How to set up an AI interview assistant on macOS

You have twenty minutes before an interview and you want to know whether this can be working in time, not read a feature tour. It usually can, and the part that goes wrong is always the same one, permissions. This section is the short prerequisite list; the next is the exact order of operations that avoids the common failure.

• A supported macOS version. Check the current requirement on the download page rather than trusting a number written months ago. Both Apple Silicon and Intel Macs are supported, and the download page selects the right build.
• A stable internet connection. Transcription and answer generation run in the cloud, so a weak connection shows up as lag in the suggestions rather than as an error message.
• Headphones, ideally wired. This matters more than people expect. With speakers, the interviewer's voice leaves your speakers and returns through your microphone, so the assistant hears a doubled, echoing version of the conversation and transcription quality drops.
• An account, signed in before the call. Do not leave sign-in as something to do while an interviewer waits.
• Ten quiet minutes for a dry run. The single highest-value item on this list.

If you are choosing a tool rather than setting one up, the comparison of what different assistants actually do is on the best AI interview assistant page.

The order below matters, because macOS applies most privacy grants only after an app restarts. Granting a permission and continuing in the same session is the reason a setup that looks complete produces an empty transcript.

• Download the native macOS app from the official source. Step-by-step screens for SubcueAI are on the tutorial page.
• Open it and clear the first-launch prompt. Gatekeeper checks apps downloaded from the internet, and the first open may need confirmation. If it refuses to open at all, re-download from the official source rather than working around the warning.
• Grant Microphone access when prompted, or in System Settings under Privacy and Security. This captures your side of the conversation.
• Grant Screen and System Audio Recording. On modern macOS this is the permission that lets an app capture the audio your Mac is playing, which is the interviewer's voice. It is the step people miss, and its absence is silent: the app runs, the transcript just never shows the other person.
• Quit the app completely and reopen it. Not close the window, quit. This is where most failed setups actually fail.
• Sign in and confirm your plan is active so a credit prompt does not appear mid-interview.
• Position the overlay where your eyes travel least, near the top of the screen close to the interviewer's video tile, so you are not visibly reading down and to the side.

A backend engineer preparing for an L5 platform round did all of this except the relaunch, watched a silent transcript for the first four minutes of a real interview, and spent those minutes debugging instead of answering. The relaunch is not optional and takes five seconds.

A dry run converts every problem above into something you fix while nobody is waiting. Use a second device, a phone works, or a friend.

• Start a test call on the platform you will actually use, Zoom, Google Meet or Microsoft Teams. Do not test on one and interview on another.
• Speak, then have the other side speak. Both voices must appear in the transcript. Only yours means system audio capture is not active; go back and check the grant, then relaunch.
• Check the overlay is readable and stays on top when the meeting window is focused and in full screen.
• Watch the delay. Suggestions arrive a moment behind the question, which is normal. Knowing how that feels stops you pausing awkwardly to wait for text.
• Practice glancing rather than reading. Reading a paragraph aloud is audible in your delivery, and it is the most common way candidates give themselves away, not any technical signal.
• Run one full question end to end so your first real use is not your first use.

If you want that rehearsal to be more than a microphone check, mock interview mode asks real questions with follow-ups, so you practice answering and the setup at once.

Nearly every macOS issue is a permission or an audio-routing problem. In order of how often they occur:

• The interviewer's voice never appears. Screen and System Audio Recording was denied, or the app was not relaunched after granting it. Toggle it off and on in System Settings, then quit and reopen.
• Everything echoes and the transcript repeats itself. You are on speakers. Switch to headphones.
• Audio disappears when you join the call. The meeting app changed the system output or input device. Set both explicitly in the meeting app's audio settings and in System Settings, then rejoin.
• The overlay hides behind the meeting window. Full-screen mode on macOS puts the meeting in its own Space. Run the meeting windowed instead.
• The app will not open at all. Re-download from the official source; a partially downloaded or quarantined build fails this way.
• Suggestions lag badly. Check the connection first, then close bandwidth-heavy apps. On a video call the network is usually the constraint.

One limit no configuration removes: on macOS the overlay is left out of a Google Meet or Zoom share, but some apps, such as Tencent Meeting and QuickTime, can capture it, and the menu-bar icon stays visible. Sharing a single window is narrower, though window pickers and notifications still leak, and recorded or proctored sessions capture whatever a viewer would have seen. What is and is not in scope is set out on the security page. Aaron Cao, founder of SubcueAI, has been consistent that no assistant is invisible everywhere, and treating any promise of universal invisibility as a warning sign will save you a bad surprise.