Copilot Chat can now show you which memories shaped a response.
Select Sources under a response and there’s a new Personalization tab next to References. It lists the custom instructions, saved memories, generated memories and past chats that informed the answer, and from there you can return to a cited chat or review and delete saved memories. Roadmap item 574111 marks it as launched worldwide on the web.
A few days ago, I wrote that there were weaknesses in the visibility of Copilot's memory. You could ask Copilot what it knew about you, but there was nowhere to see what it had inferred. This goes a long way toward fixing that, and it puts the evidence in the right place, beside the answer it influenced rather than on a settings page you’d rarely visit.
Seeing what Copilot has inferred is half of what I was asking for. The other half is understanding how it's inferred and being given the opportunity to correct it. The biggest risk is that Copilot gets the right answer for the wrong reasons, and then that understanding gets incorrectly applied elsewhere.
I do have some questions. First, does this change the way in which we should verify Copilot's output? This creates another set of inputs to check after all. And second, is this the right way to present this information or is a specific UI to manage this knowledge necessary?
First posted on Linkedin on 10/15/2026 → View Linkedin Post Here