Microsoft does not "spy" on you in the sense of secretly watching everything you do, but Copilot does collect and process your prompts, your conversations, and certain diagnostic data — and what happens to that data depends heavily on which Copilot you are using and which account you are signed into.
The short version: Microsoft's own privacy documentation says consumer Copilot data (the free and Pro versions tied to a personal Microsoft account) can be used to improve Microsoft's AI models, while enterprise Copilot through a work or school account is governed by different terms that generally exclude your prompts from model training. That distinction matters far more than any headline claim about "spying."
To understand why, you need to know that "Copilot" is not one product. It is a brand name Microsoft has attached to several different systems: the consumer Copilot app and website, Copilot inside Windows, Copilot in Microsoft 365 apps like Word and Outlook for personal accounts, and Microsoft 365 Copilot for business and education tenants.
Each one has its own data handling rules, and the rules are set by the account you sign in with, not by the app icon. When you use a personal Microsoft account, you are typically accepting the Microsoft Services Agreement and the Microsoft Privacy Statement, which describe how data may be used to operate and improve services.
When you sign in with an organisational account, your employer's agreement with Microsoft usually takes precedence, and Microsoft's documentation for business Copilot states that prompts and responses are not used to train the underlying foundation models. Same interface, different contract.
That is the mechanism, and it explains why two people typing identical prompts can have completely different privacy outcomes.
There are also specific controls worth knowing by name, because the general "you can manage your data" advice is useless without them. In the consumer Copilot settings, there is a toggle for whether your conversations are saved to your history, and a separate option to delete your Copilot activity history.
If you turn off conversation history, Microsoft's documentation indicates your chats are not retained for the purpose of improving the service in the same way. On the enterprise side, the relevant controls live in the Microsoft 365 admin centre and in Purview, Microsoft's compliance platform, where administrators can set retention policies, audit Copilot interactions, and apply sensitivity labels that restrict what Copilot can surface from your organisation's files.
For Windows Copilot specifically, there are also device-level settings and, in newer builds, on-device processing features that keep some requests local rather than sending them to the cloud. The practical takeaway: if you care about this, you need to look at settings in two places — your Copilot account settings and, if you are at work, whatever your IT department has configured.
A concrete example makes the difference clear. Imagine two people ask Copilot the same question: "Summarise our Q3 sales strategy and suggest improvements." Person A is signed in with a personal Microsoft account on a home laptop.
Their prompt, the document they attach, and Copilot's response are processed in Microsoft's cloud, subject to the consumer privacy terms, and may be retained and used to improve the service unless they have disabled history and deleted their activity. Person B is signed in with a work account at a company that has deployed Microsoft 365 Copilot.
Their prompt is scoped to the tenant, access is governed by the company's permissions, administrators can audit the interaction, and Microsoft's business terms state the content is not used to train foundation models. Same words, same button, very different exposure. If Person B pastes confidential client data into a personal Copilot account on the same laptop, they have just moved themselves from the enterprise bucket into the consumer one — a mistake that happens constantly and is the real privacy risk, not some secret Microsoft surveillance programme.
Now the honest limits. First, consumer and enterprise Copilot terms genuinely differ, and Microsoft updates its privacy documentation regularly, so any summary — including this one — can go stale; the only authoritative source is Microsoft's current Privacy Statement and the specific product terms for the Copilot you are using.
Second, "not used for training" in enterprise agreements does not mean "never stored" — retention, auditing and compliance logging still apply, and your administrator may be able to review interactions. Third, none of this is unique to Microsoft. Google's Gemini, OpenAI's ChatGPT and Anthropic's Claude all draw a similar line between consumer and business data handling, and all of them change that line from time to time.
The useful habit is not to memorise a verdict about Microsoft, but to check three things before typing anything sensitive: which account am I signed in with, is conversation history on, and does my organisation have a policy covering this tool. According to our AI tool database, which tracks 360 AI tools with pricing and capability snapshots verified as recently as 2026-09-18, account type and data-handling terms are among the variables that change most often between tools — which is exactly why a blanket "does it spy on me" answer ages badly.
If you want a broader walkthrough of keeping your data out of AI training pipelines, see How to Use AI With Your Privacy Intact.