Candidates regularly ask: 'how does a real-time AI interview copilot stay invisible when I share my screen on Zoom?' The answer is not a clever positioning trick or a transparent CSS overlay. It is an operating-system-level mechanism that filters specific windows out of the screen-capture pipeline at the kernel layer. This explainer walks through what InterviewLift does and what extension-based copilots cannot do.
The Underlying OS APIs
macOS: NSWindowSharingNone
macOS exposes NSWindowSharingType on every NSWindow. Setting it to NSWindowSharingNone tells the WindowServer to exclude that window from every screen-capture pipeline — including ScreenCaptureKit (used by modern Zoom / Teams / Meet), CGDisplayStreamCreate (older), and AVCaptureScreenInput (used by some recording tools). The exclusion is enforced at the compositor, so the window never reaches the capture buffer regardless of capture client.
Windows: SetWindowDisplayAffinity
Windows has the SetWindowDisplayAffinity API. Setting WDA_EXCLUDEFROMCAPTURE on a window tells the DWM (Desktop Window Manager) to render a black rectangle in place of the window in any capture stream. This affects PrintWindow, BitBlt, Windows Graphics Capture (the API used by modern Teams, Zoom, Meet), and even DRM-protected video paths.
Why Browser-Extension Copilots Cannot Match This
A browser-extension copilot renders its overlay inside the Chrome process. When Zoom or Teams captures the Chrome window, every Chrome-rendered pixel is captured, including the extension's overlay. The extension's workaround is to render outside a defined region or to detect screen-share and hide. Both workarounds are easily defeated by: (a) full-screen-share mode, (b) Chrome enterprise policies that enumerate extensions, (c) any modern proctoring tool that scans for extension IDs, and (d) any third-party recording tool not running inside Chrome.
What InterviewLift Specifically Does
Native desktop binary, not an extension
InterviewLift ships as a signed native macOS app (notarised by Apple) and an EV-signed Windows binary. There is no browser extension, so Chrome enterprise policies and proctoring tools have nothing to enumerate. The app runs as a separate OS process outside the browser sandbox.
OS-level overlay exclusion on every window
Every InterviewLift overlay window (transcription panel, answer scaffold panel, control bar) has the OS-level display-affinity flag set at creation time, before the window ever becomes visible. There is no race condition where the window could be captured between creation and flag-setting.
Audio capture via OS-level system audio, not Zoom plugins
InterviewLift captures system audio at the OS-level audio routing layer (Core Audio on macOS, WASAPI loopback on Windows). It does not install a Zoom Marketplace plugin, does not require any Zoom permission, and does not appear in the Zoom Apps panel. Your interviewer sees nothing.
Regression Testing
Every InterviewLift release is regression-tested against the latest Zoom, Microsoft Teams, and Google Meet client builds. The methodology: a candidate machine runs InterviewLift with the overlay positioned visibly to itself, a recruiter machine receives the screen-share, and a third machine records the recruiter view. We diff the recorded video against the candidate's actual screen and verify the overlay never appears in any frame.
If you are choosing between an extension-based interview AI and a native app, the architecture is the difference to check. The dated comparisons at /alternatives/final-round-ai and /alternatives/cluely show what each competitor ships today.
由 InterviewLift 客户成功团队审阅，他们每周都在帮助候选人准备真实面试。