Copilot Vision at Work: Check Screen and Camera Sharing Before Rollout

Giving Copilot access to what is on your screen is useful for explaining a problem. It also means users need to think about everything else that happens to be visible at the same time.

Microsoft's June Copilot update described administrative controls for Vision with the user-facing capability rolling out in July. That makes this a good moment to decide how your organization wants to introduce the feature instead of waiting for a support question after someone discovers it themselves.

Start with the actual business use case

A sensible pilot might be a helpdesk scenario where a user shares an application screen so Copilot can help explain an error message or guide them through a non-sensitive configuration step.

I would not start with payroll, patient information, customer records or an administrator portal. There is no prize for choosing the most sensitive screen possible for the first test.

Write down what the feature is expected to help with. For example: “Help a user understand the visible settings on a test application without changing anything.” That gives the pilot a clear purpose.

Review the tenant control before user training

Check the relevant Copilot settings in your Microsoft 365 administration experience and verify which users are included. The exact control names and rollout status can change, so use Microsoft's current documentation and Message Center information for your tenant instead of relying on an old screenshot.

If you decide not to allow the capability yet, document that decision and the condition for reviewing it again. “Off forever” and “off while we complete a pilot” are different governance choices.

Teach users to prepare the screen first

The most practical instruction is simple: clean the screen before sharing it. Close unrelated chats, email notifications, customer records and browser tabs. Check the visible taskbar and secondary monitors. If the session does not need a camera, do not enable it.

This is familiar behaviour from ordinary screen sharing, but AI changes the expectation. A colleague on a call may ignore an unrelated line of text. An AI experience is being asked to reason about what it sees, so the user should be deliberate about the visible context.

I would also tell users not to assume that “I did not mention it in my prompt” means it is outside the shared visual context.

Test the stop button as well as the start button

In a controlled test, verify how the user starts and stops visual sharing, what indication remains visible while it is active, and what happens when the application changes. Switch to a screen that is intentionally outside the test scope and confirm the user understands how to stop sharing before doing so.

Use synthetic information for the pilot. The objective is to understand the behaviour and user experience, not to prove that a sensitive workflow can be exposed safely on day one.

Include it in your Copilot training

A two-minute demonstration of good screen preparation is probably more useful than a page of policy text. Show a messy desktop, then show the cleaned-up version you actually want users to share.

Connect the guidance to the same data handling rules employees already know. Vision is a new input method, not an excuse to invent a completely separate classification system.

Final thoughts

Vision can make Copilot feel much more practical because the user no longer has to describe every visible detail. The rollout should be equally practical: check the setting, pick a low-risk scenario, teach users to control the visual context and make stopping the share part of the demo.