Comparison · 8 min read
Copilot vs ChatGPT: memory compared
Both Microsoft Copilot and OpenAI's ChatGPT promise an AI that "remembers you." They mean fundamentally different things by that. Here's what each one actually stores, where it stores it, and which one is the better memory system for what.
Three products, one word
"Copilot" is now Microsoft's brand for several distinct products, and each treats memory differently. To compare cleanly:
- Consumer Copilot (copilot.microsoft.com): the direct ChatGPT equivalent. Has an opt-in memory layer that stores explicit notes about you.
- Microsoft 365 Copilot (inside Word, Outlook, Teams, etc.): draws context from your Microsoft Graph — the live enterprise data behind your Microsoft 365 account.
- GitHub Copilot: no personal memory; scoped entirely to the current repo and open files.
The head-to-head
| Dimension | ChatGPT | Copilot (Consumer + M365) |
|---|
| Personal notes layer | Yes, robust | Yes (consumer), lighter |
| Work-context integration | Weak (connectors only) | Deep (M365 Graph) |
| Cross-chat retrieval | Yes, opt-in | Limited |
| Memory panel edit-ability | Explicit UI | Privacy dashboard |
| Enterprise admin controls | Limited | Strong (Purview, DLP) |
| Portability | Weak — export exists but no import path anywhere | Weak — tied to Microsoft 365 |
Which to use when
Use ChatGPT for the "know me as a person" layer.Cross-topic preferences, long-running projects, voice — it's the strongest of the three at maintaining continuity about you rather than about your work.
Use M365 Copilot for the "know my work" layer.Summarize this thread, draft an email in the tone of my last five, find the deck I sent to X last quarter — these are all things M365 Copilot does natively because it can see the actual files.
Use GitHub Copilot inside the repo, and don't expect it to remember you outside it.
The portability trap either way
Both systems lock their memory to themselves. If you build up months of context in ChatGPT and want to try Copilot — or the reverse — the memory doesn't transfer. And if either provider deprecates the model you'd tuned to, the retrieval shifts even if the notes don't. This is the case for keeping a canonical persona in a portable format both ChatGPT and Copilot can consume as pasted context: one edit, works in both places. Related: best AI for long-term memory walks through the same comparison across all major providers.