Will Your AI Notetaker Transcribe the Video Playing in Another Tab?

Short answer: usually yes

If your notetaker does not send a bot into the call, it hears the other people through your Mac's system audio. On most tools, system audio means the sound of your whole Mac, not just Zoom or the Meet tab. While the notetaker is recording, a video playing in another tab, a voice note you replay in WhatsApp or a training clip you are half watching all go into the same stream as your meeting.

Granola says so plainly. Its transcription page says it cannot isolate audio from individual applications and captures the combined audio stream from your system, and that music or other audio played during a meeting will be included in the transcription even if it plays through a different output device. Its transcript labels your microphone as Me and system audio as Them, so a stray video arrives on the same side of the transcript as your client.

Fireflies' desktop app records without a bot using what it calls your computer's system audio. Its help pages list supported apps such as Zoom, Google Meet, WhatsApp and FaceTime, but that list is about detecting a call and offering to record it. The pages we read do not say whether audio from other apps is kept out once a recording is running, so treat it as included until a test shows otherwise.

Why the tools do not just listen to the meeting app

It is not a technical impossibility. Apple's Core Audio taps, the API most Mac notetakers now use, can capture the output of one process or a chosen group of processes. So why take everything?

Because the meeting is often not one process. A call in Google Meet runs in your browser, and Chrome plays sound through helper processes that come and go. A tap that names the apps it listens to when the recording starts can miss the process that actually carries the call audio, and the result is a transcript with only your side of the conversation. Driffle's engineers hit exactly this: the capture code notes that the per process approach missed Chrome's audio, and the app now taps all system audio except its own. Granola documents the same outcome.

That is the trade every botless tool is making. Capture one app and risk silently losing the other side of a browser call. Capture the whole Mac and accept that anything else playing becomes part of the record. Losing your client's half of the call is the worse failure, so most tools choose the whole Mac. It is a reasonable choice. It just has a cost you should know about.

Music is harmless. Other people's voices are not

A notification ping or instrumental music mostly produces nothing in a transcript, or a line of noise you can ignore. Speech is the problem, because speech is what transcription is built to keep.

Consider the voice note from a different client that you replay while waiting for someone to join, or the recording of yesterday's board meeting you scrub through to find a number, or a second call you are listening to on mute. Every word of it lands in this meeting's transcript under Them. From there it can reach the summary, the action items and search, and the notes now say something your client never said.

The cost is not only accuracy. If your notetaker emails a recap to attendees, or you share the notes, another client's words can travel with them. A few seconds of stray audio is easy to miss in a long transcript and much harder to explain once it has gone out.

What helps and what does not

Headphones do not help. They change where you hear the sound, not what the notetaker captures, and Granola's docs say other audio is included even on a different output device.

Pausing or muting the source helps. Stop the video, or mute the tab or the app playing it, so it stops producing sound. Turning down your Mac's volume is a different thing, and whether a tool captures before or after your volume setting is not something vendors document, so do not rely on it.

Pausing the notetaker helps most, when the other audio is the point, for example when you need to replay a clip for yourself mid call. Most tools have a pause control; use it, play the clip, resume. Then read that stretch of the notes before sharing anything.

A two minute test for your trial

Start a test call with a colleague and let the notetaker record it. Halfway through, play thirty seconds of a spoken video in another browser tab, something with a distinctive phrase. Then play it again through a different output, such as AirPods while the call stays on your speakers.

When the notes arrive, search for the phrase. Check the transcript, then the summary and action items. If the phrase shows up, you know what the tool captures, and you know to pause it before you play anything with voices in it. If it does not, ask the vendor in writing how they keep other apps out, and repeat the test with the call in a browser tab to make sure your colleague's side was still captured.

What Driffle captures, and when

Driffle records a meeting only between the moment it starts and the moment you press Stop. During that time it hears your microphone and all of your Mac's system audio except Driffle's own sounds, through Apple's audio tap rather than screen recording. So the answer for Driffle is the same as for Granola: a video or voice note playing during a recording will be transcribed, and it arrives on the same side as the other people on the call.

The floating pill has Pause, Resume and Stop, and Pause stops capture of both your microphone and the call audio, so it is the control to use before you play anything else with voices in it. Audio is sent for transcription and is not stored as a recording.

Sound is separate from Driffle's screen context, which reads work context from other apps. That is off until you allow it during setup or later in Settings, and Settings lets you pause it or exclude an app. Driffle runs on Macs with Apple Silicon, and transcription and the AI run in the cloud. If you want botless notes with that control in reach, you can get Driffle at driffle.ai/download.