Transparency for AI memory
Microsoft has announced a new feature for Copilot, its artificial intelligence assistant, designed to show users why they received a specific answer. When the software uses its 'memory'—a record of past interactions and saved preferences—to tailor a response, it will now provide citations.
Currently, it can be difficult to determine if a response from an AI is based on general web data, internal company files, or a specific detail mentioned in a chat three weeks prior. This update adds a 'Sources' button beneath responses. Selecting this will open a Personalization tab, listing the specific saved memories or past chats that informed the answer.
Data management and deletion
For small business owners, the primary concern with AI is often data hygiene. If an employee discusses sensitive financial details or private personnel matters with the assistant, that information could influence future responses.
This new interface allows users to see exactly what the AI has 'learned' about their work habits or projects. Users can review these citations and delete individual memories if they are no longer accurate or if they contain information that should not be stored. It also provides a direct link back to the original chat, allowing a staff member to see the full context of the source material.
Availability and settings
These citations will only appear if the memory feature is active within the Microsoft 365 environment. If a business has disabled personalisation for compliance or privacy reasons, this change will not be visible.
Microsoft lists the general availability for this update as October 2026. While this is a long lead time, it indicates the direction of travel for Microsoft's privacy controls. We do not yet know if there will be a specific administrative toggle for these citations, but they appear to be a standard part of the memory interface. We will monitor for any changes to this timeline or specific configuration requirements as the date approaches.