to find out if any information associated with that name is available to help frame its response. The new mode of processing started on September 1, 2025. However, complete worldwide deployment is not expected to be complete until late October 2025.
Making Microsoft 365 Copilot A Bit Smarter
Anything that helps AI to do a better job is welcome. In this case, it seems like Microsoft is expanding the horizon of what Copilot looks for when it assembles information to answer a question from the limit of a single app to a more-encompassing view of what’s available to a user in the Graph. That’s a good idea.
Of course, the normal caveats over anything to do with AI applies. Copilot communication memory depends on what Copilot can find, and if the information in the Graph is inaccurate, obsolete, or just plain wrong, then the advice informed by memory could be affected. Microsoft notes that citations (source links) are provided for each source mentioned in a response (Figure 1). Quite how often people check out every citation for a Copilot response is quite another matter. Sometimes, any answer will do.
Learn more about how the Microsoft 365 applications really work on an ongoing basis by subscribing to the Office 365 for IT Pros eBook. Our monthly updates keep subscribers informed about what’s important across the Office 365 ecosystem.
They probably should have named it just “communications” source or “communication history” instead of reusing memory that is already used to describe another feature.