Can Zoom, Google Meet or Teams Detect an AI Overlay? What Recruiters Actually See

Can Google Meet detect an AI overlay? The short, technical answer is that Google Meet, Zoom and Microsoft Teams do not "detect" software on your computer at all. They transmit three things: your camera, your microphone, and whatever you explicitly choose to share. Whether an AI assistant's window ends up in front of a recruiter depends on what you shared, how the operating system composes the screen, and whether a proctoring tool is layered on top of the meeting.

This article explains each layer in plain terms, describes exactly what MockMate's Windows companion does and does not do, and ends where every honest article on this subject has to end: the capture question is separate from the permission question, and only the second one decides whether you should be using a live tool at all.

What a meeting app actually sends

All three major platforms give you the same three sharing choices, with slightly different names.

Google Meet offers "A tab", "A window" and "Your entire screen". A tab share sends only that Chrome tab (with its audio, if you allow it). A window share sends the pixels of one application window. An entire-screen share sends one display.

Zoom lets you share your entire desktop, one or more application windows (hold Ctrl on Windows to pick several), a portion of the screen marked by a green border, a whiteboard, video or audio. Zoom's dual-monitor mode lets you see shared content on one monitor and participants on the other.

Microsoft Teams lets you share your entire desktop or a specific window, present PowerPoint Live, share a whiteboard, and include system audio. Microsoft's own documentation notes that if you open chat while presenting, "your audience won't see the chat window in your shared screen".

The point of listing these is that the meeting app is a pipe. It sends the region you picked. It does not enumerate your running processes, inspect your window list, or report "an overlay is present" to the other side. A recruiter on the far end sees your camera tile and, if you are sharing, the shared region. Nothing else.

What a shared region contains, layer by layer

A tab share contains only the contents of that browser tab. An AI assistant in a different tab, a different window or a native app is outside it. This is the narrowest share and the one interviewers usually ask for when they want a coding sandbox or a document.

A window share contains one application window. On Windows, a window share is produced by asking the operating system for that window's pixels. Other windows, even ones floating above it, are generally not included, but there are edge cases: some capture methods grab the screen region under the window and pick up whatever overlaps it. This varies by app and by Windows version.

An entire-screen share contains everything drawn on that display: the taskbar, notifications, every window, and any always-on-top overlay. This is the share where an assistant window is most likely to be visible, and it is the one interviewers ask for when they want to see "everything".

There is one more layer that changes the entire-screen case.

The Windows capture-exclusion flag

Windows has an operating-system feature, documented by Microsoft as SetWindowDisplayAffinity with the value WDA_EXCLUDEFROMCAPTURE, that lets an application mark its own top-level window so that it is "displayed only on a monitor" and "everywhere else, the window does not appear at all". Microsoft's stated example is video-recording controls that should not show up in a recording. It was introduced in Windows 10 version 2004. On older versions the same flag falls back to WDA_MONITOR, which shows the window as a blank area rather than removing it.

Microsoft is careful about its limits, and so should you be. The documentation says the feature "works only when the Desktop Window Manager (DWM) is composing the desktop", and that "there is no guarantee that using SetWindowDisplayAffinity ... will strictly protect windowed content, for example where someone takes a photograph of the screen". The exclusion applies to a specific set of public capture APIs. A meeting app that captures through a different path, a virtual display driver, a remote-desktop session, a second physical device pointed at the monitor, or a proctoring tool with its own capture stack can behave differently.

This is the mechanism behind every "hidden from screen share" claim in the interview-tool category, including MockMate's capture resistance. It is real and it is useful. It is also, by Microsoft's own description, not a guarantee.

What MockMate's window actually does

We will describe it exactly as MockMate's own pages do, because the marketing in this category is often looser than the engineering.

• The Live companion is a native Windows 10/11 application, not a browser extension or a web page. It captures microphone and system audio at the operating-system level, so it works the same in Meet, Zoom or Teams.
• It is a compact tool window that stays on top, with one-key quick hide and a Privacy Mode.
• It applies capture resistance on supported Windows display paths. MockMate states that this is platform-dependent and is not a guarantee of invisibility. MockMate never uses the word "undetectable".
• Screen context is user-initiated: you take a screenshot or grab visible text deliberately, review it in a queue, and can remove it before sending. Nothing is captured silently in the background.
• Every answer is labelled with its source: transcript, manual text, screen text or screenshot.

So the honest summary is: in a tab share or a window share the companion is outside the shared region; in an entire-screen share it relies on the Windows capture-exclusion path, which works on supported configurations and is not promised to work everywhere. Details on the security page.

What a recruiter can notice without any software

The detection that actually happens in most interviews is human, not technical. Interviewers who have seen a few hundred candidates describe the same tells:

• Gaze. Eyes that keep returning to one fixed spot off-camera, especially the same spot before every answer.
• Latency. A consistent two-to-four-second pause before every answer, including easy ones, followed by a fluent response.
• Cadence. Reading rhythm is different from thinking rhythm. Reading has no false starts and no "um".
• Mismatch. Vocabulary that does not match the resume, or a perfect textbook answer followed by inability to handle a follow-up.
• Follow-ups. The follow-up question is the interviewer's detection tool. It is why MockMate's Practice mode uses follow-up pressure and personas: the goal is that you can handle the follow-up yourself.

No capture-exclusion flag changes any of these. A tool that makes you read is more visible than a tool that is technically on screen.

What proctoring software adds

Proctored assessments are a different environment from a normal video interview, and the difference is the whole point of them. Instead of a meeting app that sends what you choose, a proctoring tool runs on your machine and records what it decides to record.

HackerEarth's proctoring page says its tool "records screen, webcam", "captures periodic high-resolution screenshots of the candidate's face and desk space", "tracks gaze direction and micro-movements to flag when candidates look away", "blocks the use of other applications during the test" and "prevents candidates from sharing their test screen via screen-sharing tools". Talview's glossary on screen sharing in proctoring says that "running split screen or connecting a second monitor is monitored for, alongside screen sharing and screen mirroring, as part of a complete screen-integrity check". Secure browsers from these vendors typically require administrator rights and tell candidates not to run screen recorders or remote-desktop tools.

In that environment the question "can Meet see my overlay" is the wrong question. The proctoring tool is designed to record the whole screen through its own path, to flag a second display, and to notice your eyes leaving the test window. Our Talview, HackerEarth and Mettl explainer lists what each says it does. The advice is not "pick a better overlay"; it is do not use Live assistance inside a proctored assessment.

A practical way to think about it

The question that actually matters

Everything above is about pixels. None of it answers whether you are allowed to use the tool. MockMate's policy is explicit: use Live assistance only where the organisation, interviewer or applicable rules permit assistance and disclosure, and use Practice when permission is unclear. MockMate does not authorise cheating in exams or unauthorised interview use, and it does not promise invisibility. If your plan for an interview depends on a capture flag working, you have already decided to hide something, and our ethics article is the page to read next.

Bottom line

Google Meet, Zoom and Teams do not detect overlays; they transmit the region you share. A tab or window share excludes a separate assistant window; an entire-screen share includes everything unless the operating system's capture-exclusion path removes it, and Microsoft itself says that path is not a guarantee. Proctoring tools record through their own stack and flag second monitors and wandering eyes. Recruiters detect reading, not windows. MockMate's companion is a compact tool window with one-key hide, Privacy Mode and capture resistance on supported Windows display paths, described by MockMate as platform-dependent and never as undetectable. Use it live only where that is permitted; otherwise, practise.